长距离输水管道检修通风SRM-CoRBF多保真代理模型研究  

SRM-CoRBF Multi-Fidelity Surrogate Model for Maintenance Ventilation of Long-Distance Water Transmission Pipeline

作　　者：王晓玲[1] 丁超元 刘长欣[1] 刘来全 龚芳 余佳[1] Wang Xiaoling;Ding Chaoyuan;Liu Changxin;Liu Laiquan;Gong Fang;Yu Jia(State Key Laboratory of Hydraulic Engineering Simulation and Safety,Tianjin University,Tianjin 300072,China;Guangdong GDH Pearl River Delta Water Supply Co.,Ltd.,Guangzhou 511458,China;Guangdong Hydropower Planning&Design Institute Co.,Ltd.,Guangzhou 510635,China)

机构地区：[1]天津大学水利工程仿真与安全国家重点实验室,天津300072 [2]广东粤海珠三角供水有限公司,广州511458 [3]广东省水利电力勘测设计研究院有限公司,广州510635

出　　处：《天津大学学报（自然科学与工程技术版）》2022年第11期1182-1194,共13页Journal of Tianjin University：Science and Technology

基　　金：天津市自然科学基金资助项目(19JCYBJC22600)。

摘　　要：检修通风对于保障长距离输水管道检修安全至关重要.目前基于计算流体力学(computational fluid dynamoics,CFD)模型及高保真度代理模型的通风效果研究方法存在计算效率低、建模成本高的不足.通过融合高/低保真度分析模型的数据,多保真度代理模型能够兼顾预测精度与建模成本,然而其通过最小化预测误差获取模型超参数,忽略了结构复杂性对模型泛化性能的影响.针对上述问题,本文建立长距离输水管道检修通风改进协同径向基函数多保真度代理模型.该模型利用结构风险最小化准则(structural risk minimization,SRM)能够同时考虑经验风险和置信范围最小化的优势,采用结构风险L2范数描述协同径向基函数(cooperative radial basis function,CoRBF)的结构复杂性,推导其与模型超参数之间的微分关系对径向基函数形状因子进行优化,进而建立SRM改进的协同径向基函数(SRM-CoRBF)多保真度代理模型,并采用测试函数验证了改进模型的优越性.案例分析表明,SRMCoRBF多保真度代理模型能够实现长距离输水管道检修通风效果的高精度预测.本文所提模型一方面与径向基函数(radial basis function,RBF)单保真度代理模型相比,在检修通风风量供需比、换气次数以及通风成本方面的预测精度分别提高15.02%、9.29%、6.03%;另一方面,与传统CoRBF多保真度代理模型相比,SRM-CoRBF在保证模型预测精度的同时能够有效提高泛化性能.本文所提模型为长距离输水管道检修通风效果分析提供了一种新思路.Maintenance ventilation is essential to ensure the maintenance safety of long-distance water transmission pipelines.At present,the ventilation effect research methods based on the computational fluid dynamics(CFD)model and high-fidelity surrogate model have disadvantages of low computational efficiency and high modeling cost.The multi-fidelity surrogate model can take into account the prediction accuracy and modeling cost by integrating the data of high-and low-fidelity analysis models.However,it obtains model hyperparameters by minimizing the prediction error and ignores the influence of structural complexity on the generalization performance of the model.Given the above problems,this paper establishes an improved cooperative radial basis function(CoRBF)multi-fidelity surro-gate model for long-distance water transmission pipeline maintenance ventilation,which uses the structural risk minimization(SRM)criterion to consider the advantages of empirical risk and confidence range minimization simulta-neously.The L2 norm of structural risk is used to describe the structural complexity of the cooperative radial basis function(CoRBF),and the differential relationship between CoRBF and model hyperparameters is derived to opti-mize the shape factor of the radial basis function.The multi-fidelity surrogate model of SRM-CoRBF is then estab-lished,and the test function is used to verify the superiority of the improved model.The case analysis shows that the SRM-CoRBF multi-fidelity proxy model can realize the high-precision prediction for the ventilation effect of long-distance water transmission pipeline maintenance ventilation.Compared with the RBF single-fidelity surrogate model,the prediction accuracy of the proposed model in the supply-demand ratio of the air volume,the number of ventilation changes,and the ventilation cost is increased by 15.02%,9.29%,and 6.03%,respectively.Mean-while,compared with the traditional CoRBF multi-fidelity surrogate model,SRM-CoRBF can effectively improve the generalization performance while

关 键 词：长距离输水管道 检修通风效果 SRM-CoRBF多保真度代理模型 结构风险最小化 协同径向基函数 

分 类 号：X820.2[环境科学与工程—环境工程]