Ultra-Short-Term Photovoltaic Power Prediction Based on CNN-LSTM

Considering that the prediction accuracy of photovoltaic power strongly depends on factors such as weather patterns and climate conditions, a ultra short term photovoltaic power generation prediction method based on CNN-LSTM is proposed. Firstly, based on Convolutional Neural Network (CNN), features are selected and training and testing datasets are divided, and local features of the input data are extracted through CNN. Secondly, utilizing Long Short Term Memory (LSTM) networks to extract temporal features of data, fully mining the temporal and spatial correlations of photovoltaic measurement values, and analyzing the error distribution of prediction results to verify the robustness of subsequent control strategies. Finally, the proposed model is validated using data from a certain photovoltaic power station. The prediction results indicate that the predicted trend of photovoltaic power generation under different weather conditions is highly consistent with the actual photovoltaic power generation. Statistical analysis was conducted on the error distribution of the prediction model on the test set to obtain the distribution of photovoltaic power prediction errors and their kernel density estimates. The relative error of 90% of the prediction results was less than 7%.