若干新型智能算法优化极限学习机在区域人水和谐度评价中的应用  

Application of Several New Intelligent Algorithms to Optimize Extreme Learning Machines in Regional Human-Water Harmony Evaluation

作　　者：许建伟 崔东文 XU Jianwei;CUI Dongwen(Yunnan Water Conservancy and Hydropower Survey and Design Institute,Kunming 650021,China;Yunnan Province Wenshan Water Bureau,Wenshan 663000,China)

机构地区：[1]云南省水利水电勘测设计院,昆明650021 [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）》2023年第2期7-16,共10页Journal of China Three Gorges University：Natural Sciences

基　　金：国家澜湄合作基金项目(2018-1177-02);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)。

摘　　要：为科学量化和评价区域人水和谐度,改进极限学习机(ELM)评价性能,提出野狗优化算法(DOA)、哈里斯鹰优化(HHO)算法、均衡优化(EO)算法、卷尾猴搜索算法(CapSA)、鸽群优化(PIO)算法与ELM融合的人水和谐度评价模型,通过云南省2006-2020年人水和谐度评价实例对模型进行验证.首先,简要介绍5种算法原理,在不同维度条件下选取4个标准函数对DOA、HHO、EO、CapSA、PIO算法进行仿真验证;其次,从水系统健康、经济社会发展、人水系统协调3个方面遴选18个指标构建人水和谐度评价指标体系和等级标准;最后,采用线性内插和随机选取的方法生成样本构建ELM适应度函数,利用DOA、HHO、EO、CapSA、PIO对适应度函数进行寻优,利用寻优获得的最佳ELM超参数建立DOA-ELM、HHO-ELM、EO-ELM、CapSA-ELM、PIO-ELM模型对实例各年度人水和谐度进行评价,结果与模糊综合评价法、ELM评价结果作对比.结果表明:DOA、HHO、EO、CapSA、PIO算法对4个标准函数的寻优精度由优至劣依次是HHO>PIO>DOA>CapSA>EO;对ELM适应度函数的寻优精度由优至劣依次是HHO>DOA>EO>CapSA>PIO.DOA-ELM、HHO-ELM、EO-ELM、CapSA-ELM模型对预测样本预测的平均绝对百分比误差在0.0124%~0.0198%之间,预测精度较PIO-ELM提高26.9%以上,较ELM提高84.6%以上;对实例2006-2011年人水和谐度评价为“不和谐”,2012-2018年评价为“基本和稭”,2019-2020年评价为“较和谐”;近15年来云南省人水和谐度总体上呈上升趋势,且上升趋势显著.DOA、HHO、EO、CapSA、PIO算法均能有效优化ELM超参数,提高ELM预测性能;DOA、HHO、EO、CapSA优化效果要优于PIO算法.In order to scientifically quantify and evaluate the harmony degree of people and water in the region, and improve the evaluation performance of extreme learning machine(ELM), the wild dog optimization algorithm(DOA), the Harris Hawk optimization(HHO) algorithm, the equilibrium optimization(EO) algorithm, the capuchin monkey search algorithm(CapSA), and pigeon group optimization(PIO) algorithm are proposed to integrate with ELM method for the human-water harmony evaluation model. The models are verified by the human-water harmony evaluation data of Yunnan Province in 2006-2020. First, five algorithm principles are briefly introduced, and four standard functions are selected under different dimensional conditions to simulate and verify DOA, HHO, EO, CapSA, and PIO algorithms. From 3 aspects, 18 indicators are selected to construct the evaluation index system and grade standard of human-water harmony. Finally, the samples are generated by linear interpolation and random selection to construct the ELM fitness function, and the fitness function is determined by DOA, HHO, EO, CapSA, and PIO. The function is optimized, and the DOA-ELM, HHO-ELM, EO-ELM, CapSA-ELM, and PIO-ELM models are established by using the best ELM hyperparameters obtained by the optimization to evaluate the harmony degree of human and water in each year of the example. The comprehensive evaluation method and ELM evaluation results were compared. The results show that: The optimization accuracy of DOA, HHO, EO, CapSA and PIO algorithms for the four standard functions is HHO>PIO>DOA>CapSA>EO in order from good to bad, while the optimization accuracy of the ELM fitness function in order from good to bad is HHO>DOA>EO>CapSA>PIO. The mean absolute percentage error of DOA-ELM, HHO-ELM, EO-ELM, and CapSA-ELM models for forecasting samples is between 0.0124% and 0.0198%. The prediction accuracy is 26.9% higher than that of PIO-ELM and 84.6% better than that of ELM. The evaluation of the human-water harmony in 2006-2011 is “disharmonious”, the evaluati

关 键 词：人水和谐度 极限学习机 群体智能算法 仿真测试 云南省 

分 类 号：TV213.4[水利工程—水文学及水资源]