Short-term output prediction of wind-photovoltaic power based on time-frequency decomposition

This paper proposes a short-term wind and photovoltaic power forecasting framework considering time-frequency decomposition based on bidirectional long short-term memory networks. First, the seasonal and trend decomposition using loess is applied to the original wind and photovoltaic data for time domain decomposition, obtaining trend, seasonal, and residual components. Then, the residual component undergoes variational mode decomposition to further extract features of different frequencies. Next, the maximum information coefficient is used to select features, which is highly correlated with wind and photovoltaic power as input features to the prediction model. Finally, the selected features are input into bidirectional long short-term memory networks for training and prediction. Experimental validation using actual data from a photovoltaic station and a wind power station in Hebei Province, China from July to August 2023, which shows that the proposed method achieves high accuracy and reliability in photovoltaic and wind power output prediction. The proposed time-frequency decomposition with the smallest root mean square error of 0.92 and mean absolute error of 0.58 in photovoltaic prediction, at the same time, the smallest root mean square error of 67.5 and mean absolute error of 48.16 in wind power prediction, significantly outperforming other power prediction methods.