基于多元特征参数与改进SVM算法的驾驶风格识别研究

Research on driving style recognition based on multivariate feature parameters and an improved SVM algorithm

当自动驾驶车辆在执行换道等行为时,需要准确识别人类驾驶车辆的驾驶风格类型,以保证换道等行为的安全进行。为此,提出了一种基于多元特征参数与优化支持向量机(SVM)相结合的驾驶员驾驶风格识别模型。采用改进粒子群算法(IPSO)对支持向量机模型参数进行优化,搭建IPSO-SVM驾驶风格识别模型,并用UCI数据库中的数据集对其进行验证,结果表明:IPSO-SVM模型在准确性、实时性及收敛性方面均优于CV-SVM模型和PSO-SVM模型。在此基础上,进一步采用NGSIM数据库中真实交通流的数据进行驾驶风格识别测试。首先处理多元特征数据,滤除异常值;其次采用主成分分析法对数据进行降维和简化,并用K-means算法对其进行聚类;最后将降维简化的数据作为输入,聚类得到结果作为输出,采用IPSO-SVM识别模型进行仿真实验。结果表明:提出的IPSO-SVM模型准确率可达97.96%,均方误差降低约84%,绝对误差降低约81%,运行时间平均减小30%,且ROC曲线的AUC值最大模型性能最优,仿真结果验证了该模型对驾驶风格有更好的识别效果,其具有一定的可行性。

Driving style refers to the different driving characteristics of drivers in different driving environments.The acceleration of vehicles with an aggressive driving style changes sharply,and the pursuit of speed gains is higher;while vehicles with a conservative driving style run more smoothly,and the pursuit of safety gains is higher.In a word,different driving styles have a great impact on vehicle lane changing,so it is necessary to accurately identify driving styles.When automatic driving vehicles perform lane changing and other operations,it is necessary to accurately identify the driving style of the surrounding vehicles on road to ensure the safety of lane changing and other operations.Traditional recognition of driving style tends to use data obtained from questionnaires or collected by on-board sensors.This paper proposes a driving style recognition model based on the combination of multivariate feature parameters and optimized Support Vector Machine(SVM).SVM has a simple structure,a strong learning ability and versatility,which can effectively avoid“dimensional disasters”,but there are still some problems,such as easiness to fall into local optimization.However,the Improved Particle Swarm Optimization(IPSO)algorithm overcomes this shortcoming,making the selection of relevant parameters of SVM algorithm more accurate,and the recognition accuracy of driving style higher.First of all,this paper uses IPSO to optimize the parameters of the SVM model,builds IPSO-SVM driving style recognition model,and verifies the model with the data set in the UCI public database.The results show that IPSO-SVM model is superior to CV-SVM model and PSO-SVM model in terms of accuracy,real-time or convergence.On this basis,the data of real traffic flow in NGSIM database is further used for driving style recognition test.Firstly,the multivariate feature data are processed to filter out the outliers;secondly,principal component analysis is used to reduce the dimension and simplify the data,and K-means algorithm is used to cluster them;finally,the reduced dimension and simplified data are used as input,and the clustering results are used as output.IPSO-SVM recognition model is used for simulation experiments.The results show that,compared with CV-SVM model and PSO-SVM model,IPSO-SVM model proposed in this paper overcomes the disadvantage of easily falling into the local optimum,effectively improves the identification accuracy,reduces the errors,and speeds up the convergence speed.Its accuracy rate can reach 97.96%,the mean square error decreases about 84%,the average absolute error decreases about 81%,and the running time decreases about 30%.At the same time,the AUC value of the ROC curve further proves the performance and reliability of the model.The simulation results show that the model has better recognition effects on driving style,which is a feasible solution.