Solar forecasting using ground-based sky image offers a promising approach to reduce uncertainty in photovoltaic (PV) power generation. However, existing methods often rely on deterministic predictions that lack diversity, making it difficult to capture the inherently stochastic nature of cloud movement. To address this limitation, we propose a new two-stage probabilistic forecasting framework. In the first stage, we introduce I-GPT, a multiscale physics-constrained generative model for stochastic sky image prediction. Given a sequence of past sky images, I-GPT uses a Transformer-based VQ-VAE. It also incorporates multi-scale physics-informed recurrent units (Multi-scale PhyCell) and dynamically weighted fuses physical and appearance features. This approach enables the generation of multiple plausible future sky images with realistic and coherent cloud motion. In the second stage, these predicted sky images are fed into an Image-to-Power U-Net (IP-U-Net) to produce 15-min-ahead probabilistic PV power forecasts. In experiments using our dataset, the proposed approach significantly outperforms deterministic, other stochastic, multimodal, and smart persistence baselines models, achieving a superior reliability–sharpness trade-off. It attains a Continuous Ranked Probability Score (CRPS) of 2.912 kW and a Winkler Score (WS) of 33.103 kW on the test set and CRPS of 2.073 kW and WS of 22.202 kW on the validation set. Translating to 35.9% and 42.78% improvement in predictive skill over the smart persistence model. Notably, our method excels during rapidly changing cloud-cover conditions. By enhancing both the accuracy and robustness of short-term PV forecasting, the framework provides tangible benefits for Virtual Power Plant (VPP) operation, supporting more reliable scheduling, grid stability, and risk-aware energy management.

This paper examines the difficulties of managing distributed power systems, notably due to the increasing use of renewable energy sources, and focuses on voltage control challenges exacerbated by their variable nature in modern power grids. To tackle the unique challenges of voltage control in distributed renewable energy networks, researchers are increasingly turning towards multi-agent reinforcement learning (MARL). However, MARL raises safety concerns due to the unpredictability in agent actions during their exploration phase. This unpredictability can lead to unsafe control measures. To mitigate these safety concerns in MARL-based voltage control, our study introduces a novel approach: Safety-Constrained Multi-Agent Reinforcement Learning (SC-MARL). This approach incorporates a specialized safety constraint module specifically designed for voltage control within the MARL framework. This module ensures that the MARL agents carry out voltage control actions safely. The experiments demonstrate that, in the 33-buses, 141-buses, and 322-buses power systems, employing SC-MARL for voltage control resulted in a reduction of the Voltage Out of Control Rate (%V.out) from 0.43, 0.24, and 2.95 to 0, 0.01, and 0.03, respectively. Additionally, the Reactive Power Loss (Q loss) decreased from 0.095, 0.547, and 0.017 to 0.062, 0.452, and 0.016 in the corresponding systems.