Abstract: Aiming at the problem of low accuracy of photovoltaic power generation, the paper proposes a model based on optimized decomposition, noise reduction and error correction. The model is divided into three stages.In the first stage, the global search-based whale optimization algorithm（GSWOA） is used to select the parameters of the variational modal decomposition（VMD）,and then the optimized VMD is used to decompose the original data; then the high-frequency components are reconstructed using mutual correlation analysis, and finally wavelet soft-threshold noise reduction（WTSD） is applied to the high-frequency components; in the second stage, a gated recurrent unit（GRU） is used to predict each component, and the initial prediction results are obtained by superposing all the component prediction results; in the third stage, the initial prediction results are subjected to error correction（EC）. In order to verify the effectiveness of the model, the paper utilizes the measured PV data from Ningxia Sun Mountain PV power station in January, April, July and October 2021 to conduct experiments, and the experimental results show that the hybrid model exhibits better performance than LSTM, GRU, and VMD-LSTM.