A Short-Term Multi-Site Photovoltaic Power Forecasting Method Based on Hybrid Inception-SpatioTemporal TimesNet

This paper proposes HIST-Net, a hybrid model for short-term photovoltaic (PV) power forecasting across multiple sites. It combines Inception-based multi-scale convolution and spatial self-attention to capture temporal patterns and inter-site dependencies, enhanced by a TimesNet-like structure. Experiments on real PV datasets show that HIST-Net outperforms existing models in both accuracy and generalization.