Research On The Calibration Method Of Non-Motorized Social Force Model Parameters

Social force models have been widely used in the field of non-motorized traffic simulation,and their parameter calibration methods are crucial to the accuracy of simulation results.To address the lack of sensitivity analysis in determining the parameters to be calibrated and the calibration indexes in existing studies,as well as the problem of focusing on a single microscopic calibration index and ignoring macroscopic data validation,this study proposes a parameter calibration method that integrates microscopic and macroscopic calibration indexes.The simulation model is implemented in Python language and combined with NSGA-Ⅱ optimization algorithm to realize automated parameter calibration in order to improve the simulation fidelity of the model at micro and macro levels.The details of the study are as follows:(1)Analysis and determination of the parameters to be calibrated.A more general nonmotorized social force model is proposed through an in-depth analysis of the improvement principles of existing non-motorized social force models in terms of individual attributes,basic motion situations,self-organization,and simulation scenarios.According to the structural classification of perception layer,decision layer and execution layer,the parameters to be calibrated in each layer are determined,and these parameters are divided into observable parameters and parameters that need to be optimized by intelligent algorithms.(2)Calibration index selection and sensitivity analysis.By analyzing the limitations of existing calibration metrics,four microscopic calibration metrics based on trajectories are proposed,including absolute position difference,stepwise position difference,relative position difference,and path length difference,and two macroscopic calibration metrics based on traffic flow,flow-density basic map and overtaking rate.Information entropy and genetic algorithm are used to analyze the parameter sensitivity of these microscopic and macroscopic calibration indicators to verify the reasonableness of these indicators,and the results show that the parameter sensitivity of stepwise position difference and flow-density basic map is high.(3)Parameter calibration method design.The process of parameter calibration is optimized,and a parameter calibration method that integrates microscopic and macroscopic calibration indicators is proposed.A multi-objective optimization function is constructed,and a generalized non-motorized social force simulation model is successfully implemented using Python language,which is combined with NSGA-Ⅱ optimization algorithm to realize automated parameter calibration.(4)Example analysis and validation of the calibration method.Taking the intersection non-motorized vehicle crossing data as an example,all the parameters to be calibrated in the non-motorized vehicle social force model were parametrically calibrated,and then the performance of the model after parameter calibration was comprehensively evaluated in three ways: trajectory distribution validation,behavior validation,and basic graph validation.By comparing and analyzing the parameter calibration method with a single calibration index and the integrated microscopic and macroscopic calibration index proposed in this paper,it is found that the integrated method proposed in this paper achieves more satisfactory results at both microscopic and macroscopic levels.