Research On Reinforcement Learning Adaptive Cruise Control Strategy Based On Driving Style Recognition

In recent years,with the increase of car ownership,frequent driving accidents and high urban road congestion rate have become an urgent problem in modern transportation.Therefore,adaptive cruise control technology can be introduced to shorten the following distance while ensuring driving safety to improve the traffic efficiency of urban roads.In addition,leading vehicle driving style has an important impact on vehicle driving decision-making.It is positive to optimize adaptive cruise control by adjusting the self-driving speed and vehicle spacing according to the driving style of the front vehicle and preventing road unsafe factors in advance.Aiming at ensuring driving safety and improving road traffic efficiency,and taking vehicle longitudinal driving as the scene,an adaptive cruise control strategy based on leading vehicle driving style recognition is proposed in this paper.The main research contents are as follows:Firstly,a driving style recognition method based on semi-supervised gaussian mixture model is proposed to identify the driving style of leading vehicles accurately and solve the strong randomness of driving style and few driving data labels.The driving data of the leading vehicle is collected and processed.The speed,acceleration and impact are selected to construct the multi-dimensional driving style characteristic parameters.And the kernel principal component analysis method is used to reduce the dimension of the characteristic parameters.A driving style recognition model based on semi-supervised gaussian mixture clustering is established,and the expectation maximization algorithm is used to solve the model parameters.The proposed method carries out auxiliary training on the driving style classification model through less label sample information,which reduces the dependence on the training data label,and improves the recognition accuracy of the driving style of the leading vehicle.Secondly,aiming at the problem that adaptive cruise control is difficult to achieve optimal decision in complex and changeable traffic environment,an adaptive cruise control optimization strategy based on deep deterministic policy gradient is proposed.The variable spacing model is designed based on the driving style of the leading vehicle,and the multi-objective optimization problem of adaptive cruise control for driving safety,speed consistency and riding comfort is constructed.The multi-objective optimization problem is transformed into a markov decision process,which is solved based on deep deterministic policy gradient algorithm.The next generation traffic simulation data set is used to train the deep deterministic policy gradient network,and the adaptive cruise control strategy based on the driving style of the front vehicle is obtained,which improves the driving safety and riding comfort of the adaptive cruise control.Finally,a Python simulation environment is built to identify the driving style of the leading vehicle,and the constant speed cruise,following cruise,side vehicle entry and front vehicle departure of autonomous vehicle are simulated to verify the feasibility and effectiveness of the proposed method.