Photovoltaic (PV) power forecasting plays a crucial role in renewable energy integration and power system operation, as its accuracy directly affects the security and economic efficiency of the grid. However, PV output is strongly influenced by meteorological conditions and thus exhibits pronounced volatility and intermittency, posing significant challenges for ultra-short-term forecasting. To address this issue, this study proposes a hybrid forecasting model that integrates a parallel architecture combining a multi-scale convolutional neural network and a bidirectional gated recurrent unit to effectively capture both local features and temporal dependencies. In addition, a Kolmogorov–Arnold network is incorporated to enhance the model's ability to represent complex nonlinear patterns. To evaluate the generalization and robustness of the proposed approach, we conduct ablation studies, seasonal and weather-condition comparisons against multiple baseline methods, and multi-step forecasting experiments. The results demonstrate that the proposed method consistently achieves superior forecasting accuracy and stability across diverse scenarios, effectively mitigating the uncertainty arising from the volatility and intermittency of PV power generation.
Jiang et al. (2026) studied this question.