We introduce a physics-informed neural network for power flow (PINN4PF) that effectively captures the nonlinear dynamics of large-scale modern power systems. The proposed neural network (NN) architecture consists of two important advancements in the training pipeline: (A) a double-head feed-forward NN that aligns with power flow (PF), including an activation function that adjusts to the net active and reactive power injections patterns, and (B) a physics-based loss function that partially incorporates power system topology information through a novel hidden function. The effectiveness of the proposed architecture is illustrated through 4-bus, 15-bus, 290-bus, and 2224-bus test systems and is evaluated against two baselines: a linear regression model (LR) and a black-box NN (MLP). The comparison is based on (i) generalization ability, (ii) robustness, (iii) impact of training dataset size on generalization ability, (iv) accuracy in approximating derived PF quantities (specifically line current, line active power, and line reactive power), and (v) scalability. Results demonstrate that PINN4PF outperforms both baselines across all test systems by up to two orders of magnitude, not only in terms of direct criteria, e.g., generalization ability, but also in terms of derived physical quantities.

Electric-vehicle smart charging requires quick decision-making under uncertainty while enforcing strict electricity grid and user requirements. Mathematical optimization becomes too slow at scale, while online reinforcement learning struggles with sparse rewards and safety. This paper proposes GNN-DT, a topology-aware Decision Transformer that combines graph neural network embeddings with sequence modeling to learn charging policies from offline trajectories. The method operates over variable numbers of vehicles and chargers without retraining. Evaluated on realistic smart charging scenarios, GNN-DT achieves near-optimal performance, reaching rewards within 5 percent of an oracle solver while using up to 10× fewer training trajectories than baseline methods. It consistently outperforms online and offline reinforcement learning approaches and generalizes to unseen fleet sizes and network topologies. Inference runs in milliseconds, making the approach suitable for real-time deployment in large-scale charging systems.

The increasing penetration of Electric Vehicles (EVs) and renewable energy sources is placing significant stress on existing power grid infrastructure. This work investigates the application of vehicle-to-grid (V2G)-enabled smart charging in workplace environments from the perspective of EV aggregators, using real-world charging data from Dutch business parking lots. To address the limitations of conventional deep Reinforcement Learning (RL) methods in enforcing operational constraints, we propose a Safe RL method using the Constrained Variational Policy Optimization (CVPO) algorithm, specifically designed to reduce constraint violations and enhance reliability. Empirical results show that CVPO outperforms classic RL baselines and rule-based policies, closely approximating the performance of an optimal offline benchmark while exhibiting strong generalization to unseen scenarios.

Integrating electric vehicles (EVs) into the power grid can revolutionize energy management strategies, offering both challenges and opportunities for creating a more sustainable and resilient grid. In this context, model predictive control (MPC) emerges as a powerful tool for addressing the complexities of Grid-to-Vehicle (G2V) and Vehicle-to-Grid (V2G) enabled demand response management. By leveraging advanced optimization techniques, MPC algorithms can anticipate future grid conditions and dynamically adjust EV charging and discharging schedules. However, no standard tools exist to evaluate novel energy management strategies based on MPC approaches. This work focuses on harnessing the potential of MPC in G2V and V2G applications by providing open-source algorithms that allow the maximization of EV flexibility and support demand response initiatives while mitigating the impact on EV battery health. Through extensive simulation and analysis, we demonstrate the efficacy of our approach in maximizing the benefits of G2V and V2G while assessing the impact on the longevity and reliability of EV batteries. Specifically, the proposed methods enable the optimization of EV charging and discharging schedules in real-time, taking into account fluctuating energy prices, grid constraints, and EV user preferences.

Rapidly expanding Electric Vehicle (EV) adoption necessitates robust, large-scale charging strategies to meet global decarbonization targets. Traditional methods, such as heuristics and mathematical programming-based approaches, struggle to scale effectively and adapt to EV dispatch’s complexity, uncertainty, and variability. Reinforcement Learning (RL) offers a promising alternative due to its ability to handle complex optimization problems, process substantial real-time data, and learn continuously without explicit retraining for every scenario. This study proposes a novel end-to-end RL framework that leverages the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm, enhanced by Graph Neural Networks (GNN), to capture spatial interactions among charging nodes. A key contribution is integrating a safety layer designed to ensure grid stability, preserve EV charging requirements, and enforce power limits. The RL agent was trained and evaluated using real EV charging sessions, offering a realistic assessment of its performance. The results indicate that the proposed method can efficiently coordinate large fleets of EVs, ensuring stable power grid operation and fair distribution of charging resources.

As the adoption of electric vehicles (EVs) accelerates, addressing the challenges of large-scale, city-wide optimization becomes critical in ensuring efficient use of charging infrastructure and maintaining electrical grid stability. This study introduces EV-GNN, a novel graph-based solution that addresses scalability challenges and captures uncertainties in EV behavior from a Charging Point Operator’s (CPO) perspective. We prove that EV-GNN enhances classic Reinforcement Learning (RL) algorithms’ scalability and sample efficiency by combining an end-to-end Graph Neural Network (GNN) architecture with RL and employing a branch pruning technique. We further demonstrate that the proposed architecture’s flexibility allows it to be combined with most state-of-the-art deep RL algorithms to solve a wide range of problems, including those with continuous, multi-discrete, and discrete action spaces. Extensive experimental evaluations show that EV-GNN significantly outperforms state-of-the-art RL algorithms in scalability and generalization across diverse EV charging scenarios, delivering notable improvements in both small- and large-scale problems.

The transition to Electric Vehicles (EVs) introduces challenges for power grid integration, particularly due to the growing demand for charging infrastructure. To support research on smart charging strategies and bidirectional charging applications, this study presents an open-access dataset containing 142 EV charging profiles obtained in a laboratory environment. The dataset includes static charging and discharging scenarios alongside dynamic profiles where the charging power is varied over time. These scenarios are applied to eight commercially available EVs, three of which support bidirectional charging. It features tests in alternating current and direct current charging modes and includes high-resolution time series of grid and vehicle parameters at sub-second intervals. The dataset is technically validated by assessing charging efficiency, reactive power injection, harmonics, and its suitability for development of digital EV models. This dataset supports applications like model validation, grid integration simulations in the context of Vehicle-to-Grid (V2G), charging infrastructure planning, and smart charging strategy development.

Interest in integrating distributed energy resources (DERs) into the electric distribution system (EDS) is growing due to the economic and operational benefits that DERs can provide. Consumer-sited photovoltaic (PV) generation is one of those DERs that have been penetrating the EDS over the years; the owner can sell the energy surplus to the EDS, and for the EDS operator it brings the advantage of clean energy from solar irradiation; since the EDS operator doesn't have control of it, the growing of the penetration can leading into technical problems to the EDS. Energy storage systems (ESS) can be used to manage non-dispatchable renewable energy providing ancillary services to the EDS like peak shaving and voltage regulation. However, investors for the ESS planning (allocation and sizing) must consider their cost, EDS constraints, and the future scenarios of PV penetration and uncertainties inherent to the EDS; for this purpose, a stochastic mixed integer linear programming model for the ESS planning is presented integrating a temporal method to model the PV penetration growth over the years. The decision variables are the size and the EDS node where the ESS should be connected. The model is implemented in AMPL and the optimal solution is found with CPLEX for a case study based on the IEEE 33 node test system.

Accurate and efficient power flow (PF) analysis is crucial in modern electrical networks’ operation and planning. Therefore, there is a need for scalable algorithms that can provide accurate and fast solutions for both small and large scale power networks. As the power network can be interpreted as a graph, Graph Neural Networks (GNNs) have emerged as a promising approach for improving the accuracy and speed of PF approximations by exploiting information sharing via the underlying graph structure. In this study, we introduce PowerFlowNet, a novel GNN architecture for PF approximation that showcases similar performance with the traditional Newton–Raphson method but achieves it 4 times faster in the IEEE 14-bus system and 48 times faster in the realistic case of the French high voltage network (6470rte). Meanwhile, it significantly outperforms other traditional approximation methods, such as the DC power flow, in terms of performance and execution time; therefore, making PowerFlowNet a highly promising solution for real-world PF analysis. Furthermore, we verify the efficacy of our approach by conducting an in-depth experimental evaluation, thoroughly examining the performance, scalability, interpretability, and architectural dependability of PowerFlowNet. The evaluation provides insights into the behavior and potential applications of GNNs in power system analysis.

Decarbonizing the transportation sector involves adopting electric vehicles (EVs); a shift that introduces significant challenges in energy distribution management and raises concerns about grid stability. Charge Point Operators (CPOs) are important in this transition as they control the EV charging process by balancing the needs of EV users and the grid. This study presents a smart-charging model from the perspective of CPOs for handling EVs located in a commercial parking lot to minimize the Power Setpoint Tracking (PST) error. To solve this sequential decision-making problem, a Markov Decision Process (MDP) model is designed and solved using Deep Deterministic Policy Gradient (DDPG), a Deep Reinforcement Learning (DRL) algorithm. The proposed model can effectively manage the uncertainties associated with EV arrivals and fluctuating charging demands by structuring the action and state space to incorporate power constraints. The experimental evaluation using realistic EV behavior data shows that the proposed approach significantly outperforms uncontrolled charging, reducing PST error while effectively managing multiple EV chargers and EVs with varying battery capacities and power limitations.

As electric vehicle (EV) numbers rise, concerns about the capacity of current charging and power grid infrastructure grow, necessitating the development of smart charging solutions. While many smart charging simulators have been developed in recent years, only a few support the development of Reinforcement Learning (RL) algorithms in the form of a Gym environment, and those that do usually lack depth in modeling Vehicle-to-Grid (V2G) scenarios. To address the aforementioned issues, this paper introduces EV2Gym, a realistic simulator platform for the development and assessment of small and large-scale smart charging algorithms within a standardized platform. The proposed simulator is populated with comprehensive EV, charging station, power transformer, and EV behavior models validated using real data. EV2Gym has a highly customizable interface empowering users to choose from pre-designed case studies or craft their own customized scenarios to suit their specific requirements. Moreover, it incorporates a diverse array of RL, mathematical programming, and heuristic algorithms to speed up the development and benchmarking of new solutions. By offering a unified and standardized platform, EV2Gym aims to provide researchers and practitioners with a robust environment for advancing and assessing smart charging algorithms.

Multiclass image classification is a complex task that has been thoroughly investigated in the past. Decomposition-based strategies are commonly employed to address it. Typically, these methods divide the original problem into smaller, potentially simpler problems, allowing the application of numerous well-established learning algorithms that may not apply directly to the original task. This work focuses on the efficiency of decomposition-based methods and proposes several improvements to the meta-learning level. In this paper, four methods for optimizing the ensemble phase of multiclass classification are introduced. The first demonstrates that employing a mixture of experts scheme can drastically reduce the number of operations in the training phase by eliminating redundant learning processes in decomposition-based techniques for multiclass problems. The second technique for combining learner-based outcomes relies on Bayes’ theorem. Combining the Bayes rule with arbitrary decompositions reduces training complexity relative to the number of classifiers even further. Two additional methods are also proposed for increasing the final classification accuracy by decomposing the initial task into smaller ones and ensembling the output of the base learners along with that of a multiclass classifier. Finally, the proposed novel meta-learning techniques are evaluated on four distinct datasets of varying classification difficulty. In every case, the proposed methods present a substantial accuracy improvement over existing traditional image classification techniques.

In this paper, we put forward a novel DER aggregation framework, encompassing a multiagent architecture and various types of mechanisms for the effective management and efficient integration of DERs in the Grid. One critical component of our architecture is the Local Flexibility Estimators (LFEs) agents, which are key for offloading the Aggregator from serious or resource-intensive responsibilities-such as addressing privacy concerns and predicting the accuracy of DER statements regarding their offered demand response services. The proposed aggregation framework allows the formation of efficient LFE cooperatives. Our experiments verify its effectiveness for incorporating heterogeneous DERs into the Grid in an efficient manner-showing that the use of appropriate mechanisms results in higher payments for participating LFEs.

Factorizing a multiagent system refers to partitioning the state-action space to individual agents and defining the interactions between those agents. This so-called agent factorization is of much importance in real-world industrial settings, and is a process that can have significant performance implications. In this work, we explore if the performance impact of agent factorization is different when using different learning algorithms in multiagent coordination settings. We evaluated six different agent factorization instances - or agent definitions - in the warehouse traffic management domain, comparing the performance of (mainly) two learning algorithms suitable for learning coordinated multiagent policies: the Evolutionary Strategies (ES), and a genetic algorithm (CCEA) previously used in this setting. Our results demonstrate that different learning algorithms are affected in different ways by alternative agent definitions. Given this, we can deduce that many important multiagent coordination problems can potentially be solved by an appropriate agent factorization in conjunction with an appropriate choice of a learning algorithm. Moreover, our work shows that ES is an effective learning algorithm for the warehouse traffic management domain; while, interestingly, celebrated policy gradient methods do not fare well in this complex real-world problem setting.

Urban waste management is a most challenging issue for modern societies. Reducing pollution and saving environmental resources provides significant opportunities for local, national and international economic growth. In Greece, the recycling rates are currently low compared to other European countries. The current study proposes an autonomous, intelligent robotic system for categorizing and separating recyclable materials aiming to contribute in increasing the recycling rates in Greece. The system is a series connection of an optical sub-system and a robotic sub-system. The optical subsystem receives input from a ordinary RGB and an NIR camera. These are processed in combination for the identification and categorization of recyclables into predefined material types. The output of the optical sub-system provides a list of potential targets (recyclables) to be picked and sorted. This is forwarded to the robotic subsystem, which undertakes the physical separation of the materials to the appropriate bin. The proposed system, named ANASA system, has been deployed in two different urban waste management industrial units, in DEDISA, Chania, Crete, Greece (processing recyclable wastes) and in ESDAK, Heraklion, Crete, Greece (processing composite wastes), where the system's reliability and validity is experimentally tested in real industrial environments. The advantages over the existing ordinary recycling systems are significant: high reliability in object recognition (material detection), short separation cycle (high speed), significantly low installation volume, low cost and ease of application to both old and new recycling industries. The combination of the above features provides a potential for exploitation as a complete commercial commercialization.

The PowerTAC competition provides a multi-agent simulation platform for electricity markets, in which intelligent agents acting as electricity brokers compete with each other aiming to maximize their profits. Typically, the gains of agents increase as the number of their customers rises, but in parallel, costs also increase as a result of higher transmission fees that need to be paid by the electricity broker. Thus, agents that aim to take over a disproportionately high share of the market, often end up with losses due to being obliged to pay huge transmission capacity fees. In this paper, we present a novel trading strategy that, based on this observation, aims to balance gains against costs; and was utilized by the champion of the PowerTAC-2020 tournament, TUC-TAC. The approach also incorporates a wholesale market strategy that employs Monte Carlo Tree Search to determine TUC-TAC’s best course of action when participating in the market’s double auctions. The strategy is improved by making effective use of a forecasting module that seeks to predict upcoming peaks in demand, since in such intervals incurred costs significantly increase. A post-tournament analysis is also included in this paper, to help draw important lessons regarding the strengths and weaknesses of the various strategies used in the PowerTAC-2020 competition.