Recent advancements in deep neural network (NN) models have made them promising candidates for positioning devices in environments lacking line-of-sight (LoS) links. We propose several novel components and design choices for complexvalued NNs (CVNNs) to process signals received from distributed multiple input multiple output (MIMO) wireless systems for precise positioning in non-LoS (NLoS) environments. We design effective preprocessing solution to handle the wide dynamic range of radio frequency (RF) data; novel MIMO RF data tokenization approach with 1D convolutional layers to enable joint processing of all antennas; and novel components for CVNN designs. We further propose three highly effective$\mathbf{R F}$data augmentation approaches to enhance model performance and reduce needs of large training datasets. Extensive comparison/ablation study is provided to support the proposals. With the proposed designs, we demonstrate the achievability of 2D positioning error less than 0.21 m more than 90 % of the time in a large NLoS environment.

Recent advancements in machine learning have improved device positioning in challenging environments where traditional methods often fall short. This work introduces a deep learning-based approach to predict the location of user equipment in a wireless radio network. The model leverages data samples derived from channel impulse responses as input features. Additionally, the proposed sample-based method is compared to conventional path-based data estimated through a channel estimator. Performance evaluations are conducted under different conditions to demonstrate the impact of input types, network densification, and available bandwidth in an indoor factory scenario.

Prolonged rotating outages and exorbitant energy bills, recently experienced in California and Texas, have exposed the limitations and need for modernizing electric power systems. The occurrence of such events is a consequence of peak loads, often due to extreme outside temperatures that simultaneously trigger Heating Ventilation Air Conditioning (HVAC) systems. Leveraging pervasive computing technologies, such as smart meters and smart thermostats, this paper introduces a comprehensive approach to perform residential HVAC power conservation and prevent these catastrophic events. Differently from previous solutions, our approach models realistic user behavior and HVAC dynamics of individual homes. Specifically, we formulate a novel reverse auction-based problem, called POwer Conservation Optimization (POCO). The goal is to perform power conservation by motivating users to temporarily adjust their HVAC thermostat settings in exchange for financial rewards. We prove that POCO ensures truthfulness and individual rationality of the auction mechanism, although it is an NP-hard problem. Therefore, we propose an efficient heuristic, called Greedy Ranking AllocatioN (GRAN), which we prove ensures the same formal properties, while incurring only a polynomial complexity. To predict power savings resulting from an HVAC thermostat adjustments, we propose a novel machine learning-based technique called Power Saving Prediction (PSP). In addition, we conduct an online survey to study the willingness to adopt the proposed system and to model realistic user behavior. Survey results show willingness of adoption above 79% and a highly heterogeneous and non-linear user behavior. We perform extensive experiments using high-fidelity simulator EnergyPlus. Results show that PSP outperforms a state-of-the-art solution obtaining 85% predictions within a 5% error margin. Furthermore, GRAN achieves near-optimal performance, outperforming a recent state-of-the-art approach obtaining results between 58% and 68% closer to the optimum.

With the acceleration of Information and Communication Technologies and the Internet-of-Things paradigm, smart residential environments , also known as smart homes , are becoming increasingly common. These environments have significant potential for the development of intelligent energy management systems and have therefore attracted significant attention from both academia and industry. An enabling building block for these systems is the ability of obtaining energy consumption at the appliance-level. This information is usually inferred from electric signals data (e.g., current) collected by a smart meter or a smart outlet, a problem known as appliance recognition . Several previous approaches for appliance recognition have proposed load disaggregation techniques for smart meter data. However, these approaches are often very inaccurate for low consumption and multi-state appliances. Recently, Machine Learning (ML) techniques have been proposed for appliance recognition. These approaches are mainly based on passive MLs, thus requiring pre-labeled data to be trained. This makes such approaches unable to rapidly adapt to the constantly changing availability and heterogeneity of appliances on the market. In a home setting scenario, it is natural to consider the involvement of users in the labeling process, as appliances’ electric signatures are collected. This type of learning falls into the category of Stream-based Active Learning (SAL). SAL has been mainly investigated assuming the presence of an expert , always available and willing to label the collected samples. Nevertheless, a home user may lack such availability, and in general present a more erratic and user-dependent behavior. In this article, we develop a SAL algorithm, called K -Active-Neighbors (KAN), for the problem of household appliance recognition. Differently from previous approaches, KAN jointly learns the user behavior and the appliance signatures. KAN dynamically adjusts the querying strategy to increase accuracy by considering the user availability as well as the quality of the collected signatures. Such quality is defined as a combination of informativeness , representativeness , and confidence score of the signature compared to the current knowledge. To test KAN versus state-of-the-art approaches, we use real appliance data collected by a low-cost Arduino-based smart outlet as well as the ECO smart home dataset. Furthermore, we use a real dataset to model user behavior. Results show that KAN is able to achieve high accuracy with minimal data, i.e., signatures of short length and collected at low frequency.

Smart Connected Communities (SCCs) is a novel paradigm that brings together multiple disciplines, including social-sciences, computer science, and engineering. Large-scale surveys are a fundamental tool to understand the needs and impact of new technologies to human populations, necessary to realize the SCC paradigm. However, there is a growing debate regarding the reproducibility of survey results. As an example, it has been shown that surveys may easily provide contradictory results, even if the subject populations are statistically equivalent from a demographic perspective. In this paper, we take the initial steps towards addressing the problem of reproducibility of survey results by providing formal methods to quantitatively justify apparently inconsistent results. Specifically, we define a new dissimilarity metric between two populations based on the users answers to non-demographic questions. To this purpose, we propose two algorithms based on submodular optimization and information theory, respectively, to select the most representative questions in a survey. Results show that our method effectively identifies and quantifies differences that are not evident from a purely demographic point of view.

Renewable, heterogeneous and distributed energy resources are the future of power systems, as envisioned by the recent paradigm of Virtual Power Plants (VPPs). Residential electricity generation, e.g., through photovoltaic panels, plays a fundamental role in this paradigm, where users are able to participate in an energy sharing system and exchange energy resources among each other. In this work, we study energy sharing systems and, differently from previous approaches, we consider realistic user behaviors by taking into account the user preferences and level of engagement in the energy trades. We formulate the problem of matching energy resources while contemplating the user behavior as a Mixed Integer Linear Programming (MILP) problem, and show that the problem is NPHard. Since the solution of such problem requires the knowledge of the user behavioral model, we propose an heuristic based on reinforcement learning with bounded regret to learn such model while optimizing the system performance. Comparison with the state-of-the-art approaches using realistic simulations based on real traces shows that our method outperforms existing schemes in several efficiency metrics. Besides, the results reveal that increasing the amount of produced energy improves the learning ability of the system even in a short period. It gives practical insights for implementation of energy sharing systems.

Matching power demand during peak load hours is a well-known problem in power systems. In fact, the cost of producing electricity increases very rapidly when the demand is high, due to the need for starting backup generators and enhancing transmission system. Incentive-based Demand Response (DR) program is a new approach, enabled by recent advances in smart grid technologies, designed to deal with such problem. According to DR, the utility company can provide economical incentives to users in order to temporarily reduce their energy consumption during peak hours. It is, however, challenging to determine the procedure to distribute such incentives, as well as to ensure that users will be sufficiently engaged and satisfied to make the DR program effective. In this paper, we propose a reverse auction mechanism to enable an incentive-based DR program. We formulate the DR reverse auction as an integer linear programming (ILP) problem, which integrates a perceived-value utility, to model the user perception of electrical appliances, as well as the financial objectives of the utility company. We adopt a Vickrey-Clarke-Groves (VCG) based reverse auction mechanism to guarantee the truthfulness and individual rationality properties. Since the VCG auction requires to optimally solve the NP-Hard ILP problem, we propose a heuristic algorithm named Reverse Auction DemAnd Response (RADAR), and prove that RADAR preserves truthfulness. Extensive simulations using real power consumption data of several homes show that RADAR is effective in reducing demand peaks while outperforming previous solutions in terms of users' perceived utility.

Smart homes offer new possibilities for energy management. One key enabler of these systems is the ability to monitor energy consumption at the appliance level. Existing approaches rely mainly on data from aggregated smart meter readings, but lack sufficient accuracy to recognize several appliances. Conversely, smart outlets are a suitable alternative since they can provide accurate electrical readings on individual appliances. Previous approaches for appliance recognition based on smart outlets use passive machine learning, which are deficient in the flexibility and scalability to work with highly heterogeneous appliances in smart homes. In this paper, we propose a stream-based active learning approach, called K -Active-Neighbors (KAN), to address the problem of appliance recognition in smart homes. KAN is an interactive framework in which the user is asked to label signatures of recently used appliances. Differently from previous work, we consider the realistic case in which the user is not always available to participate in the labeling process. Therefore, the system simultaneously learns the signatures and also the user willingness to interact with the system, in order to optimize the learning process. We develop an Arduino-based smart outlet to test our approach. Results show that, compared to previous solutions, KAN achieves higher accuracy in up to 41% less time.

Residential energy consumption has been rising rapidly during the last few decades. Several research efforts have been made to reduce residential energy consumption, including demand response and smart residential environments. However, recent research has shown that these approaches may actually cause an increase in the overall consumption, due to the complex psychological processes that occur when human users interact with these energy management systems. In this article, using an interdisciplinary approach, we introduce a perceived-value driven framework for energy management in smart residential environments that considers how users perceive values of different appliances and how the use of some appliances are contingent on the use of others. We define a perceived-value user utility used as an Integer Linear Programming (ILP) problem. We show that the problem is NP-Hard and provide a heuristic method called COndensed DependencY (CODY). We validate our results using synthetic and real datasets, large-scale online experiments, and a real-field experiment at the Missouri University of Science and Technology Solar Village. Simulation results show that our approach achieves near optimal performance and significantly outperforms previously proposed solutions. Results from our online and real-field experiments also show that users largely prefer our solution compared to a previous approach.

Activity recognition has been introduced as a means of detecting an action from a series of observations. Although in the literature, the terms "activity recognition" and "human activity recognition" are mostly used interchangeably, there exist several interesting applications for non-human subjects. In this work, we study animal activity recognition with special focus on horse gaits recognition. The on-body and unobtrusive system developed in this paper has several potential applications which can raise attention towards financial, emotional and veterinary health issues related to animals. Leveraging the pervasive use of smartwatches for activity tracking, we develop a smartwatch application to collect accelerometer data. The application is based on novel outlier detection and feature extraction techniques, in conjunction with state-of-the-art machine learning approaches. We perform real life experiments with two horses to evaluate the performances of our proposed system. To this aim, we place the monitoring device both on the horse saddle and the rider's wrist. The results show a high accuracy in both scenarios, which allows a seamless and unobtrusive use of our wearable device application by the rider. In addition, we study the effects of sliding window size and sampling frequency, providing useful insights for future research in horse gaits recognition.

In this paper, we study the Energy Efficiency (EE) and Area Spectral Efficiency (ASE) of green random cellular networks which utilize the Cell Zooming (CZ) techniques. To this aim, using stochastic geometry, we derive a tractable expression for the Ergodic Capacity in a Poisson Voronoi Tessellation (PVT) random cellular network in which both Base Station (BS) and Mobile User (MU) locations are drawn randomly from two independent Poisson Point Processes (PPPs). The performance of this network is examined under different MU densities and two CZ algorithms. Numerical evaluations show that there is an optimum transmission power, which maximizes EE in PVT random cellular networks. On the other hand, increasing the transmission power and the cell size does not improve ASE much more after passing a threshold. The tradeoff between EE and ASE is also presented.

In this paper, we investigate the achievable secrecy rates in an Energy Harvesting communication system composed of one transmitter and multiple receivers. In particular, because of the energy constraints and the channel conditions, it is important to understand when a device should transmit or not and how much power should be used. We introduce the Optimal Secrecy Policy in several scenarios. We show that, if the receivers demand high secrecy rates, then it is not always possible to satisfy all their requests. Thus, we introduce a scheme that chooses which receivers should be discarded. Also, we study how the system is influenced by the Channel State Information and, in particular, how the knowledge of the eavesdropper's channel changes the achievable rates.