基于TVFEMDⅡ-十种鱼群算法-DHKELM模型的日含沙量预测  

Daily Sediment Concentration Prediction Based on TVFEMDⅡ-Ten Fish Swarm Algorithm-DHKELM model

作　　者：邓智予 谢静 崔东文 DENG Zhi-yu;XIE Jing;CUI Dong-wen(Yunnan Water Resources and Hydropower Survey and Design Institute Co.,Ltd,Kunming 650021,Yunnan Province,China;Yunnan province Wenshan Water Bureau,Wenshan 663000,Yunnan Province,China)

机构地区：[1]云南省水利水电勘测设计院有限公司,云南昆明650021 [2]云南省文山州水务局,云南文山663000

出　　处：《中国农村水利水电》2025年第3期61-70,共10页China Rural Water and Hydropower

基　　金：国家自然科学基金项目(41702278);中国地质调查局地质调查项目(DD20221758、DD20190326)。

摘　　要：为提高日含沙量时间序列预测精度,改进深度混合核极限学习机(DHKELM)预测性能,对比验证十种鱼群算法——电鳗觅食优化算法(EEFO)/成吉思汗鲨鱼优化(GKSO)算法/白鲸优化(BWO)算法/白鲨优化(WSO)算法/鲸鱼优化算法(WOA)/金枪鱼优化(TSO)算法/旗鱼优化(SFO)算法/海洋捕食者算法(MPA)/?鱼优化算法(ROA)/蝠鲼觅食优化(MRFO)算法在基准测试函数和实例目标函数上的优化效果,提出时变滤波器经验模态二次分解(TVFEMDⅡ)-十种鱼群算法-DHKELM日含沙量时间序列预测模型。首先,利用TVFEMDⅡ对日含沙量时间序列进行分解处理,得到若干分解分量,合理划分训练集和预测集;其次,基于各分量训练集构建DHKELM超参数优化实例目标函数,同时选取8个基准测试函数作为对比验证函数,利用十种鱼群算法分别对基准测试函数和实例目标函数进行极值寻优与对比分析。最后,建立TVFEMDⅡ-十种鱼群算法-DHKELM模型,通过云南省龙潭站汛期日含沙量预测实例对各模型进行验证。结果表明:(1)十种鱼群算法对基准测试函数寻优总排名与对实例目标函数寻优总排名仅有10%相同,总体上EEFO、GKSO寻优效果较好,ROA、WSO较差。(2)十种鱼群算法对实例目标函数寻优总排名与十种鱼群算法优化的各模型预测精度总排名基本一致,表明鱼群算法极值寻优能力越强,其优化获得的DHKELM超参数越优,由此构建的预测模型性能越好,日含沙量预测精度越高。(3)TVFEMDⅡ-十种鱼群算法-DHKELM模型对实例日含沙量预测的平均绝对百分比误差(MAPE)在0.927%~1.583%之间,模型计算规模小、预测精度高、稳健性能好,具有较好的实用价值和意义。(4)在分解分量十分有限的情形下,TVFEMDⅡ能将复杂的日含沙量时间序列分解为更具规律、更易建模预测的模态分量,大大改进时间序列分解效果,显著提升日含沙量预测精度。In order to improve the prediction accuracy of daily sediment concentration time series,and improve the predictive performance of deep hybrid kernel extreme learning machine(DHKELM),this paper compares and verifies the optimization effects of ten fish swarm algorithms,namely electric eel foraging optimization algorithm(EEFO)/Genghis Khan shark optimization algorithm(GKSO)algorithm/beluga whale optimization(BWO)algorithm/white shark optimization(WSO)algorithm/whale optimization algorithm(WOA)/tuna optimization(TSO)algorithm/sailfish optimization(SFO)algorithm/marine predator algorithm(MPA)/remora optimization algorithm(ROA)/manta ray foraging optimization(MRFO)algorithm,on benchmark test functions and instance objective functions,and proposes TVFEMDⅡ-Ten Fish swarm algorithms-DHKELM daily sediment concentration time series prediction model.Firstly,TVFEMDⅡis used to decompose the daily sediment concentration time series and obtain several decomposition components;Secondly,based on the training sets of each component,the DHKELM hyperparameter optimization instance objective function is constructed,and 8 benchmark test functions are selected as comparative verification functions.Ten fish swarm algorithms are used to perform extreme value optimization and comparative analysis on the benchmark test function and instance objective function.Finally,the TVFEMDⅡ-Ten Fish Swarm Algorithm-DHKELM model is established,and each model is verified through an example of predicting daily sediment concentration during the flood season at Longtan Station in Yunnan Province.The results show that:①The total ranking of the ten fish swarm algorithms for benchmark function optimization is only 10%the same as the total ranking for instance objective function optimization.Overall,EEFO and GKSO have better optimization effects,while ROA and WSO are poor.②The overall ranking of the optimization of the instance objective function by the ten fish swarm algorithms is basically consistent with the overall ranking of the prediction accur

关 键 词：日含沙量预测 时变滤波器经验模态分解 二次分解 十种鱼群算法 深度混合核极限学习机 函数优化 

分 类 号：P338.5[天文地球—水文科学]