| Popis: |
For the improvement in prediction accuracy of monthly runoff time series,a prediction model is proposed,which combines the wavelet decomposition (WD),African vultures optimization algorithm (AVOA),and deep extreme learning machine (DELM),and it is applied to the monthly runoff prediction of Yale Hydrological Station in Yunnan Province.Specifically,WD decomposes the time-series data of monthly runoff to obtain highly regular subsequence components,and AVOA is employed to optimize the number of neurons in the hidden layers of DELM;then,the WD-AVOA-DELM model is built to predict each subsequence component,and the prediction results are summated and reconstructed to produce the final prediction results of monthly runoff.Meanwhile,models based on the support vector machine (SVM) and BP neural networks are constructed for comparative analysis,including WD-AVOA-SVM,WD-AVOA-BP,AVOA-DELM,AVOA-SVM,and AVOA-BP models.The results reveal that the average absolute percentage error of the WD-AVOA-DELM model for the monthly runoff prediction of Yale Hydrological Station is 3.02%;the prediction error is far less than that of WD-STOA-SVM and WD-AVOA-BP models,and the prediction accuracy is more than one order of magnitude higher than that of AVOA-SVM,AVOA-SVM,and AVOA-BP models.The result indicates that the proposed model has good prediction performance.In this model,WD can scientifically reduce the complexity of runoff series and raise the prediction accuracy;AVOA can effectively optimize the key parameters of DELM and improve the performance of DELM networks. |
