多代理模型自动构建及大坝反演分析应用研究  

Research on Automatic Construction of Multi-surrogate Model and Application of Dam Inversion Analysis

作　　者：孙广正 戴妙林[1] 赵兰浩[1] 朱开滔 SUN Guangzheng;DAI Miaolin;ZHAO Lanhao;ZHU Kaitao(College of Water Conservancy and Hydropower Engineering,Hohai University,Nanjing 210024,Jiangsu,China)

机构地区：[1]河海大学水利水电学院,江苏南京210024

出　　处：《水力发电》2023年第7期88-94,共7页Water Power

基　　金：国家重点研发计划(2022YFC3005403);中国电建集团科技项目(DJ-ZDXM-2021-10)。

摘　　要：基于二分法进行自变量子空间的划分,依据已有模型的预测误差及预期标准,实现不同规模的子空间中样本采集及模型训练、检验和保存流程的自动执行,直至所有子空间所属代理预测模型均达到预期精度。采用支持向量回归(SVR)训练建模,并通过粒子群算法(PSO)优化获得PSO-SVR代理模型。针对显式函数的多代理模型自动构建和验证结果表明,相比于初始全局代理模型,空间划分多代理模型在全局预测精度和局部误差控制方面均有明显提高。为此,建立基于PSO-SVR多代理模型和麻雀搜索算法(SSA)的大坝参数快速反演方法,并应用于坝基渗流参数的准实时反演,根据反演成果分析坝基局部测点水位偏高的成因,为大坝实时在线监控提供有效的技术支持。Based on variable space partitioning by dichotomy,the collection of samples and the training,testing and preservation of surrogates in subspaces of scales are conducted automatically according to the prediction errors of existing models and the expected standards,until all surrogates of subspaces reach the expected precision.Surrogates are obtained by support vector regression(SVR)modeling with particle swarm optimization(PSO)optimizing model parameters.The results of automatic modeling and testing for an explicit function show that,compared with the initial global prediction model,a multi-surrogate model based on space partitioning has significantly improvement in both global prediction precision and local error control,and the algorithm proposed is reasonable and effective.In addition,aiming at the near real-time inversion of seepage parameters of dam foundation,a fast inversion method is proposed associating PSO-SVR multi-surrogate modeling with sparrow search algorithm(SSA).According to the inversion results,the causes for high water level at measuring points in local area of dam foundation are analyzed.

关 键 词：代理模型 空间划分 二分法 PSO-SVR 坝基渗流 参数反演 SSA 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TV64[自动化与计算机技术—控制科学与工程]