ABSTRACT The assessment of renewable hydrogen feasibility requires predictive models that are not only accurate but also resistant to target leakage and capable of generalizing across geographically unseen regions. This study evaluates a publicly available global renewable hydrogen dataset containing 2,535 geographically distributed observations and develops a dynamic ensemble hybrid regression (DEHR) framework for renewable hydrogen feasibility assessment. Particular attention is given to the distinction between exogenous pre‐development predictors and downstream variables that may encode information closely related to the target. Accordingly, the revised experimental protocol introduces a leakage‐free primary analysis based only on variables that can reasonably be available before project development, while models containing derived power and hydrogen‐production variables are retained only as secondary diagnostic experiments. All preprocessing, hyperparameter optimization, expert weighting, out‐of‐fold prediction generation, and meta‐learning are performed exclusively within the training data. In addition to random hold‐out testing, repeated cross‐validation and geographically structured validation are used to quantify robustness under distribution shift. The DEHR framework combines heterogeneous base regressors using out‐of‐fold performance‐based weighting and a regularized ElasticNet meta‐learner. Ablation experiments separately evaluate the contributions of expert weighting and meta‐learning. Model interpretation is performed using SHAP, LIME, partial dependence, and individual conditional expectation analyses, with explicit consideration of feature collinearity and explanation stability. The results demonstrate that near‐perfect performance obtained with the extended predictor set is substantially influenced by downstream derived variables and should therefore not be interpreted as evidence of unrestricted real‐world generalization. In contrast, the leakage‐free and spatially structured experiments provide a more conservative and practically relevant estimate of predictive performance. These findings emphasize that methodological separation, spatial validation, and leakage‐aware interpretation are essential when machine‐learning models are used for renewable hydrogen feasibility assessment.
ABSTRACT Accurate regional day‐ahead wind power forecasting is important for renewable energy integration and power system scheduling. However, existing methods often insufficiently exploit multi‐scale spatial information in high‐dimensional multi‐grid numerical weather prediction (NWP) data, and station‐level operating information available before forecasting is rarely represented explicitly. To address these limitations, we propose a multi‐branch forecasting model that integrates multi‐scale NWP representation with station‐level operating priors. The proposed model extracts local spatial features, learns extended‐area meteorological representations through a variational autoencoder and constructs adaptive station‐level neighbourhood features for regional aggregated wind power forecasting. Maintenance reduction factors and curtailment constraint factors are further embedded to characterise changes in station‐level output boundaries under planned operating constraints. The model is evaluated using two real‐world wind farm clusters and compared with representative machine‐learning and deep‐learning benchmarks. Results show that the proposed model achieves normalised mean absolute error (NMAE), normalised root mean square error (NRMSE) and R 2 values of 0.0686, 0.0946 and 0.6656 on Cluster 1, and 0.0795, 0.1101 and 0.7813 on Cluster 2, respectively. Ablation results indicate that station‐level operating priors, adaptive neighbourhood extraction and multi‐scale NWP representation contribute to improved forecasting performance.
ABSTRACT With the rapid growth of renewable energy generation, wind and photovoltaic power plants are increasingly integrated into power grids in clusters, with each plant equipped with an automatic voltage control (AVC) substation for voltage regulation. However, parallel aggregation and hierarchical step‐up connections of multiple substations introduce strong inter‐station coupling and operating uncertainty into regional voltage control, rendering traditional reactive power‐voltage sensitivity (Δ Q /Δ U ) methods developed for individual stations inadequate. Existing studies neither systematically address typical cluster topologies nor enable conventional Jacobian‐matrix‐based methods to explicitly account for active power fluctuations. This paper proposes a practical framework for calculating, simplifying, and correcting the reactive power‐voltage sensitivity of renewable energy clusters. The framework accommodates different cluster topologies, accounts for active power fluctuations and inter‐station coupling, and enables sensitivity updates following topology switching or temporary substation non‐response. First, reactive power‐voltage sensitivity models incorporating active‐power fluctuations are derived for typical structures. Subsequently, based on the dominant factors analysis, a simplified inter‐station decoupling method and a generalised sensitivity calculation algorithm applicable to hybrid cluster topologies are developed. Furthermore, a sensitivity correction strategy is proposed to account for sensitivity‐matrix variations caused by main‐grid topology changes or temporary substation non‐response. Case studies demonstrate that the proposed method can improve the ability of individual substations to follow AVC dispatch commands while enhancing regional voltage regulation performance.
ABSTRACT The problem of loss of excitation (LOE) in synchronous generators (SGs) is one of the most pressing threats to the stability of the electric power system. The increasing integration of inverter‐based resources (IBRs) and FACTS devices introduces operating conditions that can modify the voltage, reactive‐power, and apparent‐impedance trajectories used by conventional LOE protection, potentially affecting its dependability and security in specific network configurations. The current review seeks to systematically outline the protection against the problem of LOE, tracing its progression from the development of conventional protection solutions to the current state‐of‐the‐art intelligent solutions developed today. In this paper, the imperfections of the conventional solutions will be assessed, with the recent development of the most effective protection strategies that combine the physical interpretability of model‐based techniques, such as Dynamic State Estimation (DSE) and internal EMF modelling, with the adaptive pattern recognition of data‐driven methods, including machine learning (ML) and advanced signal processing. The article proposes a new taxonomy according to the logic of the detection, analyses performance benchmarks, and identifies gaps in current research, with special attention to the lack of validation in renewable‐dense grids. Finally, future directions, including sensorless detection, explainable AI (XAI), and PMU‐based real‐time solutions, are delineated.
ABSTRACT Photovoltaic thermal (PVT) systems have the potential to generate both electrical and thermal energy from the same solar collector, rendering them an ideal solution for applications with limited space for solar panels and high energy consumption requirements. However, the performance of these systems is limited by the inefficiencies created in the thermal component of the system. Artificial intelligence (AI) has the potential to offer solutions to the issues created within PVT systems. Specifically, AI algorithms can be utilised to provide improvements to the thermal management of these systems. In this paper, a critical review of the literature on AI‐enhanced PVT thermal management is performed with a specific focus on edge intelligence as the delivery framework of the AI algorithms. The review covers six categories of barriers to the implementation and deployment of AI‐enhanced PVT thermal management systems: hardware resources, data resources, real‐time requirements, generalisation issues, cybersecurity and interpretability concerns, and scalability issues. Enabling technologies examined include energy‐efficient hardware architectures, model compression methods, self‐powered edge nodes, and advanced thermal management for edge hardware. Distributed AI paradigms evaluated include federated learning, multi‐agent reinforcement learning, and edge‐cloud collaborative computing.
