The high penetration of distributed generation (DG) and renewable energy sources (RES) exacerbates voltage-voltage-ampere reactive (volt–VAR control, VVC) due to fast uncertainty-driven voltage fluctuations, topology-coupled interactions across feeder regions, and the limited real-time practicality of model-dependent optimization and iterative power-flow solvers. This paper formulates VVC as a Markov game and proposes an integrated, data-driven, topology-aware multi-agent learning architecture. From an artificial-intelligence perspective, we develop a centralized training with decentralized execution (CTDE) multi-agent graph soft actor–critic framework (MAGSAC), where graph convolution is embedded into Actor–Critic networks to enhance topology-aware coordination and stabilize multi-agent learning under partial local observations. To avoid repeatedly invoking iterative power-flow solvers during interaction, an offline-trained power flow neural network (PNN) is embedded into the environment to provide fast voltage estimation. It is trained using Latin hypercube sampling and a Levenberg–Marquardt optimizer, reducing reliance on explicit network parameters. A polygon voltage fortress reward is further introduced for fine-grained penalty shaping near operational limits. From an engineering-application perspective, the proposed architecture enables coordinated reactive-power regulation of inverter-interfaced DGs and compensators under high DG variability, improving voltage compliance and operational economy. Case studies on modified IEEE 33-bus and IEEE 69-bus systems show that MAGSAC achieves zero voltage violation on the 33-bus typical-day test and reduces network losses by 38.13 % versus no control. On the 69-bus system, it attains the lowest voltage-violation ratio among compared strategies and reduces daily loss by 27.18 %. Additional tests under topology changes and measurement-data loss support robustness and generalization under the studied settings.

Accurate wind power forecasting can significantly reduce grid dispatch costs and improve system performance. Consequently, this study proposes a bimodal-driven method based on large language model and interpretable convolutional attention-gated (BD-LLICAG) network for wind power forecast. The BD-LLICAG method utilizes the dataset collected from microsensors in the Internet of Things for training and is compared with 53 classic algorithms. The BD-LLCAG has 53.38% and 32.62% less than the average absolute error metrics of a single interpretable convolutional attention-gated (ICAG) network and DeepSeek large language model, respectively, and has 26.75% less average absolute error metrics than the state-of-the-art inception -embedded attention memory fully-connected network. Moreover, through comparative analysis of ablation experiment results and the Wilcoxon signed-rank test for BD-LLICAG, the dual-model-driven wind power forecasting model—combining DeepSeek with ICAG achieved optimal prediction performance is demonstrated. Compared to the single ICAG network and DeepSeek model, the mean absolute error metrics are reduced by 53.38% and 32.62%, respectively. Finally, the combined application of Shapley values and N-BEATS in tandem is successfully applied to visualize and analyze the extent to which the input wind power feature variables influence the prediction results of the wind power predicted by BD-LLICAG.

In this article, we propose an interpretable incremental voltage–current representation attention convolution neural network for the nonintrusive load monitoring (NILM) task. The proposed method consists of two parts: First, the voltage–current representation attention mechanism in the proposed network is designed in collaboration with the data preprocessing method. They provide the role for the classification function of neural networks; Second, this article proposes an adaptive distillation incremental learning method that introduced incremental learning into the NILM field. In this work, the public dataset plug-load appliance identification dataset is used to validate the proposed voltage–current representation attention mechanism and adaptive distillation incremental learning method in this article. In addition, the performance of the proposed algorithms is also complemented in this article using a private dataset. According to the experimental results, the performance of the proposed method in this article is better than the comparison methods.