The integration of renewable energy sources into the power grid, especially photovoltaic (PV) systems, has seen a significant upsurge due to the global push for sustainable energy. However, the variable nature of solar energy poses unique challenges in the effective management and control of PV power generation. This study introduces an innovative approach to intelligent decision-making in PV power generation control, leveraging deep learning techniques for temporal data feature modeling. By collecting and analyzing time-series data from PV systems, the proposed method utilizes feature neural computing to model the inherent characteristics of the data, facilitating effective temporal feature classification. The classification results are then employed to inform intelligent control strategies, optimizing the efficiency and reliability of PV power generation. This article proposes an end-to-end neural network structure that can simultaneously mine multi-scale local information and global temporal correlation information in sequences. This structure is primarily achieved through a module. The module consists of two parallel branches. One branch improves the InceptionTime module with depthwise separable convolution to extract multi-scale subsequence information. The other branch employs multi-head self-attention technology with added layer positional encoding to extract global temporal correlation information in sequences.