基于GA-PSO算法的冻土本构模型参数识别

Parameter identification of constitutive model for frozen soilbased on GA-PSO algorithm

遗传算法(GA)与粒子群算法(PSO)分别具有缺乏目标导向性和易陷入局部最优的缺点,但同时分别具有全局搜索能力强与能有效传递优势信息的优点。本文以GA计算步结合精英保留策略作为PSO计算步的优势信息,避免PSO算法陷入局部最优,以PSO计算步结合非精英优化策略作为GA计算步的导向信息,克服GA算法缺乏目标导向的问题,建立了GA-PSO新算法。其具体过程为,通过采用GA计算步对解空间进行全局搜索并对精英个体进行保留,进一步,将适应度较差的个体利用PSO计算步进行优化。基于多峰函数的验证结果表明,GA-PSO算法在解空间中具有更强的全局搜索能力,同时具有更快的收敛速度。将GA-PSO算法应用到冻土非正交弹塑性本构模型的参数识别中,通过模型的参数识别以及模型预测结果对比与验证,结果表明GA-PSO算法能够有效识别冻土非正交弹塑性本构模型的参数,提升了模型的预测效果。

Given the complexity of frozen soil mechanics,establishing constitutive models that reasonably de-scribe its mechanical behavior inevitably requires increasing the number of model parameters.The predictive ac-curacy of these models largely depends on the rational determination of these parameters.However,some pa-rameters in frozen soil constitutive models cannot be directly determined through experiments or empirical equa-tions.Therefore,parameter identification of frozen soil constitutive models based on limited experimental data holds significant engineering significance.Optimization algorithms provide effective tools for parameter identifi-cation in various engineering fields,and their application in geotechnical engineering has become increasingly widespread.Among them,genetic algorithm(GA)and particle swarm optimization(PSO)algorithm are two popular optimization algorithms.However,both algorithms have their own advantages and disadvantages.GA lacks target orientation,but possesses strong global search capability,while PSO is prone to local optima,but efficient information transmission.Therefore,this paper proposes a novel hybrid algorithm,the GA-PSO algo-rithm,which combines the strengths of GA and PSO while mitigating their respective weaknesses.In the GA-PSO algorithm,the incorporation of the elite preservation strategy within the GA computation step serves as ad-vantageous information for the PSO computation step,preventing PSO from getting stuck in local optima.Con-versely,the incorporation of the non-elite optimization strategy into the PSO computation step provides guiding information for the GA computation step to address the issue of lacking target orientation in the GA algorithm.The specific process involves global exploration of the solution space using GA calculation step while preserving elite individuals,followed by further optimization of poor fit individuals using PSO calculation steps.Validation results based on two standard test functions,i.e.,Griewank function and Restrigin function,illustrate that the GA-PSO algorithm exhibits superior global search capability and faster convergence speed in the solution space.Furthermore,The GA-PSO algorithm is applied to the parameter identification of the non-orthogonal elastoplas-tic constitutive model for frozen soil.The results of model parameter identification,as well as the comparison and validation of model prediction with test results,indicate that the GA-PSO algorithm is proficient in effective-ly identifying parameters of the non-orthogonal elastoplastic constitutive model for frozen soil,thereby enhanc-ing the predictive accuracy of the model.