Research On Intelligent Control Based On Deep Learning

The importance of autonomous driving system for intelligent transportation is self-evident.Local path planning and vehicle control algorithm are the key algorithms for autonomous driving.The final calculation results will affect the safety and efficiency of autonomous driving.Now most of the autonomous driving system using the traditional local path planning,vehicle control algorithm: local path planning algorithm to calculate the results of environmental perception are vehicles expected trajectory,vehicle control algorithm according to the trajectory tracking control signals are given to the signal,its safety and efficiency depends on environmental awareness and the result of local path planning.This paper mainly studies the depth of the deep learning combined with reinforcement learning to strengthen learning theory,and use this theory to environmental awareness,local path planning,vehicle control integration as a kind of intelligent control technology.The intelligent control technology to the dynamic simulation environment driving environment to safety analysis and according to the vehicles running near the obstacle information as well as the current task real-time traffic control signal is given.The driving tasks include road maintenance,parallel lines,overtaking,and roadside parking.Different simulation environments are designed for each scenario to provide a deep reinforcement learning network model to complete the corresponding training.In the prediction process,the deep reinforcement learning network model will seamlessly switch the different driving tasks.The main contents of this paper are:1.Introduce the traditional local path planning algorithm,vehicle control algorithm and their research status of some vehicles.This paper introduces the principle of the common local path planning algorithm,such as grid method,artificial potential field method and the principle of common PID control algorithm.On the basis of summarizing the advantages and disadvantages of the algorithm,the security of the algorithm is analyzed.2.Detailed introduction to the deep convolutional neural network.This paper introduces the structure and principle of deep convolutional neural network including traditional inverse propagation neural network,convolutional network and pooling operation.To summarize the advantages of deep convolutional neural network in solving environmental perception problems.3.Detailed introduction of value-based,policy-based related part deep reinforcement learning algorithm.Q value network is introduced,including depth,depth,the depth of the deterministic policy gradient algorithms,such as its principle and advantages and disadvantages of reinforcement learning algorithm,this paper use the depth of intensive study to complete the local path planning and the feasibility of vehicle control function.4.Due to the high cost of training and verification algorithm in the actual scene,this paper first constructs a simulation driving environment.The algorithm is trained and verified in the simulation environment to ensure the algorithm is feasible.Through simulation environment verification,this paper proposes a method to complete the intelligent control problem of vehicles in different driving tasks and effectively improve the efficiency and safety of vehicle autonomous driving.