Multi-step prediction of photovoltaic power based on two-stage decomposition and BILSTM

Accurate photovoltaic power forecasting is crucial for power dispatch planning, maintenance scheduling and regulation. In this paper, a hybrid photovoltaic power forecasting model is proposed based on two-stage decomposition, correlation heatmap and bidirectional long-short term memory network. Firstly, appropriate model input is selected from meteorological factors through correlation heatmap. Seconldly, the original time series is decomposed into a series of intrinsic mode functions with different characteristics via using ensemble empirical mode decomposition, and then the first subsequence of the intrinsic mode functions is further decomposed by variational mode decomposition. Thirdly, bidirectional long-short term memory network is utilized to extract the relationship between photovoltaic power sequences and environmental factors, and obtain the prediction results of each subsequences, then the eventual predicted results are obtained by reconstructing the predicted values of each subsequences. The result of example analysis shows that the proposed prediction model is superior to other algorithms in accuracy, which has great prospects in the development of photovoltaic power prediction.