Research On The Driving Decision-making Mechanism Of Autonomous Vehicle Based On Particle Swarm Optimized SVM Model

In order to reduce the incidence of traffic accidents and collision casualties,autonomous vehicles have become the focus of research in the field of traffic in the world,and driving decision-making mechanism is the core key technology to ensure the safe driving of autonomous vehicles.Therefore,this paper analyzes and studies the driving decision-making regulation of autonomous vehicles under different working conditions,and establishes the decision-making mechanism based on particle swarm optimized support vector machine model(Support Vector Machine,SVM).The main work is as follows:(1)For the normal driving conditions,in order to improve the adaptability of autonomous vehicle to the complex road environment of the city,in this paper,the road conditions are integrated into the influence index set of the driving decision,and used as the decision-making reference parameter of the autonomous vehicle with the vehicle operation status.Secondly,the sample data of normal driving decision are obtained by driving simulation experiment.Based on the PSO-SVM algorithm,the influence index set processed by data fusion is taken as the input parameter of the algorithm.And the corresponding lane change,car-following and tree driving decision-making are taken as the output parameters,and the SVM normal driving decision-making mechanism model is obtained.Finally,by verifying the sensitivity of the model to road conditions,the influence of road conditions on driving decision is quantitatively analyzed.The results show that the reasoning performance of the model has been greatly improved after considering the road conditions,and under the condition of low traffic flow density,road conditions have the greatest impact on driving decision-making,in which the road visibility has the greatest impact,followed by adhesion coefficient,road curvature and road slope,but at high traffic density,they have little impact on driving decision-making.(2)For the emergency conditions,this paper adopts the research main line of "scene analysis-collision severity prediction-optimal collision decision output".With the traffic accident data,the impact index of collision severity is taken as the input parameter of the PSO-SVM algorithm,and the collision severity under each emergency decision is taken as the output parameter.The prediction models of collision severity under braking,steering and braking+steering are established respectively,which can be used as the trade-off basis for autonomous vehicles to make decisions in emergency situations.Then,based on the same accident sample,the collision severity results of the three prediction models are compared and analyzed.the results show that the minimum collision severity results can be achieved only by braking decision in the low range of relative speed.With the increase of relative speed,the benefit of steering decision in reducing collision severity is becoming more and more prominent.When the relative speed reaches a high range,it is necessary to take braking and steering decisions at the same time in order to reduce the collision severity.(3)In the training process of the above SVM model,in order to enable SVM to independently select the form of kernel function that is most suitable for the driving decision mechanism according to the sample characteristics,a weighted mixed kernel function is proposed as the kernel function of the above SVM.Then,the prediction accuracy of SVM model with weighted mixed kernel function,SVM model with RBF kernel function and BP neural network model are compared and analyzed.The results show that,the SVM model with weighted mixed kernel function has the best classification performance than the other two models.