Short-term Power Prediction of Photovoltaic Power Generation Based on LSTM and Error Correction

In order to improve the stability of photovoltaic power grid connection and make full use of error information to correct the model prediction results, this paper proposes a short-term photovoltaic power prediction model based on long short-term memory (LSTM) and error correction. First, the data is preliminarily predicted by LSTM to generate an error sequence, and then the error sequence is decomposed into submodels of different frequencies by empirical mode decomposition (EMD). Similarity measurement is conducted according to Hausdorff distance ( HD), and each modal component is assigned weights, and then LSTM optimized by Sparrow Search Algorithm ( SSA) is used to predict error modal components. The weighted prediction error is combined with the predicted value to achieve error correction. Through experiments, it has been proven that the model proposed in this article outperforms traditional LSTM models, BP models, and SVM models in evaluation indicators such as root mean square error (RMSE) and mean absolute percentage error (MAPE), verifying the effectiveness of the combined model.