基于多峰优化Kriging模型与距离相关系数的高速列车动力学参数多输出灵敏度分析  

Sensitivity Analysis for Dynamic Parameters of High-Speed Train Based on Multimodal-Optimization Improved Kriging Model and Distance Correlation

作　　者：姜杰[1] 杨旭锋 丁国富[1] JIANG Jie;YANG Xufeng;DING Guofu(School of Mechanical Eng.,Southwest Jiaotong Univ.,Chengdu 610031,China)

机构地区：[1]西南交通大学机械工程学院,四川成都610031

出　　处：《工程科学与技术》2024年第4期250-260,共11页Advanced Engineering Sciences

基　　金：国家重点研发计划项目(2020YFB1711400,2022YFB4301303);四川省科技厅重点研发项目(2021YFG0178)。

摘　　要：高速列车动力学参数较多,有多个动力学性能的评价指标。综合考虑这些动力学指标进行参数灵敏度分析,对高速列车动力学参数的恰当匹配具有重要意义。现有的单输出灵敏度分析技术难以准确度量参数对车辆动力学综合性能的影响。为避免这一缺陷,本文引入了一种新的基于距离相关系数的多输出灵敏度分析技术。该方法通过求解单个输入与多个输出变量之间的距离相关系数,实现多输出全局灵敏度分析的目的。为提高灵敏度分析的效率,建立了多个动力学指标与输入参数的Kriging模型;为提高近似模型的精度,引入了一种新颖的多峰优化算法对Kriging模型的超参数进行全局寻优。与传统单目标智能优化算法不同,该优化算法通过引入种群差异指标,将单目标优化问题转化为双目标优化问题,从而增加了子代种群的多样性,避免了传统优化算法易于陷入局部最优解的问题。基于所提方法,以和谐号动车组动车为例,研究了多个动力学指标的多输出全局灵敏度。结果表明:由于增强了种群多样性,所提的超参数优化方法对Kriging模型精度和稳定性提升效果明显;基于距离相关系数的灵敏度分析方法能够更加合理地识别出对车辆动力学综合性能影响较大的参数,排除几乎没有影响的参数。There are many parameters that affect the dynamic performance of high-speed trains,and there are many evaluation indexes that can be used to assess this performance.Comprehensive consideration of these dynamic indexes while conducting a sensitivity analysis of the parameters is a very significant approach for optimizing high-speed train performance.It is difficult to measure the sensitivity of vehicle dynamics to its parameters comprehensively and accurately using existing single-output sensitivity-analysis techniques.To avoid the limitations of existing methods,this paper introduces a multivariant sensitivity-analysis technique based on distance correlation.By solving the distance-correlation coefficient between a single input and multiple outputs,the proposed method achieves global sensitivity analysis for multiple outputs.To improve the efficiency of this sensitivity analysis,Kriging models between multiple dynamic indexes and input parameters are established.To improve the accuracy of the approximation model,a novel multimodal optimization algorithm is introduced to globally optimize the hyperparameters of these Kriging models.Contrasting with traditional single-objective intelligent optimization algorithms,the algorithm presented here transforms a single-objective optimization problem into a double-objective optimization problem by introducing the population-diversity index.As a result,the diversity of the offspring population is enhanced and the problem of getting stuck in local minima of the optimization space is avoided.Based on the proposed method,the CRH1 electric multiple unit is taken as an example to investigate the parameter sensitivity of multiple dynamic indexes.The results show that the proposed hyperparameter-optimization technique can significantly improve the accuracy and stability of the Kriging models because it enhances the population diversity.In addition,the described sensitivity-analysis method can more reasonably identify those parameters that have large and small impacts on the vehicle

关 键 词：演化的多模优化 KRIGING模型 多输出灵敏度分析 距离相关系数 

分 类 号：TB114.3[理学—概率论与数理统计]