Q Value-Based Dynamic Programming with Boltzmann Distribution by Using Neural Network
In this paper, a feedback method using neural network is proposed with Q Value-based Dynamic Programming based on Boltzmann Distribution for static road network. The neural network can supply more distribute strategies and the feedback method chooses the best result from the strategies produced by neural network. The method distributes vehicles well on all the optimal routes from the origin to destination according to the gradual decreasing parameters, which are used in the neural network. This method can overcome local optimum problems to some extent by setting appropriate parameters at the beginning. The proposed method is evaluated by using the Kitakyushu city (Fukuoka, Japan) road network data. The simulation result shows that the better result can be obtained than conventional QDPBD method by training parameters.
KeywordsQ value Boltzmann Distribution Optimal routes Feedback Neural network Vehicle distribution
This research was supported by A Project【16ZA0131 】which supported by Scientific Research Fund of Sichuan Provincial Education Department,【2018GZ0517】which supported by Sichuan Provincial Science and Technology Department, 【2018KF003】 Supported by State Key Laboratory of ASIC & System, Science and Technology Planning Project of Guangdong Province 【2017B010110007】, the National Natural Science Foundation of China grants【61672438】.