Short-Term Wind Speed Prediction Based On Reformer

Wind energy has become the most promising clean energy because of its large reserves and good foundation.The randomness and volatility of wind speed lead to unstable wind power,which have a negative impact on the economic and safe operation of wind farms.Accurate prediction of short-term wind speed is of great significance to realize accurate prediction of wind power,so as to optimize power grid dispatching,reduce rotating reserve capacity and ensure economic operation of power grid.The inertial and time-varying characteristics of wind speed make it a very challenging task to accurately predict short-term wind speed.Most of the current short-term wind speed prediction methods are based on fixed length historical wind speed segments to predict the future wind speed.These methods have high computational overhead and cannot effectively describe the inertia and time-varying characteristics of wind speed.Therefore,the accuracy of short-term wind speed prediction cannot meet the needs of practical application.This paper focuses on these limitations in short-term wind speed forecasting.The main contents include:(1)Short-term wind speed prediction based on improved Reformer.The improved Reformer can adapt to the time series prediction task,learn the weight of each component in the historical wind speed support segment corresponding to the future wind speed through the attention mechanism,and use the powerful learning ability of the Reformer model to realize the short-term wind speed prediction task.The experimental results show that the improved Reformer model can effectively learn the quantitative relationship between historical wind speed segments and future wind speeds,and achieve accurate short-term wind speed prediction;(2)Short-term wind speed prediction based on VMD-Reformer.Wind speed is a typical non-stationary time series,and its broadband signal characteristics in the frequency domain make it difficult to predict its future change trend.The wind speed sequence is decomposed into narrow-band signals by the variational mode decomposition method,and the concept of adaptive support domain is introduced to describe the inertial characteristics of wind speed.Based on the improved self-attention mechanism,the self-adaptive support domain is estimated,the improved Reformer is used to realize the prediction of each mode,and the prediction results of each mode are finally added together to realize the short-term wind speed prediction.The experimental results show that the combined prediction model improves the accuracy of wind speed prediction while reducing the time cost of wind speed prediction.