Research On Virtual Test And Evaluation Method Of Intelligent Vehicle Based On Scenarios

In recent years,intelligent vehicles have been developing rapidly and the whole automotive industry is in a critical period of transformation from traditional vehicles to intelligent vehicles.However,with the popularity of intelligent vehicles,some issues and challenges have emerged,so before intelligent vehicles are officially used,they must undergo strict testing and verification to give a quantitative evaluation of their many properties.Aiming at the problems which there is no systematic testing and evaluation system for intelligent vehicles,and some foreign testing methods and standards are not fully compatible with the actual traffic scenarios in China.Supported by the project of National Key R&D Program "Holographic Traffic State Reconstruction and Vehicle Group Cooperative Control Testing and Verification"(2018YFB1600605),this paper discusses and researches the scenario-based intelligent vehicle testing and evaluation,and the main research contents are as follows.(1)A typical test scenario was obtained by clustering using the improved system clustering algorithm.Aiming at the problems of multicollinearity in the multidimensional indicators of the classical systematic clustering algorithm and fail to consider the variability among the factors in clustering,the classical systematic clustering algorithm was improved and optimized by using principal component analysis,information entropy assignment and determination of sample continuity.Typical elements extracted from natural driving data were used as clustering objects,and clustering was performed with two clustering algorithms respectively,The clustering results are evaluated by using the clustering effectiveness indexes DB(DaviesBouldin index)and CH(Calinski-Harabaz index).The results show that the DB value of the improved clustering algorithm is 1.0627,The CH value is 1.2166,the DB value of the classical hierarchical clustering algorithm is 1.3659,the CH value is 1.0361,which shows that the improved clustering algorithm has improved the clustering accuracy.,and four types of typical test scenarios were obtained to build the scenario basis for the subsequent virtual test.(2)Pre Scan virtual simulation platform and Simulink were used to build a virtual test environment and conduct virtual tests.Aiming at the problems of the classical TTC algorithm such as fixed time threshold and easy to be false triggered due to the fluctuation of the input signal,the method of using neural network to predict the braking distance,adding filters to smooth the signal and setting three detection modules was improved and optimized.The three aspects of safety,ride experience and efficiency and energy consumption were selected as the evaluation index parameters,and the vehicles to be tested with two different design solutions were tested under four typical scenarios,and the relevant index parameters were output.(3)A synthetic fuzzy evaluation model for intelligent vehicles is constructed,and the expandable hierarchical analysis method is used to assign weights to each evaluation index.It is verified that the optimized TTC algorithm improves safety,ride experience and efficiency and energy consumption than the classical TTC algorithm.