Short-term Wind Speed Prediction Based on the CEEMDAN-SE-GWO-LSTM Model
In order to reduce difficulty of wind speed prediction due to its nonlinearity and randomness and improve prediction accuracy,the present work proposed a combined prediction model integrating fully adaptive noise ensemble empirical mode decomposition(CEEMDAN),sample entropy(SE),gray wolf optimization algorithm(GWO),and long short-term memory neural network(LSTM)to predict short-term wind speed.The method works by using first CEEMDAN to decompose wind speed data into several modal components,second SE to screen the components and to superimpose modal components with similar SE values which form new subsequences,and third GWO-LSTM model for training and prediction of each subseries whose results of are finally superimposed.The proposed CEEMDAN-SE-GWO-LSTM model was proved by simulative experiment capable of achieving reduction in root mean square error,mean absolute error,and average relative error by 21.7%,44.5%,and 40.9%,respectively,compared with isolated LSTM model,and therefore satisfactorily accurate,stable,and effective in predicting short-term wind speed.
NETL
NSTL
万方数据
