基于高阶响应面和NSGA-Ⅲ的卷钢托架多目标优化  

Multi-objective optimization of coil steel pallet based on high-order response surface and NSGA-Ⅲ

作　　者：潘帅 袁舜 周宝宪 王振东 闫帅帅 PAN Shuai;YUAN Shun;ZHOU Baoxian;WANG Zhendong;YAN Shuaishuai(School of Traffic and Transportation,Beijing Jiaotong University,Beijing 100044,China;Department of Safety Technology,Sinotrans Ningxia Co.,Ltd.,Yinchuan 750010,China;Yinchuan Railway Logistics Center,China Railway Lanzhou Group Co.,Ltd.,Yinchuan 750021,China;China Academy of Railway Sciences Corporation Limited,Beijing 100081,China)

机构地区：[1]北京交通大学交通运输学院,北京100044 [2]中国外运宁夏有限公司安全技术部,宁夏银川750010 [3]中国铁路兰州局集团有限公司银川铁路物流中心,宁夏银川750021 [4]中国铁道科学研究院集团有限公司,北京100081

出　　处：《铁道科学与工程学报》2025年第3期1266-1278,共13页Journal of Railway Science and Engineering

基　　金：国家自然科学基金资助项目(71761023);中国铁路兰州局集团有限公司科技开发计划(LZJKY2021013-2,2022060-2)。

摘　　要：为将托架优化问题中复杂的非线性隐式关系转化为显式函数,提出基于白箱系统理论的高阶响应面模型进行多目标优化,大幅降低了技术门槛,简单实用、精准度高,更容易被现场理解、接受。制定集重装车方案,进行静载、冲撞仿真和强度试验,检验仿真可靠性。采用基于AE势能准则和自适应遗传算法组合优化的最优拉丁超立方取样方法,创建高维参数空间矩阵。利用最小二乘回归法创建含交叉项的4阶响应面模型,并通过k折交叉验证评估不同代理模型精度。采用Sobol法量化各参数对不同响应的1阶、2阶、全局灵敏度及交互效应,深入探索各参数及其2阶交叉项对响应的影响。搭载NSGA-Ⅲ获得Pareto解集。结合我国铁路卷钢运输特点,运用秩和比法赋予各响应权重矩阵,获得不同设计偏好下的最优解。研究结果表明:仿真与实测值误差低于10%,仿真准确可靠。最优拉丁环境下,系统最小势能是0.41,相较于普通拉丁降低了22.6%,样本点分布合理。高阶响应面误差低于4.5%,精度满足工程需要。从各权重矩阵中,选择方案4作为最优设计,与原方案相比,托架质量降低了9.89%,2个工况最大应力分别降低了6.79%、14.36%,托架符合铁路货运安全需求。研究结果有望为同类托架设计、改进等领域提供关键技术支撑和工程参考。To transform the complex nonlinear implicit relationships in pallet optimization problems into explicit functions,a high-order response surface model based on white-box system theory was proposed for multiobjective optimization.This approach significantly lowers the technical threshold,was simple and practical,highly accurate,and more easily understood and accepted in the field.Devising heavy loading scenarios,conducting static loading simulation,impact simulation,and strength tests to verify the reliability of simulation.Proposing an optimal Latin hypercube sampling method based on the Audze-Eglais criterion and adaptive genetic algorithm combination optimization to create a high-dimensional parameter space matrix.Using the method of least squares regression,create a fourth-order response surface surrogate model with cross-terms,and assess the accuracy of different surrogate models through k-fold cross validation.Utilize the Sobol method to quantify the first-order,second-order,and global sensitivities as well as interaction effects of each input on different responses,exploring the comprehensive impact of each input on the responses.Employing NSGA-Ⅲ to obtain the Pareto solution set,and considering the characteristics of railway coil steel transportation,utilize the rank-sum ratio method to assign weights to each input matrix,obtaining the optimal solution under different design preferences.The results show that the error between the simulation and measured values is less than 10%,indicating that the simulations are accurate and reliable.Under optimal Latin hypercube sampling,the system's minimum potential energy is 0.41,which is 22.6% lower than that of the ordinary Latin,and the sample point distribution is reasonable.The error of the high-order response surface is less than 4.5%,meeting the precision requirements of engineering.Among the weight matrices,scheme 4 is selected as the optimal design.Compared to the original scheme,the mass of the pallet is reduced by 9.89%,and the maximum stress under two cond

关 键 词：卷钢托架 高阶响应面 强度试验 k折交叉验证 Sobol法 NSGA-Ⅲ 

分 类 号：U294.3[交通运输工程—交通运输规划与管理]