群智能算法在月径流预测支持向量机建模中的适应性研究  

Adaptation of Swarm Intelligence Algorithm in Support Vector Machine Modeling for Monthly Runoff Prediction

作　　者：毛建刚 王庆杰 MAO Jian-gang;WANG Qing-jie(Xinjiang Institute of Water Resources and Hydropower Research,Urumqi 830000,China;Sichuan Water Development Investigation,Design&Research Co.,Ltd.,Chengdu 610065,China)

机构地区：[1]新疆水利水电科学研究院,新疆乌鲁木齐830000 [2]四川水发勘测设计研究有限公司,四川成都610065

出　　处：《水电能源科学》2024年第5期34-38,共5页Water Resources and Power

基　　金：新疆维吾尔自治区重大科技专项(2022A02003-5)。

摘　　要：变化环境下,径流的精准预测愈加困难,集成稳健的优化算法是近年来机器学习发展的动力,也是提高径流预测精度的有效途径。为规避支持向量机(SVM)建模时适应度函数选取的不足,提出基于群智能算法的SVM参数优化范式,并以新疆克孜尔水库月平均径流量预测为例,对粒子群算法(PSO)、差分算法(DE)、灰狼算法(GWO)、鲸鱼算法(WOA)和麻雀算法(SSA)5个典型算法进行仿真验证。结果表明,PSO建立SVM模型合格率(QR)小于60%,预测精度不达标,其余4类算法的平均绝对相对误差和纳什系数分别介于10%~20%和0.75~1之间,预测效果良好;20次独立运算结果中,PSO、WOA、SSA和DE存在预测结果较差的情况,其中PSO的稳定性最差。综合而言,GWO优化的SVM(SVMGWO)在月径流预测中精度、稳定性和可靠性更佳。Accurate prediction of runoff in changing environments is becoming more and more difficult.The integration of robust optimization algorithms has been the driving force behind the development of machine learning in recent years and is an effective way to improve the accuracy of runoff prediction.In order to prevent the inadequacy of the selection of the fitness function in the SVM optimization modeling process,the SVM parameter optimization paradigm based on swarm intelligence algorithm is proposed.Five typical algorithms,particle swarm algorithm(PSO),difference algorithm(DE),gray wolf algorithm(GWO),whale algorithm(WOA)and sparrow algorithm(SSA),are simulated and validated by taking the monthly average flow prediction of Xinjiang Kizil reservoir as an example.The results show that the qualified rate(QR)of the SVM established by PSO is less than 60%,and the prediction accuracy is not up to standard.The mean absolute relative error(MAPE)and Nash coefficient(NSE)of the SVM models established by the remaining four algorithms are between 10%-20%and 0.75-1,respectively,with good prediction results.Among the results of 20 independent operations,there are poor prediction results for PSO,WOA,SSA and DE,among which,PSO has the worst stability.In general,the support vector machine model(SVMGWO)optimized by the gray wolf algorithm has the best accuracy,stability and reliability in the monthly runoff prediction.

关 键 词：径流预测 支持向量机 群智能算法 交叉验证 参数优化 

分 类 号：TV121.4[水利工程—水文学及水资源]