Short-term Wind Speed Prediction Based on Improved Auto Encoder

Abstract In order to improve the wind power consumption level of new system with energy as main body, it is necessary accurately predict speed. The key refine dynamic trend and potential physical structure in speed sequence. Firstly, on basis library Koopman dynamics theory encoder structures, a physically constrained spatio-temporal neural network built, which generates linear evolution moment nonlinear variables farm. Secondly, approximated by matrix, forward backward are fully considered prediction process. Then, bidirectional correlation mechanism cost function adapted different objects set reduce requirements for reversibility stability Meanwhile, hidden vector feature space visualized show interval dependency system. Finally, effectiveness proposed method verified measurement data Prince Mountain Beipiao. results that has high accuracy, strong generalization ability, interpretability random fluctuation series.