基于小波包分解的CFO-LSTELM日供水量时间序列预测研究OA

To improve the accuracy of daily water supply time series prediction and avoid the problem of distortion of prediction results caused by using future information in advance,a cordyceps optimization(CFO)algorithm based on wavelet packet decomposition(WPD)-least squares twin extreme learning machine(LSTELM)prediction model was proposed.Firstly,the daily water supply time series was divided into a training set,a validation set,and a prediction set.One layer of WPD was used to decompose the training set+validation set and the prediction set,respectively,so as to obtain one low-frequency component and one high-frequency component and avoid using future information in advance.Secondly,based on the components of the training set,an LSTELM hyperparameter optimization objective function was constructed.The CFO was used to optimize the LSTELM hyperparameters of each objective function,and a WPD-CFO-LSTELM daily water supply time series prediction model was established.At the same time,11 comparative models were constructed,including the WPD grey wolf optimization(GWO)algorithm/whale optimization algorithm(WOA)-LSTELM model,WPD-CFO-Least squares extreme learning machine(LSELM)/twin extreme learning machine(TELM)/extreme learning machine(ELM)/least squares support vector regression(LSSVR)model,and WPD-LSTELM/ELM/LSSVR model,so as to compare and verify the performance of the WPD-CFO-LSTELM model.Finally,12 models were validated through time series prediction examples of daily water supply at the Old City Water Plant and Jinma Water Plant in Honghe Prefecture,Yunnan Province.The results show that:① The WPD-CFO-LSTELM model has the best model performance and the smallest fitting and prediction errors for two instance training sets,validation sets,and prediction sets,with an average absolute percentage error(MAPE)of≤0.34%,an average absolute error(MAE)of≤11.3 m3/d,a root mean square error(RMSE)of≤14.9 m3/d,and a coefficient of determination(R2)of≥0.999 6.②The optimization accuracy of CFO is significantly better than GWO and WOA,and a better optimization ability of metaheuristic algorithm indicates a higher fitting and prediction accuracy of daily water supply.③ The quality of LSTEL/LSELM/TELM/ELM/LSSVR hyperparameters has a significant impact on model performance.By optimizing hyperparameters through CFO,the accuracy of daily water supply prediction can be significantly improved.④ The WPD-COF-LSTELM model and prediction method proposed in this article provide an effective solution for the accurate prediction of daily water supply.