基于EWT-EVO/CDO-GPR模型的三峡入库月径流预测  

Monthly Runoff Prediction of Three Gorges Reservoir Based on EWT-EVO/CDO-GPR Model

作　　者：徐荣华 崔东文 XU Ronghua;CUI Dongwen(Yunnan Water Resources and Hydropower Investment Co.,Ltd.,Kunming 650051,China;Yunnan Province Wenshan Water Bureau,Wenshan 663000,China)

机构地区：[1]云南省水利水电投资有限公司,昆明650051 [2]云南省文山州水务局,云南文山663000

出　　处：《三峡大学学报(自然科学版)》2025年第2期26-32,共7页Journal of China Three Gorges University:Natural Sciences

基　　金：云南省创新团队建设专项(2018HC024);云南重点研发计划(科技入滇专项);国家澜湄合作基金项目(2018-1177-02)。

摘　　要：为提高三峡入库月径流预测精度,提出一种基于经验小波变换(EWT)和能量谷优化(EVO)算法、切尔诺贝利灾难优化(CDO)算法优化的高斯过程回归(GPR)预测模型.首先利用EWT将月径流时间序列分解为趋势项、周期项和波动项;然后介绍EVO、CDO算法原理,利用EVO、CDO优化GPR超参数;最后利用优化获得的最佳超参数建立EWT-EVO-GPR、EWT-CDO-GPR模型对月径流各分量进行预测,重构后得到最终预测结果,并构建基于粒子群优化(PSO)算法、遗传算法(GA)优化的EWT-PSO-GPR、EWT-GA-GPR模型,基于支持向量机(SVM)、BP神经网络的EWT-EVO-SVM、EWT-CDO-SVM、EWT-EVO-BP、EWT-CDO-BP模型,基于小波变换(WT)的WT-EVO-GPR、WT-CDO-GPR模型,基于经验模态分解(EMD)的EMD-EVO-GPR、EMD-CDO-GPR模型和EWT-GPR、EVO-GPR、CDO-GPR模型作对比分析,通过三峡2009至2022年入库月径流时序数据对各模型进行验证.结果表明:EWT-EVO-GPR、EWT-CDO-GPR模型预测的平均绝对百分比误差分别为0.689%、0.699%,决定系数均为0.9999,优于其他对比模型,具有更好的预测效果;EWT兼顾WT、EMD优势,可将月径流时序数据分解为更具规律的子分量,显著提升模型性能,分解效果优于WT、EMD;EVO、CDO对GPR超参数的寻优效果优于PSO、GA,通过超参数寻优,显著提升了GPR性能;在相同情形下,GPR预测性能要优于SVM、BP.To improve the accuracy of monthly runoff prediction for the Three Gorges reservoir,a Gaussian process regression(GPR)prediction model based on empirical wavelet transform(EWT),energy valley optimization(EVO)algorithm and Chernobyl disaster optimization(CDO)algorithm is proposed.Firstly,EWT is used to decompose the monthly runoff time series into trend term,periodic term and fluctuation term.Then the principle of EVO and CDO algorithms is briefly introduced,and GPR parameters are optimized by using EVO and CDO.Finally,the EWT-EVO-GPR and EWT-CDO-GPR models are established to predict the monthly runoff components by using the optimized super-parameters,and the final prediction results are obtained after reconstruction.The EWT-PSO-GPR and EWT-GA-GPR models based on particle swarm optimization(PSO)algorithm and genetic algorithm(GA)optimization,EWT-EVO-SVM,EWT-CDO-SVM,EWT-EVO-BP,EWT-CDO-BP models based on support vector machine(SVM)and BP neural network,and the non-optimized EWT-GPR model are constructed,WT-EVO-GPR and WT-CDO-GPR models based on wavelet transform(WT),EMD-EVO-GPR and EMD-CDO-GPR models based on empirical mode decomposition(EMD)and non-decomposed EVO-GPR and CDO-GPR models are compared and analyzed,and the models are verified by the monthly runoff time series data of the Three Gorges from 2009 to 2022.The results show that:The average absolute percentage errors of EWT-EVO-GPR and EWT-CDO-GPR models for the monthly runoff prediction of the Three Gorges reservoir are 0.689%and 0.699%respectively,and the determination coefficients are 0.9999,which are better than other comparison models,with higher prediction accuracy and better generalization ability;EWT takes the advantages of WT and EMD into account.It can decompose the monthly runoff time series data into more regular modal components,significantly improving the model prediction accuracy,and the decomposition effect is better than WT and EMD;EVO and CDO can effectively optimize GPR parameters,improve GPR prediction performance,and the optimization ef

关 键 词：月径流预测 高斯过程回归 能量谷优化算法 切尔诺贝利灾难优化算法 经验小波变换 三峡 

分 类 号：TV698.2[水利工程—水利水电工程] TV121