Short-term Output Prediction of Photovoltaic Power Stations Based on EWT-LSTM Method

We propose a forecasting model for improving the accuracy of short-term PV generation forecasting, which combines long and short-term memory network (LSTM) with empirical wavelet transform (EWT). Using EWT, the raw load is discretized to obtain a set of empirical mode function (EMF) subsequences with different feature scales. For each different pattern obtained by decomposition, the prediction is performed separately using the LSTM model, and the forecasting outcomes of the different patterns are combined to obtain the ultimate forecasting value. By comparing the prediction results of the BP model, ELM (Extreme Learning Machine), SVM, and LSTM models, we found that the EWT-LSTM model outperforms the other three models and exhibits good performance in short-term PV prediction.