This paper presents a real-world case study on boosting residential photovoltaic (PV) self-consumption through smart scheduling of an electric vehicle (EV) charger and a water heater, without stationary battery storage. Conducted from January 1 to May 1, 2025, the experiment involved a 3.2 kWp PV system in Alsace (France), a plug-in hybrid electric vehicle (PHEV) type of Volkswagen Tiguan eHybrid (with a 19.6 kWh battery and 1.7 kW AC charging). Power flows (PV output, household demand, and PV contribution) were logged every 10 seconds. Control automations, implemented in Home Assistant (HA) on a Raspberry Pi, leveraged real-time grid export data and weather forecasts to defer EV charging and water heating to align with peak solar production. We compare this smart control approach to an uncontrolled baseline. Results—highlighted with winter and spring daily profiles—show PV self-consumption exceeding 80% when charging is shifted to midday. We conduct a cost-benefit analysis based on local electricity tariffs, revealing seasonal savings. Flexibility strategies, user behavior, and forecast accuracy are also discussed. Our findings, supported by workflow diagrams of the control logic, demonstrate that smart coordination of flexible loads can substantially enhance PV utilization—supporting IEA-PVPS guidelines even in the absence of battery storage.

Accurate State of Health (SOH) estimation is critical for optimizing battery management, extending lifespan, and preventing unexpected failures in lithium-ion batteries. However, existing methods often face challenges related to computational complexity and data requirements, limiting their practicality in real-time, resource-constrained applications. This paper presents a novel data-driven approach for SOH estimation that combines low computational demands with high accuracy. Using Gaussian Process Regression (GPR) and features extracted from both charge and discharge voltage curves, we achieve precise SOH estimation with a Mean Absolute Error (MAE) of 0.06%. The proposed approach is validated using the Oxford Battery dataset, with results showing consistent accuracy across multiple cells and under varying training data sizes. Notably, the model achieves robust performance even with limited training data, completing estimations in just 0.05 seconds, making it highly suitable for real-time applications. This work contributes to the advancement of battery management systems by providing a computationally efficient, accurate, and real-time SOH estimation method.

Non-Intrusive Load Monitoring (NILM) offers a promising approach for identifying device-level energy usage from aggregate household measurements. Most existing methods rely on supervised learning or require labeled training data, which limits generalizability in practical scenarios. This paper introduces the Unsupervised Multi-Sequence Appliance Identification (UMSAI) algorithm, a modular pipeline designed to identify complex multi-phase appliances, such as washing machines and dishwashers, using only two weeks of unlabeled measurement data from a single measurement device. UMSAI performs robust event detection, feature-based clustering, and appliance identification through graph-based similarity analysis. Evaluation on real-world datasets demonstrates high detection accuracy for appliances with distinct cyclic patterns, highlighting the method’s applicability in data-scarce, label-free environments.

Battery Energy Storage Systems (BESS) play a crucial role in enhancing Distributed Energy Resources (DERs) efficiency and reliability. Managing these systems across diverse DER environments presents challenges due to the dynamic nature of the grid, market fluctuations, and the inherent complexities of both DERs and the batteries themselves. This paper proposes a new approach for adaptive battery management in DERs, utilizing meta learning for Deep Q Networks. We trained an autonomous agent in a reinforcement learning environment, enabling it to optimize battery operations across multiple DER locations with minimal training data. The effectiveness of the proposed method is validated through a two-stage process. First, the agent undergoes meta-learning training in the reinforcement environment, equipping it with the necessary decision-making capabilities. Second, its performance is evaluated through a simulation using real world data on energy consumption, generation, and pricing. The agent excels at handling multiple objectives simultaneously and pursues three key goals: maximizing renewable energy usage, maintaining healthy battery states of charge, and potentially reducing energy costs for consumers.

Data processed in context is more meaningful, easier to understand and has higher information content, hence it derives its semantic meaning from the surrounding context. Even in the field of acoustic signal processing. In this work, a Deep Learning based approach using Ensemble Neural Networks to integrate context into a learning system is presented. For this purpose, different use cases are considered and the method is demonstrated using acoustic signal processing of machine sound data for valves, pumps and slide rails. Mel-spectrograms are used to train convolutional neural networks in order to analyse acoustic data using image processing techniques.

Abstract The aim of this article is to design a new telehealth system with secured wireless transmission and classification of multiple biosignals using e‐Health sensors platform and Xbee modules with Arduino Uno and Raspberry Pi as acquisition and processing units, respectively. The collected data, such as temperature, airflow, position, Galvanic skin response and oxygen in the blood can be evaluated in order to monitor patient health state using threshold detection. The prediction of the cardiac state based on automatic identification of arrhythmias is validated by the classification of ElectroCardioGram (ECG) signals using Artificial Intelligence (AI) by exploiting TensorFlow and Keras tools. Different AI algorithms and a combination with different Machine Learning (ML) basing to transfer learning approach are tested. These algorithms include Artificial Neural Network (ANN), Convolutional Neural Network (CNN), Support Vector Machine (SVM), K‐Nearest Neighbour (KNN) and Random Forest (RF). At first, ANN and CNN are used to classify ECG‐scalogram images using softmax, then the used CNN model (VGG16) is employed to extract features and pass them to other traditional classifiers (SVM, KNN and RF) allowing to evaluate and select the best classifier, such that the ECG signal can be classified into four categories namely Normal Sinus Rhythm (NSR), Atrial Fibrillation (AF), Congestive Heart Failure (CHF) and other cardiac arrhythmia (ARR). The proposed method has been evaluated using real recorded signals and four PhysioNet databases. A Graphical User Interface (GUI) has been designed with C# under Visual Studio IDE allowing to display the results using personal computer (PC) or a network linked phone, which makes it possible to transfer the diagnosis with the prediction results to a remote clinic control room as Internet of Things (IoT) system application. The best classification accuracy of 99.56% is attained, confirming that the designed system allows a good trade‐off between low cost and performances in addition, it is easy to use with quick access to multiple biosignals. It has improved vital characteristics monitoring and diagnosis services quality under a robust secured wireless transmission using lightweight chaos‐based algorithm, thus preventing loss of life during critical health situations.

Currently, the need for more efficient use of energy is in the spotlight more than ever. For optimal energy management the forecast of energy consumption is of great interest.This paper takes a novel approach for forecasting the energy demand in households by using a hybrid AI approach. On the one hand, we use an interpretable model creation by using fuzzy rules. Those rules are then combined with an evolutionary strategy to create new simulation data which calibrates the reality and sometimes uncertainty behind the data. Based on this newly created data, a simple artificial neural network (ANN) model is created. It is shown, that there is no need to create an unnecessarily complex deep learning architecture for achieving good results. Simple ANN models can achieve excellent results, when using data created by inferred fuzzy rules. Of great advantage is, that one part of this hybrid AI approach can still be interpreted by humans and furthermore improved by adding the knowledge of human domain experts in form of fuzzy rules.

This paper presents an improved incremental conductance (Imp-IC) based maximum power point tracking (MPPT) algorithm for photovoltaic (PV) systems. This algorithm is based on the slope condition of the current-versus-voltage characteristic of the PV panel. Its principle consists in comparing the PV panel instantaneous conductance with the incremental conductance of the DC-DC boost converter to determine whether to turn on or turn off the converter power switch in order to achieve the maximum power point. The resulted algorithm is simple, easy to implement and does not require any parameter or step size for its adjustment. Tracking performances of the proposal are experimentally evaluated using the prEN 50530 standard dynamic tests. A comparison with perturb and observe algorithm is performed. Superiority of the proposed algorithm in terms of extracted power, tracking features, convergence speed and oscillatory behaviors cancellation is clearly proven.

The Li-ion batteries are commonly used for Electric Vehicles (EVs) and aerospace applications. One of the essential parameters in Li-ion batteries is state of charge (SOC) that shows the available energy in a battery. Various methods were proposed for SOC estimation. Since the battery has a nonlinear equations, it is important to use a method that does not require the system model. In the present study, a new Adaptive Online Gated Recurrent Unit (GRU) method is proposed for the State of Charge (SOC) estimation. It is a kind of deep Recurrent Neural Network(RNN) which solved the vanishing gradient problem in RNNs with GRU units. For Optimization a robust adaptive Online gradient learning method is used. This method is able to tune online the learning rate in the process. Adaptive GRU is a nondependent method from the nonlinear batteries model and simplifies the mathematical computation. The proposed technique is implemented on the real dataset of LifePO4 Li-ion batteries for finding SOC estimation. The exprimental result indicate that the Adaptive GRU method is more accurate than simple RNN.

Datasets are essential for the development of new algorithms in the research area of Non-Intrusive Load Monitoring. Therefore, a method will be investigated to generate simulation datasets based on real measurements. For this purpose, a method is developed using the FIT-PS transformation to extend switching cycles of devices to any length. It is investigated how well the recorded signals can be reconstructed from individually composed switching cycles. Then, signal expansion and superposition are used to generate simulation datasets with different degrees of complexity, which are available under the name HELD2 datasets. The first benchmarks for event detection and classification conclude the paper.

This paper proposes a method to combine high frequency sampled measurements of devices to a total current signal. Therefore, single and combination measurements can be used. The resulting total current signal is frequency, phase, and amplitude correct due to the use of Frequency Invariant Transformation of Periodic Signals. Measurements can be added from different data sets, which increases the number of devices significantly. For non-complex or dynamic loads, it is possible to extend the running time in the simulations arbitrarily. In addition to a large number of devices and switch-on and switch-off cycles, the advantage is the accuracy of the reference data as the individual consumption of each device is known precisely at any time. With the method proposed here, any realistic scenarios can be simulated using individual measurements. Details of the individual waveforms within a period are retained with this method.

Full bridge inverter are widely used as DC-AC power conversion interfaces in many areas such as PV application or interruptible power supply. The first and more important factor to evaluate the inverter is the quality of its output voltage. Indeed, this voltage should be sinusoidal with low total harmonic distortion (THD), even when feeding non-linear loads. In addition, the inverter is expected to have good transient response under load variation. This paper presents an enhanced control strategy to meet these requirements. The single-phase full bridge converter is modeled in stationary frame where the quantities are AC to make the system very simple without using any transformation. The adopted control is performed by using only one proportional regulator and only one measuring device to sense the output voltage. The injected voltage is obtained by the sum of the output voltage and a compensation term which is proportional to the error between the reference voltage and the feedback signal represented by the output voltage. The proposed control scheme is simulated and tested under linear and nonlinear load. It's found that the THD, steady-state error, response time of the output voltage still zero for any cases of connected loads.

This paper describes the optimization of parameters of an event detection method for Non-Intrusive Load Monitoring (NILM). The input signal consisting of voltage and current was processed with FIT-PS. An event detection method is presented with regard to the adjustable parameters. For parameter optimization the methods simulated annealing and pattern search are used. By using automatic parameter optimization methods, previous results based on manually selected parameters can be significantly improved up to 11.5 %. In the runtime investigation, pattern search has clear advantages over simulated annealing for comparable or better results. In addition, it is possible in the future to adapt this method very quickly to other boundary conditions.

This paper presents a smart battery charger based on the cascaded boost-buck converter supplied by a photovoltaic (PV) panel. This topology offers a great advantage since it operates in boost mode when the battery voltage is higher than the PV output voltage and operates in buck mode when the battery voltage is lower than the PV voltage output. Due to the presence of inductances at input and output of the cascaded boost-buck converter, the currents across the PV panel and the battery presented very low ripples. In this work, the control of the two converters is simplified since they are driven simultaneously within a same duty cycle. This reduces the implementation cost and complexity. The proposed battery charger operates intelligently; depending on the sun irradiance level and the battery state of charge, the algorithm automatically switches to maximum power point tracking mode, constant current mode, constant voltage mode, or floating charging mode. To verify the overall operation of the system, numerical simulations are performed for two battery voltage levels and including the different phases of charges. According to the obtained good results, the suggested smart charger can be inserted as battery management device in PV energy system.

New sensorless virtual-flux (VF) based predictive direct power control (PDPC) (VF-PDPC) of a three-phase pulse-width modulation (PWM) rectifier is developed in this paper. The VF estimation is achieved through a simple neural filter based integrator (NF-I) in series with a multi-output adaptive linear neuron (MO-ADALINE). The NF-I leads to cancel dc offset and harmonic distortions in the estimated VF. Thereafter, positive sequence components of the VF are extracted by mean of a MO-ADALINE. The estimated VF is then used for robust sensorless VF-PDPC with a constant switching frequency and tested through numerical simulations. Simulation results illustrated superiority of the proposed sensorless VF-PDPC compared with the conventional PDPC under ideal and non-ideal grid conditions. The proposed sensorless control strategy achieved good stability and the grid currents are quasi-sinusoidal, even if the grid voltages are highly unbalanced and distorted.

Smart meters based on Non-intrusive load monitoring (NILM) will be an essential part of our homes and buildings in the near future. In this paper, a novel algorithm for signal analysis in the context of NILM is proposed. The total electrical consumption is decomposed adaptively into narrow banded modes, which can be analyzed to extract useful information about the individual consumption of each device, including its ON-OFF events. The algorithm delivered very good results when used for the purpose of event detection on a publically available dataset (BLUED) [1].

In this paper, an adaptive voltage-oriented control (VOC) for three-phase grid-connected pulse-width modulation (PWM) rectifier is developed. As the control loop behavior is affected by the PWM rectifier operating conditions, the effects associated to the PWM rectifier losses and uncertainty in the inductance should be inserted in the model. Thus, equivalent resistance and inductance are estimated online by means of an adaptive linear neuron (ADALINE). The proposed ADALINE estimator is inserted in the current control loop to perform an adaptive VOC scheme. Hence, an online updating of the decoupling terms and the current controller gains is achieved. The ADALINE capability to estimate accurately the equivalent inductance and resistance as well as the adaptive VOC scheme performances are finally evaluated by experiments. Experimental results show improved performances and accurate estimation.

This paper presents the implementation of the time-frequency analysis tools, based on the S-transform and the modified S-transform, into a Raspberry Pi for a real-time computation. The aim of this approach is to provide an embedded system acquisition and measurement at low costs that would permit higher sampling resolution, recording and extraction of the data measurements in order to analyze electrical harmonics and transients. We discuss the modified S-transform proposed in literature and the different optimized Gaussian windows forms in order to obtain better energy concentration measurement. The implemented algorithm is described and applied to synthetic and real electrical signals. The modified S-transform used brings to light the enhancement of the time-frequency resolution. We conclude with future enhancements and perspectives of this study.

It is the aim of non-intrusive load monitoring (NILM) to determine the individual energy consumption of multiple devices on the basis of the total energy consumption. The energy consumption of a single device is measured at a central point without the need of individual measuring instruments on the devices themselves. We present a new method for signal separation, decomposition of the signal into individual states and feature extraction. Frequency invariant transformation of periodic signals (FIT-PS) is based on a signal diagram similar to the concept of trajectories [1]-[3] utilizing the periodicity of voltage and current, and their correlation. A major advantage of this approach is the dimensional expansion of the signal. Thereby smallest changes cif the current can be identified easily and the associated change of state can be detected. This approach breaks down the current signal into its individual periods using the voltage as a reference signal for the determination of trigger points. Thereby, the phase information between current and voltage is maintained and is inherently part of the new signal representation. The efficiency of the new signal representation and signal separation is demonstrated by the example of an event detection algorithm. For testing the BLUED dataset [4] was used reaching a sensitivity in event detection of 99.31%.

This paper presents a novel method for power quality analysis based on S-Transform. The objective is the detection of certain disturbances that pollute the electrical network such as interruptions, sags and swells. S-Transform provides frequency-dependent resolution while maintaining a direct relationship with the Fourier spectrum. The proposed approach consists in the application of Shannon Energy computed on each local spectrum obtained with the S-Transform. Using simulated and real data, the proposed method has been evaluated according to the accuracy of detecting the start and the end of each considered disturbance.

A new method for adaptive ac filter parameters identification of three phase pulse-width modulated rectifiers is presented in this paper. The proposed method is based on three adaptive linear neurons to identify online the ac filter resistances and inductances. The main advantage of this method is its simplicity and it requires low computational cost. The identification algorithm is implemented and three experimental tests have been realized in order to identify in real time the ac filter resistances and inductances. First test is performed in steady state operation. Two other tests are conducted to testing the track ability of the proposed method relative to parameters variations. Experimental results demonstrate that the proposed method provides accurate ac filter parameter values and tracks well the parameters variations. The obtained parameter values can be exploited for diagnosis purposes of ac filter status or in a control strategies.

Our goal is to successively build and publish a catalog of appliance characteristics. This is a fundamental basis for Non-Intrusive Load Monitoring (NILM) because it helps to structure the multitude of different appliances by grouping them in classes. In this paper we investigate time series signatures of electric household appliances, so called "fingerprints". We describe a methodology to find similarities in the loads' signatures that currently relies on visual inspection. First results are discussed on the basis of three types of equipment.

In this paper the Non-Intrusive Appliance Load Monitoring (NALM) will be described as a technique to improve the current smart meter systems. The development and verification of a measurement system is described. With that system it is possible to simulate and measure the energy consumption of appliances in residential buildings. These measurements are used to develop and verify the disaggregation algorithms for the event detection and the classification. A first event detection algorithm, based on the calculation of the difference values of the power signals is presented. A comparison with another algorithm is shown. The usefulness of that algorithm for the event detection is discussed at the end of the paper.

In this paper, a new adaptive neural phase-locked loop (AN-PLL) based on adaptive linear neuron networks for grid-connected converters synchronization is presented. The proposed AN-PLL architecture contains three stages. Firstly, the frequency of polluted and distorted grid voltage is tracked online. Then, the grid-voltages are filtered as well as the voltage vector amplitude is detected. Finally, the phase angle is estimated by means of PLL. The whole AN-PLL architecture is implemented under dSPACE DS1104 and applied to a real three phase power supply. The performances and the robustness of the proposed AN-PLL under voltage sag and two-phase fault are compared to the conventional PLL. The obtained experimental results demonstrate that the proposed architecture presents overwhelming advantages in terms of tracking accuracy and immunity to grid-voltage disturbances.

The goal of that paper is to show a possibility for the disaggregation of electrical appliances in the power profile of residential buildings. The advantage is that the measurement system is at a central point in the household. So the installation effort decrease. For the disaggregation of the appliances out of the load curve, an approach for the development of a system based on pattern recognition is presented. One method for the classification of appliances is to use Artificial Neural Network. This idea is the main part of that paper. It is shown a method, to classify one kind of appliances. At the end, the first results and a comparison with the famoust approach, for the disaggregation of electrical appliances, from Hart is presented.