三峡大学学报（自然科学版）2025，Vol.47Issue(4)：1-10,10.DOI:10.13393/j.cnki.issn.1672-948X.2025.04.001
基于"十二生肖"算法优化的加权极限学习机月径流预测
Weighted Extreme Learning Machine for Monthly Runoff Prediction Optimized Based on the"Twelve Zodiac"Algorithm
摘要
Abstract
To improve the accuracy of monthly runoff time series prediction and enhance the predictive performance of weighted extreme learning machine(WELM),the optimization effects of the"Twelve Zodiac"algorithm on benchmark test functions and instance objective functions were compared and verified.The combined empirical wavelet transform quadratic decomposition(EWTⅡ)technique-"Twelve Zodiac"algorithm-WELM monthly runoff time series prediction model was proposed.Firstly,the empirical wavelet transform(EWT)is used to decompose the monthly runoff time series,obtaining two decomposition components,EWT1 and EWT2;Fuzzy entropy is used to calculate the fuzzy entropy values of EWT1 and EWT2 components,and EWTⅡ is used to perform secondary decomposition on the EWT1 component with higher fuzzy entropy values,resulting in three components,i.e.,EWT1-1～EWT1-3.Secondly,based on the EWT1-1～EWT1-3 and EWT2 component training sets,four optimization instance objective functions for WELM input layer weights and hidden layer biases(hyperparameters)were constructed.At the same time,six benchmark test functions were selected as comparative validation functions,and the"Twelve Zodiac"algorithm was used to perform extreme value optimization and comparative analysis on the six benchmark test functions and four instance objective functions.Finally,the combined EWTⅡ-"Twelve Zodiac"algorithm-WELM model was established,and all 12 models were validated through a monthly runoff prediction example of the Nandong underground river in Yunnan Province.The results showed that:The overall ranking of the"Twelve Zodiac"algorithm for optimizing six benchmark test functions was inconsistent with the overall ranking for optimizing four instance objective functions.Overall,the crested porcupine optimization(CPO)and the dingo optimization algorithm(DOA)had better optimization effects,while the chameleon swarm algorithm(CSA),the beetle antenna searching(BAS),and the self-learning antelope migration algorithm(SAMA)had poorer optimization effects.The overall ranking of the"Twelve Zodiac"algorithm for optimizing the objective functions of four instances is basically consistent with the overall ranking of the prediction accuracy of 12 models,indicating that the stronger the extreme value optimization ability of the"Twelve Zodiac"algorithm,the better the WELM hyperparameters obtained,and the better the performance of the constructed prediction model.The EMAP,EMA and ERMS predicted by the EWTⅡ-CPO/CSO/DOA/CapSA/WHO-WELM models are between 0.422%～0.485%,0.022～0.026 m3/s,and 0.028～0.032 m3/s,respectively,which are better than other comparative models and have better prediction performance.关键词
月径流预测/经验小波变换/二次分解/"十二生肖"算法/加权极限学习机/函数优化Key words
monthly runoff prediction/empirical wavelet transform/quadratic decomposition/"Twelve Zodiac"algorithm/weighted extreme learning machine/function optimization分类
水利科学引用本文复制引用
韩艳,崔东文..基于"十二生肖"算法优化的加权极限学习机月径流预测[J].三峡大学学报（自然科学版）,2025,47(4):1-10,10.基金项目
国家自然科学基金项目(41702278) （41702278）
国家重点研发计划项目(2019YFC0507500) （2019YFC0507500）
中国地质调查局地质调查项目(DD20221758,DD20190326) （DD20221758,DD20190326）
