Photovoltaic Power Forecasting Based on IEWT-Attention-LSTM Model

The photovoltaic power generation forecast helps the power dispatching department to make dispatch plans and risk aversion in advance, thereby improving the safety and stability of the power system. Aiming at the defects of unreasonable spectrum division of traditional empirical wavelet transform and the lack of effective analysis of historical power sequences in existing forecasting methods, we proposed an improved empirical wavelet transform method (IEWT), and build a power forecasting model based on it. First of all, the original data set is standardized and correlated to simplify the calculation process. Secondly, IEWT is used to decompose the historical power sequence, and the input vector is constructed according to the decomposition results and related variables. At last, LSTM neural network combining Attention mechanism was used to realize component forecasting and reconstruction, and the final forecast results were obtained. Analysis of calculation examples shows that IEWT divides the sequence spectrum more reasonably and improves the over-decomposition and under-decomposition phenomena. The power forecasting model has higher forecasting accuracy than the single LSTM forecasting model and the EWT-LSTM model.