一种基于物理核函数高斯过程回归的月径流预报模型及其应用  

Monthly Runoff Prediction Model and Its Application Based on GPR with Physically Composite Kernel

作　　者：孙娜 张楠 张帅 彭甜 周建中[4] 张海荣 SUN Na;ZHANG Nan;ZHANG Shuai;PENG Tian;ZHOU Jian-zhong;ZHANG Hai-rong(Faculty of Automation,Huaiyin Institute of Technology,Huaian 223003,China;Faculty of Mechanical and Material Engineering,Huaiyin Institute of Technology,Huaian 223003,China;Xi’an Shufeng Technological Information Ltd.,Xi’an 710054,China;School of Civil and Hydraulic Engineering,Huazhong University of Science and Technology,Wuhan 430074,China;China Yangtze Power Co.,Ltd.,Yichang 443133,China)

机构地区：[1]淮阴工学院自动化学院,江苏淮安223003 [2]淮阴工学院机械与材料工程学院,江苏淮安223003 [3]西安数峰科技信息有限公司,陕西西安710054 [4]华中科技大学土木与水利工程学院,湖北武汉430074 [5]中国长江电力股份有限公司,湖北宜昌443133

出　　处：《水电能源科学》2023年第4期39-43,共5页Water Resources and Power

基　　金：江苏省高校自然科学基金项目(20KJD480003);江苏省双创计划(JSSCBS(2020)31035);江苏省自然科学基金(BK20201069);国家自然科学基金项目(91547208,51909010)。

摘　　要：鉴于传统的单一径流预报模型很难描述径流未来变化规律,将自适应变分模态分解(AVMD)与基于组合物理核函数的高斯过程回归(GPR-CK)相结合,构建了AVMD-GPR-CK预报模型,该模型采用AVMD将实测径流分解为多个子序列,对子序列依据其自身特点分别建模,子序列预报结果叠加重构即为最终预报结果。模型应用于金沙江流域向家坝站未来1~12个月的径流预报的结果表明,所有预见期AVMD-GPR-CK模型的确定性系数均大于0.94,平均绝对百分比误差(M_(MAPE))在±17%以内,预见期在10个月以内时,M_(MAPE)在±10%以内;预报精度明显优于常见的BP、GRNN、RBF、RELM模型。In light of the difficulty of traditional single runoff prediction models to describe future variation in runoff,a monthly runoff prediction model named AVMD-GPR-CK based on adaptive variational modal decomposition(AVMD)and Gaussian process regression(GPR-CK)with physically composite kernel was proposed.In the proposed model,the runoff series was decomposed into several subseries using AVMD.Then subseries were separately modeled according to their own characteristics,and the final prediction result was the superposition of the subsequence prediction results.The AVMD-GPR-CK was applied to forecast the future 1-12 months runoff at Xiangjiaba station in the Jinsha River basin.The results show that the deterministic coefficient of the AVMD-GPR-CK model is greater than 0.94,and the mean solute ab-percentage error(M_(MAPE))is within±17%for all leading times,and the M_(MAPE) is inside±10%for leading times within 10 months.Furthermore,the accuracy of the AVMD-GPR-CK is significantly better than those of the commonly used BP,GRNN,RBF,and RELM Key models.

关 键 词：月径流预报 变分模态分解 高斯过程回归 组合核函数 

分 类 号：TV124[水利工程—水文学及水资源] P338[天文地球—水文科学]