多车型回程车辆调度问题的ADP算法研究

采用近似动态规划(ADP)方法对钢铁物流运输过程中的车辆调度问题进行了分析, 设计了车辆和运输货物的状态向量空间、动作向量空间等, 充分考虑运输成本和能力约束, 建立状态转移函数、目标函数, 并对近似动态规划算法进行改进。在基于决策后状态的ADP算法的基础上, 采用Boltzmann探索策略对所有的状态空间进行遍历, 避免局部最优和低效问题。通过对比实验, 比较Q学习算法、基于决策后状态的ADP算法以及采用Boltzmann探索策略的ADP算法的实验结果, 证明了采用Boltzmann探索策略的ADP算法具有更快的收敛速度, 执行效率更高。

Approximate dynamic programming algorithm research on multi-types vehicle scheduling problem with backhauls

This paper analyzed the transportation scheduling problem with backhauls for iron and steel logistics industry using approximate dynamic programming, and designed the state vector space and action vector space of logistic vehicles and loads, and defined state transfer function and objective function in consideration of transportation cost and capacity constraint, and improved the approximate dynamic programming algorithm. On the basis of the ADP algorithm based on the post-decision state, it searched all the state space using Boltzmann explore strategy and avoid the local optimum and inefficient problem. Through the experiments, it analyzed and compared the test result of the Q-Learning algorithm, the normal ADP algorithm based on post-decision state and the improved ADP algorithm based on Boltzmann explore strategy, and prove that the improved ADP algorithm using Boltzmann explore strategy has faster convergence speed and higher running efficiency.