Incentivizing flexible consumption of end-users is key to maximizing the value of local exchanges within Renewable Energy Communities. If centralized coordination for flexible resources planning raises concerns regarding data privacy and fair benefits distribution, state-of-the-art approaches (e.g., bi-level, ADMM) often face computational complexity and convexity challenges, limiting the precision of embedded flexible models. This work proposes an iterative resolution procedure to solve the decentralized flexibility planning with a central operator as a coordinator within a community. The community operator asks for upward or downward flexibility depending on the global needs, while members can individually react with an offer for flexible capacity. This approach ensures individual optimality while converging towards a global optimum, as validated on a 20-member domestic case study for which the gap in terms of collective bill is not more than 3.5% between the decentralized and centralized coordination schemes.

This paper provides an in-depth analysis on how different aspects of the dynamic operating envelope (DOE) formulation impact the computation and allocation of network capacity. We show that the envelopes are significantly affected by the power flow model (non-linear or linear), binding network constraint (thermal or voltage) and by the calculation case (import or export envelope). We also propose a novel DOE algorithm (LACE) that presents transparent and scalable computation that is useful for larger networks or to act in tandem with other optimization engines. We run numerical simulations with different test feeders, including a realistic low-voltage feeder with real-world data from Belgium. This paper provides crucial insights and tools to distribution system operators (DSOs), stakeholders and academics alike to make sure DOE calculation achieves desirable and efficient outcome.

Indirect Demand Response (IDR) programs that incentivize consumer participation through unidirectional price-based mechanisms offer a promising way to mobilize small-scale flexibility. The main challenge in such IDR programs lies in modeling the uncertain price-response relationship of demand response resources (DRRs), which complicates DR pricing. Inverse Optimization (IO) provides an effective method for capturing historical price-response patterns with interpretability and seamless integration into retailers' decision-making frameworks. While traditional IO prioritizes forecast accuracy, recent research advocates for value-oriented forecasting, which prioritizes decision quality to mitigate the practical impacts of forecast errors. Despite their advantages, fully integrated value-oriented approaches can become computationally intensive for IO. To address this, we propose a novel value-oriented IO (VOIO) framework that facilitates a data-driven, value-oriented identification of DRRs' aggregate price-response parameters by shifting IO hyperparameter evaluation metrics to decision-making regret minimization. The proposed method was applied in a case study examining a Balance Responsible Party & sacute; (BRP) participation in the Belgian single-price imbalance market. The results demonstrated that the VOIO approach was able to reduce overall financial losses from forecast errors compared to a forecast-oriented IO benchmark, achieving overall profit gains in the validation and test sets through more conservative pricing.

Increasing electrification and distributed energy integration exacerbate congestion and voltage deviations in distribution networks. Renewable energy communities (RECs) have emerged as a promising bottom-up solution, enabling local coordination and grid support through participation in local energy and flexibility markets. However, market inefficiencies, such as strategic manipulation of the baseline, and the heterogeneity of user preferences introduce substantial complexity to their design and operation. Motivated by these challenges, this paper proposes a novel three-stage framework that jointly considers user behavior and system-level needs. In the first stage, RECs engage in local energy markets (LEMs), where electricity prices are competitive and consider collective users' diverse preferences to ensure efficient energy trading. In the second stage, RECs offer baseline-constrained flexibility services to the distribution system operator (DSO) through the local flexibility market (LFM), which allocates rewards based on actual grid support delivered. In the third stage, internal revenue sharing among REC members is determined through a fairness-guaranteeing method that accounts for marginal flexibility contributions. Simulation results show that, beyond users' preferences, all RECs benefit from participation in the framework; however, these benefits vary according to the preferences and seasonal conditions. These factors also influence REC performance as well as DSO operation and associated costs. The LFM results for a representative winter day indicate the financially-driven member achieves the highest total flexibility revenue (€140.88), while the comfort-driven member earns lower revenue (€35.27). Furthermore, DSO congestion management costs decrease from 2738.83 €/day in winter to 795.17 €/day in summer, which shows seasonal dependence and user behavior in grid operation.

This paper proposes a robust optimization formulation to calculate dynamic operating envelopes (DOEs) to safely operate unbalanced three-phase distribution systems. Unlike conventional formulations that satisfy network constraints only at the envelope bound, the robust formulation covers the entire envelope range. We formulate a robust non-linear programming (NLP) problem with the full AC power flow equations, as well as an approximate linear programming (LP) model. Numerical simulations are run with real-world data from Belgium and two different distribution test feeders. The paper compares the conventional approaches with their robust counterparts and examines the trade-off between constraint violation and envelope size as well as accuracy and solve time aspects.

As the role of offshore wind generation increases in modern power systems, the need for enhanced modelling techniques becomes critical. In problems involving iterative computations, simplified offshore wind farm models often disregard complex aerodynamic effects in order to reduce computational time. However, those effects have a significant impact on the produced electricity. To address this current modelling limitation, we develop a new topology-aware wind farm surrogate using supervised machine learning techniques. This surrogate model is trained to capture the complex relationship between free flow wind information and electrical output, irrespective of the wind farm layout. This allows bypassing the need to train a different model for each wind farm, while enabling the simulation of farms lacking historical data (such as those in construction phase). This is achieved by enriching the input space of the machine learning surrogate with novel geometric and physics-informed features, making it adaptable to any offshore wind farm configuration.

The increasing penetration of renewable energy introduces variability and uncertainties in power systems. These challenges can be mitigated through the coordinated operation of gas and electricity resources, leveraging the flexibility of gas-fired power plants and the inherent storage capacity of gas pipelines. A key difficulty of this approach lies in accurately modeling gas flow dynamics that are typically represented by the nonconvex Weymouth equation. To address this, the paper presents a two-stage Learning-to-Optimize (L2O) framework that bypasses the complexity of explicit gas flow modeling, enabling fast and reliable solutions for the coordinated day-ahead power and gas dispatch problem. The proposed L2O process uses advanced recurrent neural network architectures to forecast the next-day operational states of the gas system based on historical system data. Then, the predicted operating points are used to linearize the Weymouth equation, allowing the resulting linear constraints to be embedded into the optimization problem for improved tractability. We finally introduce a trust region to constrain solutions within a physically valid neighborhood of the predicted state. Numerical results demonstrate that our L2O approach achieves cost efficiency comparable to state-of-the-art formulations, while significantly reducing computation time.

Optimizing day-ahead wind power forecasts is critical for reliable grid integration of renewable energy. Day-ahead wind power prediction is challenging because wind output is nonlinear and nonstationary; depends on multiple changing meteorological inputs; and many classical or single-model approaches struggle with both short-term variability and long-term temporal dependence, especially during extreme events. This study proposes a hybrid model to mitigate these challenges that integrates the data-cleaning pipeline (FCM–Mahalanobis distance–ANN) with CNN-LSTM for high-accuracy day-ahead wind power forecasting, thereby strengthening performance through improved preprocessing and training to enhance both accuracy and speed, while also comparing its performance against classical and state-of-the-art models. All models were trained on the same real-world SCADA measurements, using a consistent seven-day sliding window. Prior to training, data underwent a uniform three-stage preprocessing pipeline, including (1) fuzzy C-means clustering was employed to partition the dataset into clusters, (2) Mahalanobis distance filtering to remove gross outliers, and (3) an ANN-based refinement stage to detect residual anomalies. The model CNN–LSTM used convolutional layers to extract spatial features and LSTM units to model temporal dependencies. Despite recent advances in transformer-based and attention-based models, the proposed hybrid model remains a robust and competitive approach for wind power forecasting, particularly in operational or resource-constrained environments, providing an effective balance between performance, architectural simplicity, and readability with lower computational costs and faster deployment readiness. The proposed hybrid model achieves a MAPE of 7.5%, an MAE of 80 kW, an RMSE of 105 kW, and an R² of 0.95, which are comparable to state-of-the-art models such as Autoformer (6.8%, 74 kW, 98 kW, 0.96) and Attention-LSTM (7%, 76 kW, 100 kW, 0.95) while exceeding the performance of classical architectures, such as SVR (11.2%, 135 kW, 155 kW, 0.89) and ANN (12%, 145 kW, 165 kW, 0.87). Furthermore, on days with extremely high or low power output, the proposed hybrid model achieves a Combined Accuracy (CA) score of 90.1%, which integrates multiple evaluation metrics to provide a more comprehensive assessment of performance, surpassing classical models and competing with recent models. The results indicate that using three-phase preprocessing on a CNN–LSTM lowered the MAPE by about 26%, enhanced key error metrics by 10–12% and raised combined accuracy by more than 3 percentage points. These findings highlight the critical importance of accurate prediction in wind power forecasting.

Heating, Ventilation, and Air Conditioning (HVAC) is a major electricity end-use with a substantial potential for providing grid services, such as demand response. Harnessing this flexibility requires accurate modeling of the thermal dynamics of buildings, a difficult task because nonlinear heat transfer and recurring daily cycles make historical data highly correlated and insufficient to generalize to new weather, occupancy, and control scenarios. This paper presents an HVAC management system formulated as a Mixed Integer Quadratic Program (MIQP), where Neural Network (NN) models of thermal dynamics are embedded as exact mixed-integer linear constraints. Unlike traditional training approaches that minimize prediction errors, we employ Decision-Focused Learning (DFL) to learn the NN parameters with the objective of directly improving the HVAC cost performance. However, the discrete nature of MIQP hinders DFL, as it leads to undefined and discontinuous gradients, thus impeding standard gradient-based training. We leverage Stochastic Smoothing (SS) to enable efficient gradient computation without the need to differentiate the MIQP. Experiments on a realistic five-zone building using a high-fidelity simulator demonstrate that the proposed SS-DFL approach outperforms conventional identify-then-optimize (i.e., the thermal dynamics model is identified on historical data then used in optimization) and relaxed DFL methods in both cost savings and grid service performance, highlighting its potential for scalable, grid-aware building control.

Accurate short-term wind power forecasting is crit-ical for ensuring grid stability and optimizing energy market operations. In this work, we propose a two-stage feature selection framework tailored to high-dimensional, metadata-rich fore-casting tasks. Our method integrates a group-based backward elimination strategy, using a structured feature metadata such as data source, quantities, and forecast delay, with a weighted mutual information criterion that balances predictive relevance and redundancy. All selection parameters, including backward pruning strategies and mutual information weights, are jointly tuned within a unified cross-validation framework. We evaluate our approach on a real-world wind power forecasting problem using extensive historical and meteorological data. Results show that the proposed method significantly outperforms two indus-trial forecasters used by Belgian TSO, one based on numerical weather prediction (NWP) and another on machine learning model.

Accurately modeling the thermal dynamics within buildings is critical for optimizing thermostatically controlled loads (TCLs), yet remains challenging due to system nonlinearities. Physics-based and data-driven approaches represent two primary paradigms for characterizing complex thermal dynamics. Notably, data-driven methods can directly model system behaviors from operational data without requiring explicit domain knowledge. Current data-driven approaches typically train one thermal dynamics model independently for each building, facing the generalization issue under new and unknown temperature sequences in buildings. In this work, we propose a generalizable thermal dynamics modeling method by coordinating multiple buildings. In particular, we formulate a federated learning framework to facilitate collaborative modeling in a privacy-preserving way. We further identify and represent dual heterogeneity in both model structures and data distributions for federated learning, and then propose a two-level personalization strategy combining similarity matrices and adaptive weighting to alleviate the impacts. We simulate the performance in incentive and price-based demand response scenarios. Results show that the proposed method can enhance the accuracy of temperature regulation and reduce energy consumption costs by improving the generalization of the trained model.

Energy communities offer various socio-economic advantages to their members, such as competitive internal electricity prices. Nevertheless, the efficacy of these competitive prices might be influenced by the bounded rational behavior of the members. This study explores the concept of limited observability as a paradigm of bounded rational behavior toward the internal pricing strategies of the community manager who is exposed to incomplete information regarding the real-time consumption of community members as well. Stochastic robustness near-optimality bilevel programming problem is proposed for energy sharing management in centralized energy communities. The proposal enables the users to mitigate the decision uncertainties arising from limited observability, as well as data uncertainties related to PV power generation. The investigation focuses on understanding the near-optimal decision-making of community members due to limited observability and potential impacts on community energy exchanges. Simulation results show that mitigating limited observability and PV generation uncertainties leads members to reduce their total consumption by an average of 14.64%. Therefore, the CM lowers internal purchasing prices by 24.82% to promote energy purchases by the members. Additionally, limited observability improves user social welfare by an average of 12.65% and reduces the CM’s daily power purchases from the external supplier by approximately 17%.

The Optimal Power Flow (OPF) problem is the cornerstone of power systems operations, providing generators' most economical dispatch for power demands by fulfilling technical and physical constraints across the power network. To ensure safe and reliable operation of power systems, grid operators must steadily solve the nonconvex nonlinear OPF problem for immense power networks in (near) real-time, which poses tremendous computational challenges. The enormous amount of available data created by power systems digitalization and recent breakthroughs in machine learning have opened up new opportunities for grid operators to build shortcuts to predict or solve the OPF problem close to real-time. This survey overviews recent attempts at leveraging machine learning algorithms to solve the transmission-level OPF problem. On this basis, the groundwork is laid for commonly employed machine learning approaches leveraged to address the OPF problem. Subsequently, the frequently used performance evaluation metrics in learning-based OPFs are delineated to judge efficiency from diverse aspects (e.g., optimality in terms of the dispatched cost, feasibility concerning technical constraints, and computational efficiency) compared to conventional approaches. Next, the trend and progress of recently developed algorithms are discussed. Finally, the challenges and open problems at the interface of machine learning and OPF problems are highlighted.

The potential of Energy Communities (ECs) to foster local private investment in renewable energy production has been highlighted in various recent studies. Almost all these works assume that all investment decisions are taken at year 1, as well as static EC memberships throughout its lifetime. However, as part of a wider energy system, ECs may see their composition evolve with time as founding members may leave or other end-users may join. This uncertainty on the EC’s dynamic composition induces uncertainty on the real cost savings of its members. From this perspective, this work aims at quantifying the impact of newcomers on the profits from investments decided by founding members. To this end, an initial optimal sizing problem is solved before processing a Monte-Carlo analysis on the ECs composition’s evolution. Results collected on a test case composed of 92 end-users show that founding members can lose up to 25% of their expected savings by welcoming new members if no recourse actions are taken on top of the initial optimal investment.

Wind farm layout optimization usually aims at maximizing annual energy production by placing wind turbines in a strategic way to avoid wake losses. However, this might not lead to optimal profits because of the volatility of electricity prices. Moreover, with the growing unpredictability and variability of future power systems due to the increase in renewable electricity production, wind farm operators will have a more important role in balancing the system through participation in reserve markets. This study presents a new formulation for wind farm layout optimization where the objective function aims at maximizing revenues from both day-ahead and reserve markets. It uses stochastic gradient descent for the optimization and probabilistic forecasts for wind power and electricity prices. The new formulation is applied to a test case based on a real-life offshore wind farm in Belgium. An important conclusion is that annual revenues are expected to increase in a significant way when accounting for participation in reserve markets, while exhibiting a lower supplied energy production. Moreover, layouts optimized for revenue maximization with reserve participation tend to show better yearly revenues than when considering the day-ahead market only in the objective function. Expected revenues are also higher by 0.18 % on average for the new methodology than when using the maximization of annual energy production, widely used in the literature, as the objective function.

Battery Energy Storage Systems (BESSs) are particularly well-suited to deepen the decarbonisation markets, traditionally dominated by non-renewable generators. BESSs operators often rely on Predict Optimise (PTO) methods to participate in these markets, which focus on forecasting market conditions directly considering the impact of subsequent decisions during training. Recently, learning models have to incorporate decision outcomes during training, known as Decision Focused Learning (DFL) methodologies, which have the potential to increase market benefits. This paper introduces a DFL approach that integrates decision-making process of BESSs when participating in reserve markets into the training of their models. By expressing the optimisation problem as a primal-dual mapping using the Karush-Kuhn-Tucker (KKT) conditions, the proposed DFL method enables the regressor to learn from the BESS's decisions, its predictions based on observed outcomes, improving decision accuracy and market performance. show that the proposed DFL approach outperforms traditional PTO methods, with up to a 9.5% in profits for a case study based on the Belgian secondary reserve market, highlighting its effectiveness managing the complexities of dynamic market conditions.

Compressed Air Energy Storage (CAES) and Cryogenic Energy Storage (CES) are emerging as promising technologies for sustainable grid-scale applications. To surmount the capacity and geological limitations of traditional CAES systems, this study capitalizes on the hybridization of above-ground CAES with CES, utilizing energy conversion between compressed and liquid air. Here, we develop a comprehensive mathematical model for the operation of the hybrid CAES-CES plant, incorporating discrete constraints to manage internal energy transfers and coordination. The model is leveraged to develop the: i) look-ahead dispatch schedule over the following days to enhance adaptability in managing stored energy to maximize benefits, and ii) strategic behavior in electricity markets through unified offers/bids submission. The dispatch problem is structured as a bi-level optimization, with the lower-level addressing market-clearing processes and the upper-level handling storage profit maximization. We reformulate the bi-level setup into a mixed-integer programming model using a mathematical program with equilibrium constraints. To mitigate the computational burden associated with the large number of integer variables in the optimization, we implement a learning-assisted framework for warm-starting these variables. Numerical results show that the hybrid plant can yield up to a 9.08% profit improvement over the standalone alternative under the look-ahead strategy. Further, results demonstrate that under the bi-level setup, the warm-start strategy effectively reduces computation time by 29.30% and 13.35% in the 24- and 118-bus networks, respectively.

This paper addresses important challenges in wind energy prediction caused by outliers in wind data, which distort the wind turbine power curve and lead to inaccurate performance assessments and suboptimal operation strategies. The major difficulty here is detecting and eliminating these outliers from complex wind datasets, as inaccurate data can significantly impact forecasting and related activities. To overcome this challenge, the paper proposes a hybrid model combining fuzzy C-means clustering, Mahalanobis distance, and Artificial Neural Networks (ANN) to detect and remove outliers far more accurately than any individual method or other traditional hybrid method, decreasing false alarms and misses. It improves data quality and boosts the reliability of turbine performance analysis, resource assessment, and forecasting, supporting more efficient and sustainable wind-power operations. The results show (1) that the proposed hybrid model achieves 15.4 % more accuracy than the other traditional hybrid models in detecting and removing outliers. (2) The proposed hybrid model gives an overall ≈ 116.1 % improvement in outlier-detection accuracy over the individual models. (3) Adding the ANN to the proposed hybrid model boosts the outlier-detection accuracy to about a 69.5 % relative improvement. (4) Detecting and cleaning outliers by the proposed hybrid model cuts the RMSE from 2.38 to 1.27, reducing prediction error by 46.6 %. (5) The advanced hybrid model used in this study for comparison purposes achieves nearly identical accuracy to the proposed hybrid model; it reduces RMSE by ∼0.015 and MAPE by ∼0.04 pp and boosts R² by ∼0.001 while maintaining almost perfect outlier detection (99 % vs. 100 %). Although the advanced model offers a marginal edge in reconstruction quality, the lightweight, scalable proposed hybrid model remains better appropriate for real-world deployment due to its lower computational overhead and more straightforward maintenance.

Power exchanges at Transmission-Distribution interfaces are crucial for both the Transmission System Operators (TSOs) and the Distribution System Operators (DSOs). In the past, simple hypothesis as a constant power factor sufficed for characterizing distribution networks and predicting power flows at Transmission- Distribution interfaces. However, the growing integration of distributed energy resources has led to an increased volatility in both active and reactive power flows, rendering traditional models less effective. This study presents a novel Physics-Informed Machine Learning (PIML) model designed to enhance the prediction of power exchanges at Transmission-Distribution interfaces. A novelty of the model lies in its combination of an Inverse Load Flow formulation, which defines an equivalent model of the distribution network (by calculating equivalent resistance and reactance using load flow equations), with classical data-driven regression techniques. Simulation results conducted on a modified version of the Oberrhein MV network highlight the superiority of the proposed PIML approach in front of full ML based methods, as demonstrated by a statistical indicator and an application-oriented evaluation. In addition, this research adopts the TSO perspective through a 2-step Optimal Power Flow analysis that integrates interface power predictions and enables the calculation of production and deviation costs. This multifaceted approach provides valuable insights into the practical implications of the power prediction accuracy on the TSO decision-making process and underscores the significance of accurate power exchange forecasts in the evolving electricity landscape.