基于Q学习的智能交通预测与多路径规划研究

发布时间：2018-05-10 05:43

本文选题：模糊神经网络 + 平均车速预测　；参考：《中南大学》2014年硕士论文

【摘要】：摘要：车辆路径规划技术是解决城市交通拥堵的有效手段之一。传统的路径规划算法,通常只给出最优路径,难以避免车辆所经过路段偶然瘫痪导致没有可选路径的问题。引入多路径规划技术,可保证车辆在任何情况下都有可选路径,提高路径规划稳定性。但是目前的多路径规划技术实时性不高,且算法效率较低,设计高效实时的多路径规划算法颇具挑战性。本文以提高多路径规划的实时性和稳定性为目标,对预测机制和多路径规划这两个关键技术进行研究。首先,为了给路径规划提供实时可靠的数据,本文拟采用模糊神经网络预测机制,精确预测下一时刻交通路网状况。由道路传感器收集平均车速数据,建立模糊神经网络模型预测未来车速,从而计算每条路段的未来平均通过时间,为路径规划提供未来的道路状况信息。模糊神经网络能真实反应交通信息的非线性特性,且预测精度很高。此外,本文拟引用Taguchi方法,在一定预测精度要求下,使用尽可能少的传感器数据,提高预测效率。其次,在预测数据基础上,为了提高多路径规划的算法效率和稳定性,提出了基于Q学习的多路径规划算法。根据Q学习思想对复杂城市路网进行建模。利用了Q值能反映当前路口距离目的地的长期反馈特性,推导出最优路径。然后确定合适的参考值,选取满足条件的次优Q值,实现多候选路径的选取。此外引入多路径集的稳定性约束,保证任何情况下至少存在一条候选路径可供车辆选择；引入协同机制,均衡未来路网负载。最后,利用MATLAB仿真工具分析了模糊神经网络预测机制的精度,基于增强学习的多候选路径算法的效率和稳定性,验证了所提的路径规划方案。图24幅,表8个,参考文献60篇。
[Abstract]:Abstract: vehicle path planning is one of the effective means to solve urban traffic congestion. The traditional path planning algorithm usually only gives the optimal path, and it is difficult to avoid the problem that there is no alternative path caused by the accidental paralysis of the road sections. The introduction of the multi-path planning technique ensures that the vehicle has an optional path in any case. The stability of path planning is improved. However, the current multipath planning technology is not very real-time, and the efficiency of the algorithm is low. It is very challenging to design the efficient and real-time multi-path planning algorithm. This paper aims at improving the real-time and stability of the multipath planning, and studies the two key technologies of the prediction mechanism and the path planning.
First, in order to provide real-time and reliable data for path planning, this paper uses the fuzzy neural network prediction mechanism to accurately predict the state of traffic network at the next moment. The road sensor collects the average speed data and establishes a fuzzy neural network model to predict the future speed, thus calculating the future average passing time of each section, as the path. The planning provides the future road information. Fuzzy neural network can truly reflect the nonlinear characteristics of traffic information, and the prediction accuracy is very high. In addition, this paper is proposed to use the Taguchi method to use as little sensor data as possible to improve the prediction efficiency under the requirements of certain prediction accuracy.
Secondly, on the basis of the prediction data, in order to improve the efficiency and stability of the algorithm for multipath planning, a multi-path planning algorithm based on Q learning is proposed. Based on the Q learning idea, the complex urban road network is modeled. The Q value can reflect the long-term feedback characteristics of the destination of the intersection distance, and the optimal path is derived. The suboptimal Q value satisfying the condition is selected to select the multi candidate path. In addition, the stability constraint of the multipath set is introduced to ensure that at least one candidate path is available for vehicle selection in any case, and the cooperative mechanism is introduced to balance the load of the road network.
Finally, the accuracy of the fuzzy neural network prediction mechanism is analyzed by using the MATLAB simulation tool. Based on the efficiency and stability of the enhanced learning multi candidate path algorithm, the proposed path planning scheme is verified. Figure 24, table 8, and 60 references.

【学位授予单位】：中南大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：U495;U492.22

【参考文献】