| Wind energy is an excellent energy,which is necessary to get short-term wind speed prediction in order to get stable,safe and effective wind energy.Wind farms are important places to obtain the wind resources,which are susceptible to the influence of the wind speed fluctuation,besides,reduces stability and conversion efficiency of the farms.Ensemble Empirical Mode Decomposition(EEMD)is an adaptive decomposition algorithm that can reduce the mutual interference between components,the combination method is widely use in wind prediction.The full text takes EEMD as the core to describe research work,the main contents of the thesis are as follows:Firstly,the research on status of wind speed prediction and empirical mode decomposition at home and abroad.Two basic theories of the time series and its analysis methods are summarized: traditional time series and neural network method.The autoregressive mean moving average(ARMA)and BP model are explained in detail,In addition,expounded the general ideas of EEMD-based short-term wind forecast algorithm.Secondly,in view of the low accuracy of the traditional EEMD-ARMA model,the EEMD-KF-ARMA model is proposed.System state equation and the observation equation are obtained by ARMA model,to get more accurate results,The Kalman filters(KF)algorithm is used to correct the sub-wind speed sequence ARMA through EEMD decomposition.Compared with the traditional ARMA,MSE is reduced by 55.5,while the MAPE 38.89%.Thirdly,in order to make full use of the advantages of long-term and short-term memory network(LSTM)for time series processing,EEMD-LSTM,a combination of LSTM and EEMD is proposed.the model decomposes the wind sequence into several sub-wind results,then applied the LSTM algorithm to obtain the sub-prediction results.The final results are obtained by simple linear superposition of the sub-prediction.Experiments show that compared with the traditional EEMD-ARMA model,the MSE of EEMD-LSTM is reduce by 68.18% while MAPE is 53.13%,which is effective method to predict wind power.Finally,improved EEMD-LSTM model is raised to solve partial low-accuracy of EEMD-LSTM components.The key ideas of the model are important components value,the larger value of it,the larger potential improved possibilities.Practically,the improved EEMD-LSTM has less MSE and MAPE,compared to EEMD-LSTM,it reduced about 23.41%,25.36% respectively.In summary,the improved EEMD-LSTM which is proposed in this paper is a effective method with high precision,has a positive significance for predicting short-term wind speed. |
