The increasing demand for reliable AI models is also evident in Hierarchical Energy Management Systems (HEMSs). Especially in complex energy systems like Renewable Energy Communities (RECs), efficient cost minimization is a strong incentive for potential users to participate. Nevertheless, trust in AI stems not only from its performance but also from AI Explainability (XAI). In this regard, the new Kolmogorov-Arnold Network (KAN) XAI paradigm offers significant advantages over Multi-Layer Perceptrons (MLPs). In this work, a KAN is optimized by a Genetic Algorithm (GA) to serve as an inference engine in a realistic REC HEMS. The main goal is to validate KANs in that application domain. Secondly, a custom coefficient space quantization is proposed to enable efficient KAN-GA encoding. Specifically, the generic KAN model is encoded as a GA individual and optimized to minimize the operational cost of the REC. Then, an explainable AI model is extracted from the original KAN by fitting its connection splines with simpler function forms and by applying the Kolmogorov-Arnold Theorem to get the output. The results show a high precision of the KAN models with a low computational cost. In addition, the explainable model performance is very close to that of the original KAN in terms of precision, while being significantly better in terms of computational efficiency, with about 10% time-saving. Therefore, developing fast and explainable KAN-based models is worthwhile in that application field.

Accurate forecasting of energy time series is essential to support Energy Management Systems (EMS) in taking real time decisions of energy flows in microgrids. In the traditional two-stage methodology, forecasts are made upstream and provided as input to the EMS by a prediction model trained in advance on energy prices, loads and generation time series. This study adopts an innovative approach of Decision Focused Learning (DFL) for energy demand, production and price forecasting in a microgrid environment, wherein for each time series an LSTM neural network is trained end-to-end by embedding the optimization cost directly into the loss function. Results show that the proposed DFL approach reduces the operational cost by 11% compared to the conventional two-stage method.

Accurate energy consumption forecasting is critical for modern power systems, supporting both operational reliability and cost optimization. Deep learning approaches, particularly Long Short-Term Memory (LSTM) networks, have become increasingly prevalent in time-series load forecasting. However, typical LSTM-based solutions do not explicitly model inter-feature correlations and often undervalue peak-load periods, which are crucial for grid stability and energy market operations. This paper proposes a Graph-Aug LSTM (Graph-Augmented LSTM) with a weighted loss function. First, we construct a feature-level correlation graph and employ a Graph Attention Network (GAT) to learn a global embedding that captures cross-feature relationships. This embedding is concatenated at each LSTM time step, providing the model with explicit awareness of inter-feature dependencies. As an additional novelty, we introduce a Weighted Mean Squared Error (MSE) that emphasizes peak consumption intervals, thereby reducing the risk of underestimating high-demand periods. Validation on multiple U.S. city datasets demonstrates that our approach consistently outperforms naive baselines, a standard LSTM, and XGBoost in both overall error metrics and peak-load accuracy. These results highlight the value of integrating graph-based feature embeddings with a peak-focused loss function for more reliable and interpretable load forecasting.

Predictive maintenance is critical for ensuring the reliability and efficiency of Medium Voltage (MV) power grids. This paper presents a novel framework combining Kolmogorov–Arnold Networks (KANs) with SHapley Additive ex-Planations (SHAP) to predict and interpret real-world faults detected in Azienda Comunale Energia e Ambiente (ACEA)’s MV grid in Rome (Italy). The KAN model captures complex nonlinear interactions between Constitutive Parameters (CPs) and Exogenous Causes (ECs) measured through smart sensors, achieving high predictive performance with a ROC AUC of 0.993 over several traditional classifiers. SHAP analysis enhances interpretability, revealing ECs, such as mean and maximum currents, as dominant predictors, while CPs, including cable length and material composition, highlight structural vulnerabilities. Insights from clustering and dependence analyses enable targeted maintenance strategies by distinguishing fault scenarios driven by environmental or structural factors. This integrative approach bridges predictive accuracy and actionable interpretability – within the explainable AI (XAI) paradigm, providing a robust tool for smart grid management and condition-based maintenance.

Hierarchical processing and information granulation have proven essential for intelligent systems, as exemplified by Large Language Models (LLMs) backed by Transformer architectures, which leverage stacked attention modules to learn progressively richer semantic features. In this work, we offer an investigation of the role of attention layers in the hierarchy through a GPT-2 layer ablation methodology, which recalls the deactivation of the HAL 9000 computer modules in the iconic scene of the film "2001: A Space Odyssey". The adopted methodology is based on the measurement of appropriate indices (Dale-Chall Readability, BLEU and Text Flow – a measure of the coherence within the flow of sentences) characterizing the text produced following the removal of a combination of layers assisted by a single-way analysis, to characterize these combinations. Subsequently, through a "machine-in-the-loop" procedure, we let GPT-4 to judge the texts produced by GPT-2. The obtained results are in line with the basic hypothesis according to which the hierarchical organization of the Transformers is the ground of the high semantic performances, opening the path to further insights and application hypotheses such as Explainable AI and the analytical characterization of the texts produced by Generative AI models.

The current worldwide use of generative models in the field of Artificial Intelligence creates new possibilities but also new challenges and problems. In such an interconnected world, data produced by machines can become part of the training phases of new Language Models (LMs), and this can result in a problem that some scholars are starting to call "model collapse". In this paper, we discuss the problem from three points of view. Firstly, we show how the accumulation of data can cause a given probability distribution obtained through a sampling procedure to degenerate towards a single-state-like equilibrium as the iterations of accumulations go by. The second perspective concerns the analysis through semantic, syntactic and complexity theory-related measures of texts produced through a recursive procedure involving an LSTM that is driven by its own generated data. The aim is to measure the quality of the text and the possible level of degradation. The third point concerns the synthesis of a simplified differential model that frames this scenario in which the models feed themselves with their data and which allows do estimates, albeit rough, about the future. Computational results show a model drift and degradation of the text structure that evolves towards simplified forms from the lexical point of view and degraded in terms of destruction of long-term correlations (one of the main ways to generate meaning in a sentence). Ultimately, the study demonstrates that the problem exists and, although far in the future thanks to the power of the modern architectures underlying LLMs, should not be underestimated.

Renewable Energy Communities (RECs) have emerged as a promising solution to the intermittent nature of Renewable Energy Sources (RES). The stochastic nature of RES demands the presence of Energy Storage Systems (ESS) to increase self-consumption. Such systems, however, must be properly managed to enhance battery life, minimize wear costs, and ensure safe operating conditions. RECs face challenges such as low self-consumption and limited battery lifespan. This paper proposes an enhanced Soft Actor-Critic (SAC) algorithm tailored to optimize battery dispatch in RECs. Unlike standard SAC, which primarily relies on action clipping, our approach directly penalizes constraint violations within the agent’s objective, guiding it toward more feasible and profitable dispatch strategies. The method maintains optimal battery State of Charge (SoC), extends battery life, and maximizes economic returns from solar energy usage. Compared to the standard SAC model, our Lagrange-SAC approach achieves an 18.2% improvement in the mean Self-Sufficiency Ratio (SSR), significantly increasing the efficiency of solar energy utilization. These findings highlight the potential of advanced reinforcement learning techniques to enhance real-world energy systems, promote sustainable practices, and improve the resilience of energy infrastructures.

Accurate forecasting in time series data is crucial, especially in the energy sector, where prediction precision significantly influences decision-making and operational efficiency. This study investigates the efficacy of integrating second derivative data into forecasting models for energy consumption. We employ four distinct energy consumption time series datasets, each exhibiting varied characteristics and trends. The core of our methodology is the innovative incorporation of second derivative data to improve the accuracy of energy forecasting. This approach is applied to two widely recognized forecasting algorithms: Long Short-Term Memory (LSTM) and Extreme Gradient Boosting (XGBoost). Our research introduces the second derivative of energy data as an additional input for these algorithms. This supplementary feature is used to provide deeper insight into the acceleration and deceleration trends in energy consumption, aspects often overlooked in standard models. We compare the performance of these enhanced models against their traditional counterparts, which do not utilize second derivative data. The results demonstrate a significant improvement in forecasting accuracy, particularly in peak regions, for both LSTM and XGBoost models with the inclusion of second derivative data. This research holds broad practical applications, notably in energy management systems and smart grid technologies, where it can contribute to more efficient energy distribution decisions.

Smart Grids (SGs) represent a prominent energy efficiency strategy in the fight against Climate Change. In 2018, with RED II, the European Union made SGs realization more concrete thanks to the definition of Renewable Energy Communities (RECs), i.e. local electrical grids equipped with Renewable Energy Generators (REGs) and batteries. REC participants receive incentives proportionally to the green energy they share, aiming at more affordable technology. Therefore, proper Energy Management Systems (EMS) are required to minimize energy losses and costs to exploit accurate real-time power forecasting. In this context, battery management plays a crucial role because of its function as an energy buffer and its not negligible costs. That said, fast and accurate battery models are needed to synthesize realistic EMSs from the perspective of efficient embedding in the field. Equivalent Circuit Models (ECM) offer speed, reasonable accuracy and good explainability as grey-box models. In this work, a Thevenin ECM is synthesized with the aim of increasing realism in a REC EMS model, which relies on a simple battery linear model. The above ECM objective is to achieve flexibility and embedding suitability. First, one single trained model is synthesized for different ambient temperatures, and, secondly, promising results towards only one set of optimal parameters (instead of two, one for charging and one for discharging) are optimized, leading to possible future developments aiming at a faster and more general battery model for online EMS applications. The EMS results from simulations with the new ECM model are compared to the previous linear battery model, showing that the ECM model does not compromise the REC EMS thesis according to which auto-consumption is worse than a cost-minimization-oriented solution.

Modeling and recognizing events in complex systems through machine learning techniques is a challenging task. Especially if the model is constrained to be explainable and interpretable, while ensuring high levels of accuracy. In this paper, we adopt a bilinear logistic regression model in which the parameters are trained in a data-driven fashion on a real-world dataset of power grid failure data. The bilinear white-box model - grounded on a specific neural architecture - has been proven effective in classifying faulty states with a performance comparable to several classifiers in technical literature. Additionally, the low computational complexity of the bilinear model, in terms of the number of free parameters, allows gaining insights into the fault phenomenon correlating the events that impact the power grid (exogenous causes) with its constitutive characteristics, thence eliciting the relational information hidden in the data. The proposed model is also able to estimate a vulnerability vector that can be associated, as a suitable characteristic "label", to power grid components, opening the way, as will be deeply demonstrated in the following, not only to predictive maintenance programs or condition monitoring tasks but also to risk assessment and scenario analyses in line with the explainable AI paradigm.

The introduction of Transformer architectures - with the self-attention mechanism - in automatic Natural Language Generation (NLG) is a breakthrough in solving general task-oriented problems, such as the simple production of long text excerpts that resemble ones written by humans. While the performance of GPT-X architectures is there for all to see, many efforts are underway to penetrate the secrets of these black-boxes in terms of intelligent information processing whose output statistical distributions resemble that of natural language. In this work, through the complexity science framework, a comparative study of the stochastic processes underlying the texts produced by the English version of GPT-2 with respect to texts produced by human beings, notably novels in English and programming codes, is offered. The investigation, of a methodological nature, consists first of all of an analysis phase in which the Multifractal Detrended Fluctuation Analysis and the Recurrence Quantification Analysis - together with Zipf's law and approximate entropy - are adopted to characterize long-term correlations, regularities and recurrences in human and machine-produced texts. Results show several peculiarities and trends in terms of long-range correlations and recurrences in the last case. The synthesis phase, on the other hand, uses the complexity measures to build synthetic text descriptors - hence a suitable text embedding - which serve to constitute the features for feeding a machine learning system designed to operate feature selection through an evolutionary technique. Using multivariate analysis, it is then shown the grouping tendency of the three analyzed text types, allowing to place GTP-2 texts in between natural language texts and computer codes. Similarly, the classification task demonstrates that, given the high accuracy obtained in the automatic discrimination of text classes, the proposed set of complexity measures is highly informative. These interesting results allow us to add another piece to the theoretical understanding of the surprising results obtained by NLG systems based on deep learning and let us to improve the design of new informetrics or text mining systems for text classification, fake news detection, or even plagiarism detection.

Accurate energy consumption forecasting is essential for optimizing resource allocation and ensuring a reliable energy supply. This paper conducts a thorough analysis of energy consumption forecasting using XGBoost, SARIMA, LSTM, and Seasonal-LSTM algorithms. It utilizes two years of hourly electricity demand data from Italy and the PJM region (USA), categorizing algorithms into seasonality and non-seasonality groups. Performance metrics like RMSE, MAE, R2, and MSPE are employed. The study underscores the importance of considering seasonality, with SARIMA and Seasonal-LSTM achieving high accuracy in the seasonality group. In the non-seasonality group, XGBoost and LSTM perform competitively. In summary, this research aids in choosing suitable forecasting algorithms for building an Energy Management System for smart energy management in microgrids, considering seasonality and data attributes. These insights can also benefit energy companies in efficient resource management, promoting sustainable energy practices and urban development.

In the era of generalist social media, finding users who share the same diseases and the same related experiences during their course is one of the main objectives of patients. In this reference framework, in applications related to recommender systems or infoveillance, just to name a few, it is useful to synthesize language models capable of capturing the semantic relationships in short texts written by patients in various posts, with the dual goal of training well-performing classification systems. In this work, a series of semantic text representation approaches - both traditional and advanced - are compared through NLP techniques, in order to classify Italian users belonging to discussion groups on medical topics. The classification and semantic evaluation experiments of the models are satisfactory above all, especially by considering that the collected dataset is unbalanced.

The study of languages' structure and their organization in a set of well-defined relation schemes is a delicate matter. In the last decades, the convergence of traditional conflicting views by linguists is supported by an interdisciplinary approach that involves not only genetics or bio-archelogy but nowadays even the science of complexity. In light of this new and useful approach, this study proposes an in-depth analysis of the complexity underlying the morphological organization, in terms of multifractality and long-range correlations, of several modern and ancient texts pertaining to various linguistic strains (including ancient Greek, Arabic, Coptic, Neo-Latin and Germanic languages). The methodology is grounded on the mapping procedure between lexical categories belonging to text excerpts and time series, which is based on the rank of the frequency occurrence. Through the well-known MFDFA technique and a specific multifractal formalism, several multifractal indexes are then extracted for characterizing texts and the multifractal signature has been adopted for characterizing several language families, such as Indo-European, Semitic and Hamito-Semitic. The regularities and differences in the linguistic strains are assessed within a multivariate statistical framework and corroborated with a Machine Learning approach that is dedicated, in turn, to investigate the predictive power of the multifractal signature pertinent to text excerpts. The obtained results show a strong presence of persistence, i.e., memory, in the morphological structure of analyzed texts and we claim that this property has a role in characterizing the studied linguistic families. In fact, for example, the proposed analysis framework - grounded on complexity indexes - is able to easily distinguish ancient Greek texts from Arabic ones, as they belong to different language strains, i.e., indo-European and Semitic, respectively. The proposed approach has been proven effective and can be adopted for further comparative studies and for designing new informetrics for further advances in the fields of information retrieval and Artificial Intelligence.

The adoption of autonomous driving systems is increasingly widespread in land vehicles for private and public transportation and more standards have been defined (among which SAE and ADAS are the most complete). Although in maritime transportation automatic navigation was developed earlier with respect to similar systems developed for land vehicles, there are no standards equivalent to ADAS and SAE. Furthermore, the automation on boats is always partial and refers only to the route planning, obstacle avoiding and motion control functions. In this paper an autonomous driving system in a simulated environment for boats is presented with the attempt to help to define a standard equivalent to those mentioned above. The autonomous driving system presented takes into consideration many simulation aspects and has been developed as a single complete software solution. The design involved both classical approaches and computational intelligence techniques. We distinguish three main implementation phases relating to as many levels of control of the architecture. In this paper, we present only the simulation results related to the mid-level, giving particular attention to the tasks of boat avoidance and docks avoidance (i.e., entry and exit from ports), in which the differences between the crisp and fuzzy logic versions of the same mid-level controllers are highlighted. Results show how fuzzy controllers allow ADS to achieve a lower probability of collision and stall with the same performance as an ADS with crisp controllers. The performance indicator that has been formulated for this work is inspired by Fish Schooling Behavior, which has strong similarities with the self-driving boat problem.

Even though it is known that Renewable Energy Sources (RESs) are necessary to face Climate Change and pollution, technology is still in a developement phase, aiming at improving energy exploitation from RESs, as these type of sources suffer from low energy density and variability over time. Thus, proper ICT infrastructures equipped with a robust software, i.e., Energy Management System (EMS), are needed to ensure that Renewable Energy (RE) does not go to waste. Relatively small local electrical grids called Microgrids (MGs) represent the EMS ecosystem, since their main features are the proximity between generation and loads and the presence of Energy Storage Systems (ESSs) adopted to recover surplus energy. The Vehicle-to-Grid (V2G) paradigm helps to realize the Smart City, which in substance is an interconnection of MGs hosting electrical vehicles for an efficient energy management at a larger scale. In this context, e-boats have only recently been considered. Hence, in this work a Multi-Objective (MO) EMS is synthesized for an e-boat docked in a small Microgrid (PV generator and ESS) with the aim of maximizing the charging time of the e-boat ESS and spending as little as possible both for energy purchase and also in terms of ESS wear. A Fuzzy Inference System - Hierarchical Genetic Algorithm (FIS-HGA) is used to achieve the Pareto Front, with the HGA that is in charge of optimizing the FIS parameters. Results laid to a balanced trade-off between the two objectives, since the e-boat ESS is almost fully charged in a reasonable time and with a low cost, compatible with people transportation. Last but not least, the inference process of a FIS is easily interpretable, in the perspective of an Explainable AI.

Recent developments in communication networks have made new kind of services possible, which are even more connected due to smart Internet of Things devices and sensor networks. This complex integration of technologies had increased the attack surface on networks that are spread across multiple environments, with different protocols and services, usually based on the wireless medium. Network Intrusion Detection Systems aim to overcome these security issues, analyzing network traffic and applying multiple reconnaissance techniques. For this purpose, state-of-the-art research activities have been focused on Artificial Intelligence and Machine Learning-based approaches in order to scale more efficiently the capabilities of this kind of systems (in terms of both complexity, speed and training data required). Our work further analyzes these security issues, in particular, by introducing a consolidated approach based on the Granular Computing information processing paradigm applied to sequences of WiFi frames. We focused on a subset of attacks collected in the Aegean WiFi Intrusion Detection dataset, that consists in complex multi-step attacks or simpler control frames flooding attacks, not recognizable using a single-frame processing strategy. This kind of structured data heterogeneity implies a non-exclusive labelling strategy, as each frames' sequence could bring information related to different attack classes, making the whole supervised problem more challenging. We show that the proposed solution provides interesting results in terms of classification performances, limited embedding complexity and peculiar white-box trained models.

The ability to perform predictive maintenance, as one of the main asset of Industry 4.0, is known to help improve downtime, costs, control and production quality. Modern predictive maintenance programs involve machine learning techniques, within the AI umbrella, that work in a data-driven fashion. This is true in all machinery where, through intelligent sensors, it is possible to collect data to be processed to detect faults or carry out anomaly detection activities. This paper presents a system for the detection of anomalies in the railway context and, specifically, in the pressurization systems of Italian high-speed trains. The available real-world dataset is in form of unlabeled time series of fixed length of 600 samples. Hence, it is proposed a two-stage machine learning workflow where the first stage acts in an unsupervised fashion through a statistical technique validated by field experts with the aim of building a labeled dataset. In the second stage, the faced problem is conceived as a classification task in the context of a strong class imbalance problem - very likely in predictive maintenance - where are compared two feature engineering techniques. The first one considers directly the raw signals as input of a SVM algorithm. In the second, time series are subjected to an adaptive heuristic procedure of piece-wise approximation, whose output is a sequence of$\mathbb{R}^{2}$vectors (slopes and intercepts). In this case, the classification task is carried out in the so-called “dissimilarity space” for pattern recognition adopting different dimensions of the representation set obtained through a clustering algorithm. The dissimilarity measure consists of an ad-hoc edit distance capable of measuring the dissimilarity between 2-dimensional sequences. In this study a k-medoids clustering procedure is adopted for balancing the dataset together with further additional techniques for solving the challenging problem of unbalanced data, offering a deep comparison related to various experimental methodologies.

Abstract Machine Learning is currently a well-suited approach widely adopted for solving data-driven problems in predictive maintenance. Data-driven approaches can be used as the main building block in risk-based assessment and analysis tools for Transmission and Distribution System Operators in modern Smart Grids. For this purpose, a suitable Decision Support System should be able of providing not only early warnings, such as the detection of faults in real time, but even an accurate probability estimate of outages and failures. In other words, the performance of classification systems, at least in these cases, needs to be assessed even in terms of reliable outputting posterior probabilities, a really important feature that, in general, classifiers very often do not offer. In this paper are compared several state-of-the-art calibration techniques along with a set of simple new proposed techniques, with the aim of calibrating fuzzy scoring values of a custom-made evolutionary-cluster-based hybrid classifier trained on a set of a real-world dataset of faults collected within the power grid that feeds the city of Rome, Italy. Comparison results show that in real-world cases calibration techniques need to be assessed carefully depending on the scores distribution and the proposed techniques are a valid alternative to the ones existing in the technical literature in terms of calibration performance, computational efficiency and flexibility.

Graph embedding is an established and popular approach when designing graph-based pattern recognition systems. Amongst the several strategies, in the last ten years, Granular Computing emerged as a promising framework for structural pattern recognition. In the late 2000’s, symbolic histograms have been proposed as the driving force in order to perform the graph embedding procedure by counting the number of times each granule of information appears in the graph to be embedded. Similarly to a bag-of-words representation of a text corpora, symbolic histograms have been originally conceived as integer-valued vectorial representation of the graphs. In this paper, we propose six ‘relaxed’ versions of symbolic histograms, where the proper dissimilarity values between the information granules and the constituent parts of the graph to be embedded are taken into account, information which is discarded in the original symbolic histogram formulation due to the hard-limited nature of the counting procedure. Experimental results on six open-access datasets of fully-labelled graphs show comparable performance in terms of classification accuracy with respect to the original symbolic histograms (average accuracy shift ranging from -7% to +2%), counterbalanced by a great improvement in terms of number of resulting information granules, hence number of features in the embedding space (up to 75% less features, on average).

In this paper, we propose and discuss a prototypical framework for graph classification. The proposed algorithm (Graph E-ABC) exploits a multi-agent design, where swarm of agents (orchestrated via evolutionary optimization) are in charge of finding meaningful substructures from the training data. The resulting set of substructures compose the pivotal entities for a graph embedding procedure that allows to move the pattern recognition problem from the graph domain towards the Euclidean space. In order to improve the learning capabilities, the pivotal substructures undergo an independent optimization procedure. The performances of Graph E-ABC are addressed via a sensitivity analysis over its critical parameters and compared against current approaches for graph classification. Results on five open access datasets of fully labelled graphs show interesting performances in terms of accuracy, counterbalanced by a relatively high number of pivotal substructures.

This paper describes the architecture and control design of an autonomous Electric Boat, together with a specific simulation environment for training and testing the Fuzzy Inference Systems. The boat will be in charge to exit and enter from harbors, plan and follow a route, avoid obstacles such as other boats, correct its motion, perform a virtual anchor and switch between these operations autonomously. The boat is equipped with a set of smart sensors such as sonars, a Global Positioning System, a camera-based vision system and an Inertial Measurement Unit. General navigation rules are respected during the route. We propose an architecture integrating several Fuzzy Controller-based modular pipelines. Furthermore, we propose a mathematical formalization of the Fish Schooling Behavior useful for training Fuzzy Controllers through Q-Learning. Our architecture will soon be implemented on a real boat intended for navigating in inland waters.

The most distinctive trait in structural pattern recognition in graph domain is the ability to deal with the organization and relations between the constituent entities of the pattern. Even if this can be convenient and/or necessary in many contexts, most of the state-of the art classification techniques can not be deployed directly in the graph domain without first embedding graph patterns towards a metric space. Granular Computing is a powerful information processing paradigm that can be employed in order to drive the synthesis of automatic embedding spaces from structured domains. In this paper we investigate several classification techniques starting from Granular Computing-based embedding procedures and provide a thorough overview in terms of model complexity, embedding space complexity and performances on several open-access datasets for graph classification. We witness that certain classification techniques perform poorly both from the point of view of complexity and learning performances as the case of non-linear SVM, suggesting that high dimensionality of the synthesized embedding space can negatively affect the effectiveness of these approaches. On the other hand, linear support vector machines, neuro-fuzzy networks and nearest neighbour classifiers have comparable performances in terms of accuracy, with second being the most competitive in terms of structural complexity and the latter being the most competitive in terms of embedding space dimensionality.

Text mining and text classification are gaining more and more importance in AI related research fields. Researchers are particularly focused on classification systems, based on structured data (such as sequences or graphs), facing the challenge of synthesizing interpretable models, exploiting gray-box approaches. In this paper, a novel gray-box text classifier is presented. Documents to be classified are split into their constituent words, or tokens. Groups of frequent m tokens (or m-grams) are suitably mined adopting the Granular Computing framework. By fastText algorithm, each token is encoded in a real-valued vector and a custom-based dissimilarity measure, grounded on the Edit family, is designed specifically to deal with m-grams. Through a clustering procedure the most representative m-grams, pertaining the corpus of documents, are extrapolated and arranged into a Symbolic Histogram representation. The latter allows embedding documents in a well-suited real-valued space in which a standard classifier, such as SVM, can safety operate. Along with the classification procedure, an Evolutionary Algorithm is in charge of performing features selection, which is able to select most relevant symbols – m-grams – for each class. This study shows how symbols can be fruitfully interpreted, allowing an interesting knowledge discovery procedure, in lights with the new requirements of modern explainable AI systems. The effectiveness of the proposed algorithm has been proved through a set of experiments on paper abstracts classification and SMS spam detection.

The need for efficient integration of an Electric Vehicles (EVs) public transportation system into Smart Grids (SGs), has sparked the idea to equip them with Renewable Energy Systems (RESs), in order to reduce their impact on the SG. As a consequence, an EV can be seen as a Nanogrid (NG) whose energy flows are optimized by an Energy Management System (EMS). In this work, an EMS for an electric boat is synthesized by a Fuzzy Inference System-Hierarchical Genetic Algorithm (FIS-HGA). The electric boat follows cyclic routes day by day. Thus, single day training and test sets with a very short time step are chosen, with the aim of reducing the computational cost, without affecting accuracy. A convex optimization algorithm is applied for benchmark tests. Results show that the EMS successfully performs the EV energy flows optimization. It is remarkable that the EMS achieves good performances when tested on different days than the one it has been trained on, further reducing the computational cost.

With the breakthrough of pervasive advanced networking infrastructures and paradigms such as 5G and IoT, cybersecurity became an active and crucial field in the last years. Furthermore, machine learning techniques are gaining more and more attention as prospective tools for mining of (possibly malicious) packet traces and automatic synthesis of network intrusion detection systems. In this work, we propose a modular ensemble of classifiers for spotting malicious attacks on Wi-Fi networks. Each classifier in the ensemble is tailored to characterize a given attack class and is individually optimized by means of a genetic algorithm wrapper with the dual goal of hyper-parameters tuning and retaining only relevant features for a specific attack class. Our approach also considers a novel false alarm management procedure thanks to a proper reliability measure formulation. The proposed system has been tested on the well-known AWID dataset, showing performances comparable with other state of the art works both in terms of accuracy and knowledge discovery capabilities. Our system is also characterized by a modular design of the classification model, allowing to include new possible attack classes in an efficient way.

Modeling and predicting failures in the field of predictive maintenance is a challenging task. An important issue of an intelligent predictive maintenance system, exploited also for Condition Based Maintenance applications, is the failure probability estimation that can be found uncalibrated for most standard and custom classifiers grounded on Machine learning. In this paper are compared two classification techniques on a data set of faults collected in the real-world power grid that feeds the city of Rome, one based on a hybrid evolutionary-clustering technique, the other based on the well-known Gaussian Mixture Models setting. While the former adopts directly a custom-based weighted dissimilarity measure for facing unstructured and heterogeneous data, the latter needs a specific embedding technique step performed before the training procedure. Results show that both approaches reach good results with a different way of synthesizing a model of faults and with different structural complexities. Furthermore, besides the classification results, it is offered a comparison of the calibration status of the estimated probabilities of both classifiers, which can be a bottleneck for further applications and needs to be measured carefully

The most fascinating aspect of graphs is their ability to encode the information contained in the inner structural organization between its constituting elements. Learning from graphs belong to the so-called Structural Pattern Recognition, from which Graph Embedding emerged as a successful method for processing graphs by evaluating their dissimilarity in a suitable geometric space. In this paper, we investigate the possibility to perform the embedding into a geometric space by leveraging to peculiar constituent graph substructures extracted from training set, namely the maximal cliques, and providing the performances obtained under three main aspects concerning classification capabilities, running times and model complexity. Thanks to a Granular Computing approach, the employed methodology can be seen as a powerful framework able to synthesize models suitable to be interpreted by field-experts, pushing the boundary towards new frontiers in the field of explainable AI and knowledge discovery also in big data contexts.

Multi-agent systems recently gained a lot of attention for solving machine learning and data mining problems. Furthermore, their peculiar divide-and-conquer approach is appealing when large datasets have to be analyzed. In this paper, we propose a multi-agent classification system able to tackle large datasets where each agent independently explores a random small portion of the overall dataset, searching for meaningful clusters in proper subspaces where they are well-formed (i.e., compact and populated). This search is orchestrated by means of a genetic algorithm able to act in a multi-modal fashion, since meaningful clusters might lie in different subspaces. Furthermore, since agents operate independently one another, their execution is parallelized across different computational units. Tests show that the proposed algorithm, E-ABC2, is able to deal with large datasets, returning satisfactory results in terms of scalability and performances, especially when compared with our previous baseline versions.

Embedding spaces are one of the mainstream approaches when dealing with structured data. Granular Computing, in the last decade, emerged as a powerful paradigm for the automatic synthesis of embedding spaces that, at the same time, yield an interpretable model on the top of meaningful entities known as "information granules". Usually, in these contexts, one aims at finding the smallest set of information granules in order to boost the model interpretability while keeping satisfactory performances. In this paper, we add a third objective, namely the structural complexity of the resulting model and we exploit three biology-related case studies related to metabolic networks and protein networks in order to investigate the link between classification performances, embedding space dimensionality and structural complexity of the resulting model.

Natural language processing and text mining applications have gained a growing attention and diffusion in the computer science and machine learning communities. In this work, a new embedding scheme is proposed for solving text classification problems. The embedding scheme relies on a statistical assessment of relevant words within a corpus using a compound index originally proposed in ecology: this allows to spot relevant parts of the overall text (e.g., words) on the top of which the embedding is performed following a Granular Computing approach. The employment of statistically meaningful words not only eases the computational burden and the embedding space dimensionality, but also returns a more interpretable model. Our approach is tested on both synthetic datasets and benchmark datasets against well-known embedding techniques, with remarkable results both in terms of performances and computational complexity.

emissions and global warming. In this context, an effective use of electrochemical energy storage systems (ESSs) is mandatory. In particular, accurate state of charge (SoC) estimations are helpful for improving the ESS performances. To this aim, developing accurate models of electrochemical cells is necessary for implementing effective SoC estimators. Therefore, a novel neural network modeling technique is proposed in this paper. The main contribution consists in the development of a white-box neural design that provides helpful insights into the cell physics, together with a powerful nonlinear approximation capability, and a flexible system identification procedure. In order to do that, the system equations of a white-box equivalent circuit model (ECM) have been combined with computational intelligence techniques by approximating each circuit element with a dedicated neural network. The model performances have been analyzed in terms of model accuracy, SoC estimation effectiveness, and computational cost over two realistic data sets. Moreover, the proposed model has been compared with a white-box ECM and a gray-box neural network model. The results prove that the proposed modeling technique is able to provide useful improvements in the SoC estimation task with a competing computational cost.

Graphs are data structures able to efficiently describe real-world systems and, as such, have been extensively used in recent years by many branches of science, including machine learning engineering. However, the design of efficient graph-based pattern recognition systems is bottlenecked by the intrinsic problem of how to properly match two graphs. In this paper, we investigate a granular computing approach for the design of a general purpose graph-based classification system. The overall framework relies on the extraction of meaningful pivotal substructures on the top of which an embedding space can be build and in which the classification can be performed without limitations. Due to its importance, we address whether information can be preserved by performing stochastic extraction on the training data instead of performing an exhaustive extraction procedure which is likely to be unfeasible for large datasets. Tests on benchmark datasets show that stochastic extraction can lead to a meaningful set of pivotal substructures with a much lower memory footprint and overall computational burden, making the proposed strategies suitable also for dealing with big datasets.

Calibrating a classification system consists in transforming the output scores, which somehow state the confidence of the classifier regarding the predicted output, into proper probability estimates. Having a well-calibrated classifier has a non-negligible impact on many real-world applications, for example decision making systems synthesis for anomaly detection/fault prediction. In such industrial scenarios, risk assessment is certainly related to costs which must be covered. In this paper we review three state-of-the-art calibration techniques (Platt’s Scaling, Isotonic Regression and SplineCalib) and we propose three lightweight procedures based on a plain fitting of the reliability diagram. Computational results show that the three proposed techniques have comparable performances with respect to the three state-of-the-art approaches.

Accurate modeling of electrochemical cells is nowadays mandatory for achieving effective upgrades in the fields of energetic efficiency and sustainable mobility. Indeed, these models are often used for performing accurate State-of-Charge (SoC) estimations in energy storage systems used in microgrids or powering pure electric and hybrid cars. To this aim, a novel neural networks ensemble approach for modeling electrochemical cells is proposed in this paper. Herein, the system identification has been faced by means of a gray box technique, in which different and specialized neural networks are used for identifying the unknown internal behaviors of the cell. In particular, the a priori knowledge on the system dynamic is used for defining the network architecture. Specifically, each nonlinear function appearing in the system equations is approximated by a distinct neural network. The proposed model has been validated upon three different data sets both in terms of model accuracy and effectiveness in the SoC estimation task. The achieved performances have been compared with those of other computational intelligence approaches proposed in the literature. The results prove the effectiveness of the gray box scheme, achieving very promising performances in both the system identification accuracy and the SoC estimation task.

The World energy demand is progressively growing, so that many different Renewable Energy Sources (RESs) are exploited to meet the user needs and to reduce the Global Warming. In this context, an emerging RES is the sea wave energy, because it can offer a better continuity in the energy production by taking advantage of the stationary nature of the waves. Research in the energy harvesting from waves has led to the development of Wave Energy Converters (WECs). The adoption of Computational Intelligence techniques become crucial for maximizing the WECs efficiency. Therefore, the optimization of the energy transduction of a specific WEC model has been investigated in this paper. More precisely, three different Evolutionary Algorithms (EAs), namely a Genetic Algorithm (GA), a Particle Swarm optimization (PSO) and a Hybrid Genetic PSO (HG-PSO), have been considered to determine the optimal values of the main WEC parameters. Both the effectiveness and the efficiency of the above algorithms have been tested aiming at finding which of them is most suitable for the considered problem. The obtained results showed promising performances, with all the three algorithms achieving effective and robust solutions. In particular, the HG-PSO proved to be the most suitable approach, being the most effective and efficient algorithm even in front of a more rigid stop condition.

Due to the increasing amount of sensors and data streams that can be collected in order to monitor electric distribution networks, developing predictive diagnostic systems over Smart Grids demands powerful and scalable algorithms in order to search for regularities in Big Data. In this regards, Evolutive Agent Based Clustering (E-ABC) is a promising framing reference, as it is conceived to orchestrate a swarm of intelligent agents acting as individuals of an evolving population, each performing a random walk on a different subset of patterns. Each agent is in charge of discovering well-formed (compact and populated) clusters and, at the same time, a suitable subset of features corresponding to the subspace where such clusters lie, following a local metric learning approach, where each cluster is characterized by its own subset of relevant features. E-ABC is able to process data belonging to structured and possibly non metric spaces, relying on custom parametric dissimilarity measures. In this paper, a supervised version of E-ABC is proposed. This novel classification system has been employed for recognizing and predicting localized faults on the electric distribution network of Rome, managed by the Italian utility company ACEA. Tests results show that E-ABC is able to synthesize classification models characterized by a remarkable generalization capability, with adequate performances to be employed in Smart Grids condition based management systems. Moreover, the feature subsets where most of the meaningful clusters have been discovered can be used to better understand sub-classes of failures, each identified by a set of related causes.

Microgrids (MGs) development is one of the most pursued solution for the electric grid modernization into smart grids, as an effective approach to achieve the European Funding Program Horizon 2020 targets. In particular, residential grid-connected MGs demand the active role of the customer into the electric market, the fulfilment of Demand Response (DR) services, and a local control of the distribution energy balance. In order to take advantage of the local production (such as a photovoltaic generator) and of the Energy Storage System (ESS), a MG needs an Energy Management System (EMS) able to decide how to efficiently redistribute in real time the energy flows among the energy systems, in order to satisfy the customer needs which are expressed through a suitable objective function. This work focuses on a new version of an Adaptive Neural Fuzzy Inference System (ANFIS) as the core of a MG EMS, supported by an Echo State Network (ESN) based predictor. The overall training algorithm is designed to maximize the profit generated by the energy exchange with the grid, by assuming a Time Of Use (TOU) energy price policy. The main objective of this work is focused on studying the impact of the prediction system on the EMS performances. Results show that EMS performances improve of about 30% for prediction time horizons over 10 hours.

The development of electrochemical cell models and of the related system identification procedures are of utmost importance for achieving effective management of electrochemical Energy Storage Systems. Specifically, accurate models are mandatory for performing effective estimation of the State of Charge (SoC) by means of Kalman Filtering approaches. Currently, some of the most promising models are those based on the equivalent circuit technique. However, these models are based on the standard definition of the SoC, which is related to the integral of the input current. The main drawback of this approach is that it is not related to any physical characteristic of the cell. A first implementation of an equivalent circuit model based on a novel mechanical inspired definition of the SoC is proposed in this paper. Herein, the cell is assimilated to a liquid tank so that the SoC, can be defined with respect to the reservoir shape. In particular, the charge reservoir has been modeled by means of an ANFIS estimator. A flexible system identification procedure has been defined by formulating a fitting problem upon generic sequences of the measured current and voltage. Specifically, a customized Particle Swarm optimization is in charge both of training the ANFIS and of identifying the other model parameters. The proposed modeling approach has been tested upon the Randomized Battery Usage Data Set and the obtained results proved its effectiveness both in emulating the cell behavior and in the SoC estimation task.

An effective management of Electrochemical Energy Storage Systems (ESSs) is nowadays of utmost importance for the technological evolution in both automotive and sustainable power networks applications. In particular, Battery Managements Systems (BMSs) are the electronic devices devolved to this management. One of the most important task of any BMS is cells balancing, aiming at leveling the operating points of the cells composing the ESS. Therefore, a novel online balancing algorithm is proposed in this work. Differently to the most commonly used methods, the proposed approach works by leveling the State of Charge (SoC) of the cells instead of their voltages. The balancing procedure has been formulated as a zero-one integer programming to be solved online by means of a Hybrid Genetic Binary Particle Swarm optimization (BPSO). Furthermore, a sparsity regularization has been considered for improving the energetic efficiency of the algorithm. Both the baseline and the regularized balancing systems have been tested and compared with a standard voltage based approach. The results show that the proposed method achieves a better and more robust balancing of the ESS, keeping a comparable energetic efficiency with respect to the voltage based technique.

In this paper we discuss techniques for potential speedups in k-medoids clustering. Specifically, we address the advantages of pre-caching the pairwise distance matrix, heart of the k-medoids clustering algorithm, not only in order to speedup the execution of the algorithm itself, but also in order to speedup the evaluation of the well-known Silhouette Index and DaviesBouldin Index for clusters' validation. A major disadvantage of such pre-caching is that it might not be suitable for large datasets. To this end, a further contribution consists in proposing parallel and distributed implementations of both the Simplified Silhouette Index and the Davies-Bouldin Index for distributed k-clustering using the Apache Spark framework. Results on real-world pathway maps datasets show the robustness of such distributed implementations, also underlining their effectiveness for structured data.

Topological Data Analysis is a novel approach, useful whenever data can be described by topological structures such as graphs. The aim of this paper is to investigate whether such tool can be used in order to define a set of descriptors useful for pattern recognition and machine learning tasks. Specifically, we consider a supervised learning problem with the final goal of predicting proteins' physiological function starting from their respective residue contact network. Indeed, folded proteins can effectively be described by graphs, making them a useful case-study for assessing Topological Data Analysis effectiveness concerning pattern recognition tasks. Experiments conducted on a subset of the Escherichia coli proteome using two different classification systems show that descriptors derived from Topological Data Analysis - namely, the Betti numbers sequence - lead to classification performances comparable with descriptors derived from widely-known centrality measures, as concerns the protein function prediction problem. Further benchmarking tests suggest the presence of some information despite the heavy compression intrinsic to the protein-to-Betti numbers casting.

Dissimilarity spaces, along with feature reduction/ selection techniques, are among the mainstream approaches when dealing with pattern recognition problems in structured (and possibly non-metric) domains. In this work, we aim at investigating dissimilarity space representations in a biology-related application, namely protein function classification, as proteins are a seminal example of structured data given their primary and tertiary structures. Specifically, we propose two different analyses relying on both the complete dissimilarity matrix and a dimensionally-reduced version of the complete dissimilarity matrix, thereby casting the pattern recognition problem from structured domains towards real-valued feature vectors, for which any standard classification algorithm can be used. A third, hybrid, analysis uses a clustering-based one-class classifier exploiting different representations. First results conducted on a subset of the Escherichia coli proteome are promising and some of the analyses presented in this work may also dually suit field-experts, further bridging the gap between natural sciences and computational intelligence techniques.

In this paper we present an interesting application of the Decision Support System, known as the OCC_System, designed for faults recognition and classification within the real-world Medium Voltage power grid of Rome, Italy, managed by the Azienda Comunale Energia e Ambiente (ACEA) company. Given a historical data set consisting of fault patterns described by heterogeneous features related to endogenous and exogenous factors, the recognition system is trained to classify fault states assigning them a probability of fault. Disambiguating the external causes, whose dynamic is fast, and the constitutive parameters of the power grid, whose dynamic is slow, an affine algebraic model is estimated through an evolutionary technique in order to obtain a vulnerability index related to power grid equipment. The estimation procedure allows obtaining even a correlation matrix among external causes and constitutive parameters useful to better characterize the fault phenomenon under study.

The worldwide power grid can be thought as a System of Systems deeply embedded in a time-varying, non-deterministic and stochastic environment. The availability of ubiquitous and pervasive technology about heterogeneous data gathering and information processing in the Smart Grids allows new methodologies to face the challenging task of fault detection and modeling. In this study, a fault recognition system for Medium Voltage feeders operational in the power grid in Rome, Italy, is presented. The recognition task is performed synthesizing a data-driven model of fault phenomenons based on a hybridization of Evolutionary learning and Clustering techniques. The model is synthesized starting from a set of clusters obtained by partitioning the fault patterns, tuning at the same time the core dissimilarity measure. In this paper we show as clusters can be successively analyzed for mining useful information about the fault phenomenon and to build up an ad-hoc decision system to support business strategies such as Condition Based Maintenance tasks.

The efficient integration of Renewable Energy Sources (RES) in the actual electrical grid has gained recently a high attention in the Smart Grids (SGs) research topic. The evolution of existing electric distribution networks into SGs can be accomplished gradually and conveniently through the installation of local grid-connected Microgrids (MGs), usually installed nearby the RESs and provided by Energy Storage Systems (ESSs). Each MG is in charge to manage connected RES, assuring the local power demand, as well as the safety and stability of the electric grid. To this aim, the Energy Management System (EMS) must provide intelligent decision making in fixing both MG configuration and energy flows between each subsystem in real time, according to some objective functions. In this work, it is proposed a MG EMS based on a Fuzzy Inference System (FIS) optimized through a custom implementation of Multi Objective Genetic Algorithm (MO-GA). In particular, the EMS is based on a three inputs FIS and it has been designed in order to reduce the fluctuations of energy exchanged with the grid (i.e. the grid stress) and to maximize the energy auto-consumption by employing an efficient utilization of the ESS. Results show that it is possible to improve considerably the auto-consumption performance, and at the same time to reduce grid stress, improving peak shaving concerning the maximum power request from the main grid.

Microgrids (MGs) play a crucial role for the development of Smart Grids. They are conceived to intelligently integrate the generation from Distributed Energy Resources, to improve Demand Response (DR) services, to reduce pollutant emissions and curtail power losses, assuring the continuity of services to the loads as well. In this work it is proposed a novel synthesis procedure for modelling an Adaptive Neuro-Fuzzy Inference System (ANFIS) featured by multivariate Gaussian Membership Functions (MFs) and first order Takagi-Sugeno rules. The Fuzzy Rule Base is the core inference engine of an Energy Management System (EMS) for a grid-connected MG equipped with a photovoltaic power plant, an aggregated load and an Energy Storage System (ESS). The EMS is designed to operate in real time by defining the ESS energy flow in order to maximize the revenues generated by the energy trade with the distribution grid. The ANFIS EMS is synthesized through a data driven approach that relies on a clustering algorithm which defines the MFs and the rule consequent hyperplanes. Moreover, three clustering algorithms are investigated. Results show that the adoption of k-medoids based on Mahalanobis (dis)similarity measure is more efficient with respect to the k-means, although affected by some variety in clusters composition.

In this paper, we propose a novel implementation for solving the large-scale k-medoids clustering problem. Conversely to the most famous k-means, k-medoids suffers from a computationally intensive phase for medoids evaluation, whose complexity is quadratic in space and time; thus solving this task for large datasets and, specifically, for large clusters might be unfeasible. In order to overcome this problem, we propose two alternatives for medoids update, one exact method and one approximate method: the former based on solving, in a distributed fashion, the quadratic medoid update problem; the latter based on a scan and replacement procedure. We implemented and tested our approach using the Apache Spark framework for parallel and distributed processing on several datasets of increasing dimensions, both in terms of patterns and dimensionality, and computational results show that both approaches are efficient and effective, able to converge to the same solutions provided by state-of-the-art k-medoids implementations and, at the same time, able to scale very well as the dataset size and/or number of working units increase.

In this paper, an automatic music transcription (AMT) algorithm based on a supervised non-negative matrix decomposition (NMD) is discussed. In particular, a novel approach for enhancing the sparsity of the solution is proposed. It consists of a two-step processing in which the NMD is solved joining a ℓ2regularization and a threshold filtering. In the first step, the NMD is performed with the ℓ2regularization in order to get an overall selection of the notes most likely appearing in the monotimbral musical excerpt. In the second step, a threshold filtering followed by another ℓ2regularized NMD are repeatedly performed in order to progressively reduce the dictionary matrix and to refine the notes transcription. Furthermore, a user-oriented instrument learning procedure has been conceived and proposed. The proposed AMT system has been tested upon the dataset collected by the LabROSA laboratories considering the transcription of three different pianos. Moreover, it has been validated through a comparison with a regularized NMD and with three open source AMT software. The results prove the effectiveness of the proposed two-step processing in enhancing the sparsity of the solution and in improving the transcription accuracy. Moreover, the proposed system shows promising performance in both multi-F0 and note tracking tasks, obtaining better transcription accuracy than the competing algorithms in most tests.

This paper presents a novel power flow optimization strategy in Micro Grids (MGs) connected to the main grid. When the MG includes stochastic energy sources, such as photovoltaic and micro eolic-generators, it is very useful to rely on Energy Storage Systems (ESSs) to buffer energy. In fact, an ESS can be employed to perform several functionalities, related to different user requirements, such as power stability, peak shaving, optimal energy trading, etc. The Energy Management System is based on a Fuzzy Logic Controller (FLC) optimized by a Multi-Objective Genetic Algorithm in order to maximize both the total profit in energy trading with the main grid and the State of Health (SOH) of the ESS. The FLC manages the neat aggregate energy deficit and surplus inside the MG, analyzing in real time the state of the MG (aggregated energy demand and production, State of Charge of the ESS, energy sale and purchase prices). The FLC is tested on a MG composed by a photovoltaic solar generator, a domestic user and a Li-ion battery. A multi-objective genetic algorithm is in charge to find the set of solutions on the Pareto front. The results are compared with the same FLC optimized by a mono-objective Genetic Algorithm (GA) minimizing in a first case only the total profit and in the second case a convex linear combination of the total profit and a measure of the battery stress.

Nowadays an effective Energy Storage System (ESS) is a fundamental requirement for any effective innovation in the fields of energetic and transportation sustainability. One of the most important device for obtaining efficient ESSs is the Battery Management System (BMS). It includes all the electronic components and algorithms for the monitoring and management of the ESS status. The key task of the BMS is the estimation of the State of Charge. Currently the most promising methods are based on state observers, which require an accurate model of the cell. In this paper a novel technique for modeling the transient behavior of the cell is proposed. It is based on a single nonlinear dipole composed of a standard linear resistor and a nonlinear voltage driven capacitor connected in parallel. A method for the parameters identification based on a Particle Swarm Optimization algorithm has been developed. Both the identification algorithm and the proposed model are validated on a A123 cell obtaining very stable solution and better accuracy with respect to models based on linear components.

This paper presents a novel power flow optimization strategy for a Grid Connected microgrid (MG) equipped with a Battery Energy Storage System (BESS), namely a Li-Ion battery pack. A BESS can be employed to perform several functionalities, related to different user requirements, such as power stability, peak shaving, optimal energy trading, etc. In the proposed system the MG is composed by an aggregation of distributed power generators and loads and a BESS is adopted to manage the power over-production/over-demand in real time, in order to maximize the prosumer profit looking at the current energy prices and the BESS State of Charge (SOC). The Energy Management System (EMS) is based on a Fuzzy Logic Controller (FLC) with a suitable rule inference system designed by an Expert Operator (EO). The control strategy is tested with different power profiles and BESS capacities in order to verify its effectiveness and limits. Furthermore, the FLC has been optimized by a Genetic Algorithm to increase the total profit exploiting the BESS as energy buffer. The optimization results have been compared to the initial FLC designed by the EO, taking into account both the profit and the deterioration of the BESS measured through a suitable battery stress index.

Energy Storage Systems (ESS)s have become widely pervasive in several sectors, both in the civil and in the industrial fields. Among the several applications, two of the most critical concern energy storing in the future Smart Grids and microgrids and power sourcing for Electric and Hybrid Vehicles. In this context, the management of the ESS represents a crucial task in order to guarantee efficient, effective and robust energy storing. The Battery Management System (BMS) is the device designated for performing this management. It has to avoid damages to the cell, to estimate the State of Charge (SoC), the State of Health (SoH) and to perform the cell equalization. In this paper, the SoC estimation by means of state observers has been investigated. In particular, the performances obtained by the Extended Kalman Filter (EKF) and by the Square Root Unscented Kalman Filter (SR-UKF) have been compared on a prototypal BMS. Results show that the SR-UKF succeeds in coping with the nonlinearities of the battery, obtaining better and more robust estimations than the classic EKF.

Granulation of information by using type-2 fuzzy sets is receiving more attention nowadays. This is due to the superior capability of type-2 fuzzy sets in handling the data uncertainty. From a theoretical perspective, a set containing type-2 fuzzy sets has no trivial geometric structure; therefore, a proper metric cannot be easily defined. As a consequence, common pattern recognition systems, which in one way or another rely on some (geo)metric structure of the input space, are not easily applicable to a space of type-2 fuzzy sets. In this paper, we study the problem of designing a classifier in the input space of type-2 fuzzy sets. Type-2 fuzzy sets are hence interpreted as (granular) patterns forming a given input dataset. By decomposing a type-2 fuzzy set into a sequence of simpler (lower type) fuzzy sets, we explore the possibility of defining and building dissimilarity and kernel-based classification systems on input spaces of type-2 fuzzy sets. Such an interpretation provided in terms of sequences allows us to conceive an effective sequence matching strategy, which can be suitably embedded into well-established pattern recognition systems. We support the methodological developments by performing experiments on synthetically generated classification problems for datasets composed of type-2 fuzzy sets, with adjustable and controlled level of difficulty. Results are promising and suggest to further investigate on the possibility of interpreting type-2 fuzzy sets as input patterns of a given data-driven inference system.

The analysis and recognition of fault status in the Smart Grid field is a challenging problem. Computational Intelligence techniques have already been shown to be a successful framework to face complex problems related to a Smart Grid. The availability of huge amounts of data coming from smart sensors allows the system to take a fine grained picture of the power grid status. This data can be processed in order to offer an instrument in aiding humans operators to better understand the power grid status and to take decisions on grid operations. This paper addresses the problem of fault recognitions in a real-world power grid (i. e. the power grid that feds the city of Rome, Italy) with the One-Class Classification paradigm by a combined approach of dissimilarity measure learning by means of an evolution strategy and clustering techniques for modeling the decision regions between fault status and the standard functioning of the power system. In this paper we present an in-depth study of the performance of two clustering algorithms in building up the model of faults, as the core procedure of the proposed recognition system.

This paper presents a universal methodology for generating an interval type-2 fuzzy set membership function from a collection of type-1 fuzzy sets. The key idea of the proposed methodology is to designate a specific type-1 fuzzy set as the representative of all input type-1 fuzzy sets. To this end, we use a novel measure of similarity between type-1 fuzzy sets, which relies on both kernel functions and fuzzy information processing methods. Based on the selected representative type-1 fuzzy set, and with respect to the principle of justifiable granularity, an interval type-2 fuzzy set is then formed. The results of the conducted experiments demonstrate the effectiveness of the proposed methodology for generating sound interval type-2 fuzzy sets.

The generalization capability is usually recognized as the most desired feature of data-driven learning systems, such as classifiers. However, in many practical applications obtaining human-understandable information, relevant to the problem at hand, from the classidication model can be equally important. In this paper we propose a classification system able to fulfill these two requirements simultaneously for a generic image classification task. As a first preprocessing step, an input image to the classifier is represented by a labeled graph, relying on a segmentation algorithm. The graph is conceived to represent visual and topological information of the relevant segments of the image. Then, the graph is classified by a suited inductive inference engine. In the learning procedure all the training set images are represented by graphs, feeding a state-of-the-art classification system working on structured domains. The synthesis procedure consists in extracting characterizing subgraphs from the training set, which are used to embed the graphs into a vector space, enabling thus the applicability of well-known classifiers for feature-based patterns. Such characterizing subgraphs, which are derived in an unsupervised fashion, are interpretable by suitable field experts, allowing a semantic analysis of the discovered classification rules for the given problem at hand. The system is optimized with a genetic algorithm, which tunes the system parameters according to a cross-validation scheme. We show the validity of the approach by performing experiments considering some image classification problems derived from an on-line repository.

Due to the intrinsic complexity of real-world power distribution lines, which are highly non-linear and time-varying systems, modeling and predicting a general fault instance is a very challenging task. Power outages can be experienced as a consequence of a multitude of causes, such as damage of some physical components or grid overloads. Smart grids are equipped with sensors that enable continuous monitoring of the grid status, hence allowing the realization of control systems related to different optimization tasks, which can be effectively faced by Computational Intelligence techniques. This paper deals with the problem of faults modeling and recognition in a real-world smart grid, located in the city of Rome, Italy. It is proposed a suitable classication system able to recognize faults on medium voltage feeders. Due to the nature of the available data, the one-class classication framework is adopted. Experiments are presented and discussed considering a three-year period of measurements of fault events gathered by ACEA Distribuzione S.p.A., the company that manages the smart grid system under analysis. Results demonstrate the effectiveness and validity of our approach.

In this paper we propose an inexact graph matching algorithm which computes the dissimilarity of a time-varying labeled graph with respect to a static one. This approach is specifically designed for processing very large labeled graphs, which are subject to frequent edit operations that modify the topology and the labeling of restricted zones of the graph. In this scenario, repeating each time an extensive computation of the whole dissimilarity value would require too much time; moreover, since only a specific part of the graph changes, it would result also in a waste of computations. We propose a fast approach for computing the graph dissimilarity which exploits the dissimilarity value estimated in the previous time interval and the nature of the observed edit operations. We evaluate the properties of the proposed approach with respect to well-known graph matching algorithms, by simulating the dynamics of the graph. Overall, the experiments confirm the effectiveness of the approach.

We propose two variants of a general-purpose graph classification system which rely on a theoretical result that we prove in this paper. The result allows us to solve analytically the setting of a sequential clustering algorithm that is used for compressing the input labeled graphs represented in the dissimilarity space. As a consequence, we achieve a considerable asymptotic and practical speed-up of the overall classification system, maintaining state-of-the-art results in terms of test set classification accuracy on well-known benchmarking datasets of labeled graphs. The obtained speed-up makes the system one step closer towards the applicability to bigger labeled graphs and larger datasets.

Text categorization is an interesting application of machine learning covering a wide range of possible applications, from document management systems to web mining. In designing such a system it is mandatory to correctly define both a suited preprocessing procedure and an effective document representation as closely related as possible to the semantic nature of document categories. To this aim, relying on a Granular Computing approach and considering a document as an ordered sequence of words, we propose a system able to automatically mine frequent terms, considering as a term not only a single word, but also a subsequence of (a few) consecutive words. The whole classification system is tailored to process sequences of atomic elements (i.e., encoded words) by means of an embedding procedure based on clustering methods. However, when dealing with unbalanced data sets, i.e. when classes are not evenly represented in the data set, the frequent substructures search procedure must be carefully designed. We prove the effectiveness of the system over a well-known benchmarking data set, achieving competitive test set classification accuracy results, with a remarkable low structural complexity of the synthesized classification models.

In this paper we propose a classifier for generalized sequences that is conceived in the granular computing framework. The classification system processes the input sequences of objects by means of a suited interplay among dissimilarity and clustering based techniques. The core data mining engine retrieves information granules that are used to represent the input sequences as feature vectors. Such a representation allows to deal with the original sequence classification problem through standard pattern recognition tools. We have evaluated the generalization capability of the system in an interesting case study concerning the protein folding problem. In the considered dataset, the entire E. Coli proteome was screened as for the prediction of protein relative solubility on a pure amino acids sequence basis. We report the analysis of the dataset considering different settings, showing interesting test set classification accuracy results. The developed system consents also to extract knowledge from the considered training set, by allowing the analysis of the retrieved information granules.

Traffic flow classification to identify applications and activity of users is widely studied both to understand privacy threats and to support network functions such as usage policies and QoS. For those needs, real time classification is required and classifier's complexity is as important as accuracy, especially given the increasing link speeds also in the access section of the network. We propose the application of a highly efficient classification system, specifically Min-Max neurofuzzy networks trained by PARC algorithm, showing that it achieves very high accuracy, in line with the best performing algorithms onWeka, by considering two traffic data sets collected in different epochs and places. It turns out that required classification model complexity is much lower with Min-Max networks with respect to SVM models, enabling the implementation of effective classification algorithms in real time on inexpensive platforms.

In this paper we face the inexact graph matching problem from the parallel algorithms viewpoint. After a brief introduction of both graph matching and parallel computing contexts, we discuss a specific method of performing inexact graph matching based on the well known tensor product operator. We analyze the problem using two parallel computing models, following different algorithmic strategies, and performing also an experimental evaluation. The aim of this paper is to provide modeling and algorithmic strategies to extend inexact graph matching methods to graphs of high order and size, conceiving the computational problem in the more wider context of graph-based Pattern Recognition and Soft Computing systems. As a whole, the obtained results encourage more effort on this direction.

In this paper we present an innovative procedure for sequence mining and representation. It can be used as its own in Data Mining problems or as the core of a classification system based on a Granular Computing approach to represent sequences in a suited embedding space. By adopting an inexact sequence matching procedure, the algorithm is able to extract a symbols alphabet of frequent subsequences to be used as prototypes for the embedding stage. Experimental evaluation over both synthetically generated and biological datasets confirms that the modeling system is able to synthesize effective models when facing even complex and noisy problems defined by frequency-based classification rules.

In this paper we propose a music Query by Humming System made of two main functional blocks; the first implements a voice-to-midi transcription algorithm to process the query, the second implements a search engine based on a novel template matching technique for Dynamic Time Warping. The voice-to-midi algorithm transforms the sung or hummed query in a MIDI file by segmenting and identifying the notes' sequence. The search engine uses a Template Matching technique to produce a list of possible melodies that best match the searched one. In the test phase, first, we investigated performance of the search engine in retrieval using a synthetic test bench; a set of artificial queries is build placing and adjusting, in the queries, patterns of typical disturbance. Second, we use a genetic algorithm to automatically optimize the performance of the overall system using a real-life test bench. Results highlight that the proposed MIR system has good performances and is robust enough to be employed in real life applications.

Secure SHell is a TCP based protocol designed to enhance with security features telnet and other insecure remote management tools. Due to its versatility, it is often exploited to forward applications (i.e. HTTP, SCP, etc.) into encoded TCP traffic flows. The point which makes challenging the identification of the uses of SSH is that packets are enciphered and instruments based on deep packet inspection (DPI) cannot achieve this task. We approached the problem of early SSH classification with k-means based machine by studying statistical behavior of IP traffic parameters, such as length, arrival time and direction of packets. In this paper we describe tools and networks designed to collect SSH remote administration traffic as well as relevant results obtained for its classification. In particular, our tool identifies remote management traffic out of other SSH encoded appli cations with accuracy up to 90.34.

In this paper we describe an automatic voice-to-MIDI transcription procedure. In particular we propose a note segmentation method based on the analysis of the signal envelope and its derivative. The pitch of the segmented note is extracted with a novel generalized correlation function called correntropy function achieving high accuracy with the same computational cost of traditional correlation based methods. The performances of our transcription system have been measured on examples extracted by some repositories available on internet; they consist in sung melodies and hummed queries. Results show the ability of our transcription procedure to cope with query-by-humming systems, as well as with monophonic singing transcription. Performances can be easily evaluated by downloading from the authors web site the original PCM files and the corresponding MIDI files produced by the proposed transcription algorithm.

This paper deals with the musical genre classification problem, starting from a set of features extracted directly from MPEG-1 layer III compressed audio data. The automatic classification of compressed audio signals into a short hierarchy of musical genres is explored. More specifically, three feature sets for representing timbre, rhythmic content and energy content are proposed for a four leafs tree genre hierarchy. The adopted set of features are computed from the spectral information available in the MPEG decoding stage. The performance and relative importance of the proposed approach is investigated by training a classification model using the audio collections proposed in musical genre contests. We also used an optimization strategy based on genetic algorithms. The results are comparable to those obtained by PCM-based musical genre classification systems.

In this paper we propose a symbolic classification system able to solve automatically a great number of different image classification problems, without any need to adapt the preprocessing procedure to the specific problem instance at hand. The basic idea consists in considering a set of semantically defined objects (symbolic elements) that can be recognized on images. By means of a segmentation procedure, each image is represented by a set of symbolic elements. The inductive inference is performed directly in this symbolic domain through a parametric dissimilarity measure. As shown in this paper, the system is able to adapt the dissimilarity measure to the specific problem, by finding the optimal values of the dissimilarity function parameters. Moreover, a compact model representation can be obtained by representing each cluster with the corresponding set median point.

We propose a low cost prevention strategy for osteoporosis. Osteoporosis is a disease consisting in the structural deterioration of bones. This disease has a very high cost for the public health expense all over the world. Its main diagnostic tool is a radiographic analysis called computerized bone mineralometry, by which it is possible to measure the bone mineral density (BMD). Starting from the BMD value it is possible to estimate the risk of contracting osteoporosis. Although the cost of this clinical analysis is not high, a wide screening of the population can be not affordable. The proposed prevention strategy is based on the assumption that BMD can be estimated by a neural model, on the basis of some objective individual characteristics to be determined by the patient itself. We propose the use of MoG (mixture of Gaussian) neural model, trained by an automatic procedure based on maximum likelihood approach.

Classification can be considered as a basic data driven modeling problem, which allows us to define and design more complex modeling systems. The choice of an adequate classification system should take into account the automation degree of the learning procedure, especially if it must be employed as a core inference engine. Fuzzy min-max neural networks are very effective and flexible classification models, since they easily allow the design of constructive learning techniques, such as the ARC/PARC one. In this paper we propose a classification system able to generate automatically a fuzzy min-max classifier. It holds the capability to optimize both the number of neurons in the hidden layer and the set of features used to classify a pattern, without any knowledge about the test set. Its performances are evaluated through a toy problem and two real data benchmarks.

A high automation degree is one of the most important features of data driven modeling tools and it should be taken into consideration in classification systems design. In this regard, constructive training algorithms are essential to improve the automation degree of a modeling system. Among neuro-fuzzy classifiers, Simpson's (1992) min-max networks have the advantage of being trained in a constructive way. The use of the hyperbox, as a frame on which different membership functions can be tailored, makes the min-max model a flexible tool. However, the original training algorithm evidences some serious drawbacks, together with a low automation degree. In order to overcome these inconveniences, in this paper two new learning algorithms for fuzzy min-max neural classifiers are proposed: the adaptive resolution classifier (ARC) and its pruning version (PARC). ARC/PARC generates a regularized min-max network by a succession of hyperbox cuts. The generalization capability of ARC/PARC technique mostly depends on the adopted cutting strategy. By using a recursive cutting procedure (R-ARC and R-PARC) it is possible to obtain better results. ARC, PARC, R-ARC, and R-PARC are characterized by a high automation degree and allow to achieve networks with a remarkable generalization capability. Their performances are evaluated through a set of toy problems and real data benchmarks. The paper also proposes a suitable index that can be used for the sensitivity analysis of the classification systems under consideration.

A new neuro-fuzzy classifier, inspired by the min-max neural model, is presented. The classification strategy of Simpson's min-max classifier consists of covering the training data with hyperboxes constrained to have their boundary surfaces parallel to the coordinate axes of the chosen reference system. In order to obtain a more precise covering of each data cluster, in the present work hyperboxes are rotated by a suitable local principal component analysis, so that it is possible to arrange the hyperboxes orientation along any direction of the data space. The new training algorithm is based on the ARC/PARC technique, which overcomes some undesired properties of the original Simpson's algorithm. In particular, the training result does not depend on patterns presentation order and hyperbox expansion is not limited by a fixed maximum size, so that it is possible to have different covering resolutions. A toy problem and two real data benchmarks are considered for illustration.

An algorithm to train min-max neural models is proposed. It is based on the adaptive resolution classifier (ARC) technique, which overcomes some undesired properties of the original Simpson's (1992) algorithm. In particular, training results do not depend on pattern presentation order and hyperbox expansion is not limited by a fixed maximum size, so that it is possible to have different covering resolutions. ARC generates the optimal min-max network by a succession of hyperbox cuts. The generalization capability of the ARC technique depends mostly on the adopted cutting strategy. A new recursive cutting procedure allows ARC technique to yield a better performance. Some real data benchmarks are considered for illustration.

A systemic view of a productive system is presented, based on three ontologically different resources. The crucial features of this view are: (i) the weaknesses and strengths of a system reside in the interactions among the components; and (ii) the history of interactions explains the ontology of each resource. The view is illustrated through a brief description of the interactions occurring in the early development of a counting system.

We propose a constructive method, inspired by Simpson's min-max technique (1992), for obtaining fuzzy neural networks. It adopts a cost function depending on a unique net parameter. This feature allows us to apply a simple unimodal search for determining this parameter and hence the architecture of the optimal net. The algorithm shows a good behavior with respect to other methods when applied to real classification problems. Due to the adopted fuzzy membership functions, it is particularly indicated when the classes are extremely overlapped (for instance, in the case of biological data). Some results at this regard are reported in the paper.