【摘要】 针对确定随机需求车辆路径问题的最优策略,存在状态空间"维数灾"问题,基于增强学习函数近似原理,首先,设计了一个径向基函数(RBF),其次,在一给定的控制策略下,将最小平方瞬时差分(LSTD)法确定函数的权系数与交叉熵(CE)法确定隐层节点基函数的参数相结合,通过在线调整,使Bellman残差平方和性能指标达到最小,最后,根据得到的径向基函数,确定最优策略。通过仿真试验,验证了所设计方法的有效性。更多还原

【Abstract】 Due to the state space "dimension disaster" problem when determining an optimal policy for vehicle routing problem with stochastic demands, based on the function approximation principle of reinforcement learning, firstly, a radical basis function (RBF) was designed, secondly, for a fixed policy, the Bellman error as an optimization criterion was minimized by on-line tuning both least squares temporal difference (LSTD) algorithm for determining the function weight coefficients and cross entropy (CE) method for solving the basis function parameters, finally, the optimal policy was obtained by the RBF. Simulation results show the effectiveness of the proposed method for solving such problem.更多还原