基于NRBO-SVM模型的月径流预测研究  

Research on Monthly Runoff Prediction Based on NRBO-SVM Model

作　　者：黎宇杰 史国勇 廖毅 李基栋 陈学毅 黄炜斌[4] LI Yujie;SHI Guoyong;LIAO Yi;LI Jidong;CHEN Xueyi;HUANG Weibin(Nanya Branch,CHN Energy Sichuan Power Generation Co.,Ltd.,Ya'an 625400,Sichuan,China;College of Water Conservancy and Hydropower Engineering,Sichuan Agricultural University,Ya'an 625014,Sichuan,China;Chengdu Kinetic Energy Technology Co.,Ltd.,Chengdu 610041,Sichuan,China;College of Water Conservancy and Hydropower Engineering,Sichuan University,Chengdu 610065,Sichuan,China)

机构地区：[1]国家能源集团四川发电有限公司南桠河水电分公司,四川雅安625400 [2]四川农业大学水利水电学院,四川雅安625014 [3]成都动能科技有限公司,四川成都610041 [4]四川大学水利水电学院,四川成都610065

出　　处：《水力发电》2025年第1期16-21,共6页Water Power

基　　金：四川省科技厅应用基础研究项目(2021YJ0544)。

摘　　要：基于冶勒站多年月径流数据,以支持向量机(SVM)作为预测器,从模型输入、模型优化和输出环节探讨了提升月径流预测精度的方法。首先,比较了牛顿-拉夫逊优化算法(NRBO)与灰狼优化算法(GWO)在参数寻优方面的性能,发现均方误差(MSE)作为适应度函数时NRBO表现更优。其次,进一步比较了逐月预测与分月预测的效能,结果显示逐月预测具有更高的预测准确性。此外,还从模型输出环节探索了组合预测输出的效果,发现能有效提升模型的泛化性能。而在数据预处理环节,经变分模态分解(VMD)预处理能大幅降低模型预测难度,同时显著提高预测精度。具体而言,GWO-VMD-NRBO-SVM相比单一模型,平均绝对百分比误差(MAPE)和归一化均方根误差(NRMSE)的降低幅度分别超过68%和79%,而纳什效率系数(NSE)提升超过15%。研究结果对非平稳月径流预测具有一定的参考价值。Based on the multi-year monthly runoff data from the Yeller Station,the methods to improve the accuracy of monthly runoff prediction are explored in terms of model inputs,model optimization and outputs,using the support vector machine(SVM)as a predictor.Firstly,the performance of Newton-Raphson optimization algorithm(NRBO)and Gray Wolf Optimization(GWO)algorithm in parameter optimization is compared,and it is found that NRBO performs better when the mean square error(MSE)is used as the fitness function.Further comparing the efficacy of time series forecasting with split-month forecasting,the results show that time series forecasting has higher forecasting accuracy.In addition,based on above forecasting results,this study also explores the effect of combining the forecasting outputs,which is found to be effective in improving the generalization performance of the model.In the data preprocessing session,preprocessing by variational modal decomposition(VMD)can significantly reduce the difficulty of model prediction while significantly improving the prediction accuracy.Specifically,GWO-VMD-NRBO-SVM reduces the mean absolute percentage error(MAPE)and normalized root-mean-square error(NRMSE)by more than 68%and 79%,respectively,and improves the Nash efficiency coefficient(NSE)by more than 15%compared to a single model.The results of this paper are informative for non-stationary monthly runoff prediction.

关 键 词：月径流预测 支持向量机 参数优化 变分模态分解 

分 类 号：P338.1[天文地球—水文科学]