CoFMD-BiLSTM: A Wind and Photovoltaic Power Prediction Method Based on Feature Mode Decomposition and Cooperative Learning Strategy

With the increasing penetration of new energy sources such as wind and solar, accurate source-side power forecasting helps power system dispatchers to obtain new energy pre-output plans and improve the power system's new energy consumption capacity through the coordination of multiple power generation resources. The wind and photovoltaic power data are characterized by large fluctuations, strong nonlinearity, instability, and strong time dependence, etc. Through “Decomposition-Prediction”, the original sequence is decomposed into a number of sub-sequences, which reduces the volatility and randomness of the original sequence, and has been widely used in wind and photovoltaic power prediction. In order to solve the problem that traditional decomposition algorithms have insufficient decomposition capability to extract data features in a targeted and effective way, we propose a wind and photovoltaic power prediction method based on Feature Mode Decomposition and Cooperative Learning Strategy: the CoFMD-BiLSTM model. A novel emerging decomposition algorithm, Feature Mode Decomposition (FMD), originally applied in fault detection, is introduced. By designing a Cooperative Learning Strategy to learn and fit the decomposition sequence information obtained from the FMD decomposition covering different features, which is fed into the BiLSTM model to obtain the wind and light power predictions. The actual operation data of Sotavento Galicia (sg) wind farm in Spain and TuoZhuo photovoltaic farm in Yunnan, China are used. Compared with the two models, VMD-BiLSTM and BiLSTM, the validation results show that the wind and solar power prediction method based on the CoFMD-BiLSTM model can effectively improve the accuracy of wind and solar power prediction.