Short-Term Photovoltaic Power Prediction Based on GWO-CNN-LSTM-MATT Hybrid Model

This paper proposes a photovoltaic power prediction model based on GWO-CNN-LSTM-MATT. Firstly, the convolutional neural network (CNN) is used to extract the spatial features and local dependencies in photovoltaic power data,; Secondly, the time series data are processed by Long Short Memory Network (LSTM); Then, the multi attention time series transformer (MATT) module is introduced, and the grey wolf optimization algorithm (GWO) is combined to dynamically adjust the model's super parameters to further improve the adaptability of the model to complex environmental factors; Through collaborative optimization of multiple modules, short-term and long-term accurate prediction of photovoltaic power is achieved. The simulation experiment takes the data of DKASC photovoltaic power station in Australia as an example, and compares the other three typical prediction methods. The results show that the RMSE of this model in sunny days is reduced by 0.18, and the MAE is reduced by 0.05 compared with the traditional LSTM model; In rainy weather, RMSE decreased by 4.6, MAE decreased by 2.78, and fitting coefficient R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> under both conditions was more than 99%, which verified the robustness and prediction accuracy advantages of the proposed model under complex meteorological conditions.