Short-Term Wind Power Forecasting Based on VMD Decomposition, ConvLSTM Networks and Error Analysis

Abstract

Improving the predicted accuracy of wind power is beneficial to maintaining the secure operation and dispatching of the power system. Therefore, a combined model consisting of the variational mode decomposition(VMD), Convolutional Long short memory network(ConvLSTM) and error analysis is conducted for short-term wind power forecasting. Firstly, the VMD algorithm decomposes the wind power signal into an ensemble of components with different frequencies; A novel architecture embedding the convolution operation into LSTM network is procured as the preliminary forecasting engine, which is appropriate for extracting the spatial and temporal characteristics of each sub-series. Afterwards, all the predicted sub-signals would be aggregated to obtain the preliminary forecasting results; For the sake of further mining the unsteady features within the raw wind power series, LSTM modelling the trend of error sequence of the preliminary forecasting result is adopted. Eventually, the final forecasting results is obtained by integrating the forecasting error series and preliminary results. As a result, It can be easily demonstrated that by comparing with the contrastive models, the proposed model achieves the highest prediction performance for wind power series which is difficult to capture.