Real-time Prediction of Wind Power Based on Error Following Forget Gate-based Long Short-term Memory

Because of the fact that the update mode of the forget gate of standard long short-term memory (abbr. LSTM) could not reflect the correction of predicated error to the model based prediction value in real time, a real-time wind power prediction model utilizing error following forget gate (abbr. EFFG)-based LSTM was proposed. The error between the predicted value of wind power and the actual value at the previous moment was used to update the forget gate, thus the influence of the predicted error of the wind power at the previous moment on the prediction accuracy at current moment could be reduced, and the rolling prediction accuracy of wind power could be improved. The historical wind power data of a certain actual wind farm and the numerical forecasted meteorological data were utilized to verify the proposed model. Verification results show that the root-mean-square error of the value predicted by the proposed real-time wind power prediction model based on EFFG-based LSTM is less than 3%, and both accuracy and acceptability reach more than 90%, thus such results are better than the prediction accuracy by the models based on support vector machine and standard LSTM model.