Research On Intelligent Traffic Control Methods Based On Deep Reinforcement Learning

Since the 21 st century,with the increasing population,traffic congestion has become a common and urgent problem in major cities,improving and optimizing traffic control method is one of the effective ways to solve the traffic congestion problem.With the continuous development of science and technology and the deepening of research in the field of machine learning,new research directions such as deep learning and intensive learning are constantly being applied to solve various practical problems,and the research of urban intelligent traffic control has moved towards a new stage of development.It is of great significance and research value to design and develop more efficient intelligent traffic control systems based on deep learning and enhanced learning.The research content of this paper is as follows:First of all,traffic flow,as an important part of traffic control system,can provide a reference basis for intelligent traffic control,this paper proposes an ALO-LSTM-based traffic flow forecasting algorithm.The urban traffic flow data has certain characteristic laws in the timing,so this paper uses the LSTM neural network model to establish the traffic fl ow prediction model.In the process of establishing the prediction model,the randomness of the initialization of neural network parameters leads to the poor training effect of the prediction model,and the initial parameters of the LSTM model are optimize d by using the ant-lion optimization algorithm ALO,which has good global search and optimization ability,and can speed up the time of prediction model convergence.Improve the performance of the LSTM predictive model.The traffic flow is predicted by Using LSTM and ALO-LSTM algorithms and the effectiveness of the algorithm is analyzed.Secondly,a multi-junction traffic signal control method based on collaborative deep Q learning is proposed for the state space explosion and multi-intersection collaboration in the complex road network.By establishing the urban traffic signal control model,using deepreinforcement learning to explore the internal information in traffic state,learning from the high-dimensional data to learn the effective control strategy,taking into account the traffic state data of the intersection itself and the influence of the adjacent intersection,the neural network training can obtain the optimal execution phase of the current intersection signal lamp.At the same time,the experience playback mechanism and the target network strategy are used in the learning process to improve the stability of the learning algorithm.Under the GLD simulation platform,the validity of the algorithm is verified by experiments in the simulated road network environment of different congestion levels.Finally,in view of the problem that deep reinforcement learning is very time-consuming due to the need to try and error constantly in the training process,heuristic Q learning is combined with deep reinforcement learning,and a multi-junction traffic signal control method based on heuristic deep Q learning is proposed.Combining the traffic flow data in the urban road network environment to determine the heuristic function,the inspiration function is used to guide the exploration of state action space and accelerate the learning rate of the algorithm in the action selection stage of deep reinforcement learning.The validity of the algorithm was verified by experiments in the simulated road network enviro nment of different congestion levels.