Q Learning Based Prediction And Multiple Paths Planning In Intelligent Transportation Systems

Abstract:Path planning for vehicles is an effective means to solve urban traffic congestion. The traditional path planning algorithms usually give only the optimal path. Once the optimal path is invalid, no alternative paths exist for vehicles. In order to improve the stability of path planning, multiple paths planning technology is introduced to ensure that at least one optional path exists for vehicles in any case. However, the real-time capability and efficiency of pre-existing multiple paths planning technology is low, it is challenging to design real-time efficient multiple paths planning algorithm. In this paper, prediction mechanism and multiple paths planning algorithm are researched to improve the real-time capability and efficiency of multiple paths planning.Firstly, in order to provide reliable real-time data for path planning, this paper uses fuzzy neural network prediction mechanism to accurately predict the future road network traffic conditions. The future average speed is predicted using fuzzy neural network prediction model based on data collected by the road sensor. The future mean travel time of each road is deduced by these prediction data to provide road conditions for the future path planning. Fuzzy neural network can describe nonlinear characteristics of true traffic information with high prediction accuracy. In addition, Taguchi method is introduced to improve efficiency of the prediction algorithm using as little sensor data as possible under given prediction accuracy.Secondly, the Q-learning based multiple paths planning algorithm is proposed to improve the efficiency and stability of path planning algorithms on the basis of forecast data. A new model of complex urban road network is designed based on Q-learning. The Q value reflects the current long feedback from the destination intersection. This characteristic is used to derive the optimal path. Then a suitable reference value is deduced to select suboptimal Q value and look for multiple paths. And the stability constraint of multiple paths set is introduced to ensure at least one available path exists in any case. Furthermore, cooperation mechanism is introduced to balance the future burden of road networks.Finally, the accuracy of the fuzzy neural network prediction mechanism is analyzed, and the efficiency and stability of the proposed Q-learning based multiple paths planning algorithm are valuated using MATLAB simulation tool.24figures,8tables and60references are included in this paper.