| In recent years,with the continuous reduction of global fossil energy reserves and the increasingly severe global ecological and environmental issues,more and more countries have increased their research efforts on new energy power generation.Among various new energy power generation technologies,wind energy has become one of the most promising new energy power generation methods due to its characteristics of green environmental protection and large reserves.However,the inherent volatility and randomness of wind energy can pose serious challenges to the stable operation of power systems.In order to effectively address this issue and improve the comprehensive utilization rate of wind energy,accurate ultra-short term forecasting of wind power has important practical significance.In order to improve the accuracy of ultra short term wind power prediction,the following research work has been mainly done:Aiming at the problem that the uncertainty of the input weight _ce and the threshold b_c of the deep extreme learning machine(DELM)affects the prediction effect,this paper applies the improved slime mold algorithm for elite opposition-based learning strategy(ESMA)to the parameter optimization of DELM for the first time,and builds an ESMA-DELM model.The results show that compared to the standard slime mold algorithm(SMA),ESMA algorithm has faster convergence speed and smaller convergence error,which proves the feasibility of ESMA optimizing DELM parameters.The ESMA-DELM model and several other prediction models are compared for power prediction under the same experimental data,and the results show that the ESMA-DELM model has better prediction performance.To address the issue of high volatility in the original wind power data sequence,the variational mode decomposition(VMD)method is used to preprocess the sequence.The number of modes K and the penalty factorαselection in VMD are subject to human factors,and different parameter values can make the decomposition effect vary too much.Therefore,SMA is used to select these two core parameters in VMD to constitute the SVMD decomposition algorithm.After SVMD decomposition,all the subsequences obtained from the wind power data are substituted into the ESMA-DELM model for prediction,and the SVMD-ESMA-DELM combined prediction model is constructed.By comparing the error indicators of the SVMD-ESAM-DELM model with other prediction models,the feasibility of optimizing VMD parameters by SMA and the advantages of the SVMD-ESMA-DELM combined model in wind power prediction are proved.In order to utilize the regular errors generated by the combined model,reduce the workload of prediction experiments,and solve the problem of the limited ability of ESMA-DELM model to handle non-stationary components.In this paper,SVMD is used to decompose the original wind power data into time series with different frequencies,and sample entropy(SE)algorithm is used to calculate the SE values of each series.Subsequences with similar SE values are reconstructed,thereby reducing the workload of subsequent prediction experiments.Then ESMA-DELM is used to model and predict the reconstructed low-frequency signal,and long short-term memory(LSTM)is used to learn and predict the reconstructed high-frequency signal.Then these prediction results are added together,and the error correction is carried out for the results,so that the prediction accuracy can be further improved.Based on this,an SVMD-SE-LSTM-ESMA-DELM combined forecasting model with error correction is proposed.Comparing the prediction results of this model with those of other mainstream models,it is proved that the model proposed in this paper has higher prediction accuracy and can provide more powerful support for real-time scheduling of power grids. |
