基于变分模态分解的自适应交叉融合模型及其在月径流预测中的应用  

Adaptive Cross-fusion Model Based on Variational Modal Decomposition and Its Application to Monthly Runoff Prediction

作　　者：孙瑜辉 王庆杰 岳春芳[3] SUN Yu-hui;WANG Qing-jie;YUE Chun-fang(Urumqi Water Industry Investment and Development Group Co.,Ltd.,Urumqi 830002,China;Sichuan Water Development Investigation,Design&Research Co.,Ltd.,Chengdu 610065,China;College of Water Conservancy and Civil Engineering,Xinjiang Agricultural University,Urumqi 830052,China)

机构地区：[1]乌鲁木齐水业投资发展集团有限公司,新疆乌鲁木齐830002 [2]四川水发勘测设计研究有限公司,四川成都610065 [3]新疆农业大学水利与土木工程学院,新疆乌鲁木齐830052

出　　处：《水电能源科学》2025年第3期1-6,共6页Water Resources and Power

基　　金：新疆维吾尔自治区重大科技专项项目(2022A02003-5)。

摘　　要：基于“分解—集成”策略的径流预测模型是现有研究中提高预测精度的主流方式之一。分解—集成建模方式主要有后验试验(HE)、预测试验(FE)和自适应预测试验(AFE)3种方式,已有研究主要聚焦于HE的改进,忽视了各建模方式的实用性研究。基于此,在梳理各类建模方式特性的基础上,以天山山系中两条典型的内陆河为例,选用BP神经网络(BP)、支持向量机(SVM)和随机森林(RF)为基准预测模型,基于变分模态分解(VMD)和互补集合经验模态分解(CEEMD)分别构建多种分解—集成预测模型,并探索了AFE与基准模型交叉融合后的预测能力。仿真结果表明,HE建立时提前使用了测试数据信息,与预测实际不符;FE在测试数据分解时受端点效应影响严重,预测精度极低;AEF符合逐时段观测—滚动分解—实时建模预测的实际,基于VMD和CEEMD的AFE模型对径流极大值的预测精度较高。在AFE类模型中,VMD的适应性更强,可实现流域汛期月径流的高精度预报。基于VMD分解的自适应交叉融合模型能够取得与HE模型相当甚至更高的预测精度,对径流预测精度的提高具有实际意义。"Decomposition-integration"runoff prediction model is one of the mainstream approaches to improve prediction accuracy in existing studies.There are three main approaches of decomposition-integration modeling:hindcast experiment(HE),forecast experiment(FE)and adaptive forecast experiment(AFE).Previous studies have mainly focused on the improvement of HE,with little research on the practicality of various modeling approaches.This paper analyzes the characteristics of various modeling approaches.Various decomposition-integrated prediction models are constructed using three benchmark prediction models(back propagation neural network(BP),support vector machine(SVM),and random forest(RF)and two decomposition algorithms(variational modal decomposition(VMD)and complete ensemble empirical modal decomposition(CEEMD)),respectively.Finally,two typical inland rivers in the Tianshan mountain system are used as examples to verify the prediction performance of various models.In addition,the prediction capability of the cross-fusion of AFE and the benchmark model is also explored.The simulation results show that the HE model is built with test data information in advance,which is not consistent with the prediction reality;However,the prediction accuracy of the FE model is extremely low due to the severe influence of the decomposition endpoint effect;The AEF is consistent with the actual time-by-time observation-rolling decomposition-real-time modeling prediction,and the AFE model based on VMD and CEEMD has a high prediction accuracy for runoff maxima.Among the AFE-type models,VMD is more adaptable and can achieve high accuracy of monthly runoff forecasting in flood season.The adaptive cross-fusion model based on VMD can achieve a prediction accuracy comparable to or even higher than that of HE model,which has practical significance for the improvement of runoff prediction accuracy.

关 键 词：径流预测 端点效应 变分模态分解 后验试验 预测试验 自适应预测试验 

分 类 号：TV121.4[水利工程—水文学及水资源]