Q-Learning Algorithm for Fourth Party Logistics Route Optimization Considering Tardiness Risk

To solve the routing optimization problem in the fourth party logistics under the dynamic and complex environment which may lead to the delivery task not being completed on time, this paper introduces the value at risk (VaR) to measure the tardiness risk, and establishes the tardiness risk as the objective function and the delivery cost as the constraint condition. Mathematical model with the aim of providing customers with satisfactory delivery services at limited costs and with minimal risk of delays. Based on the nonlinear and NP-hard characteristics of the problem, the Q-learning algorithm is combined with the fourth party logistics routing optimization problem (4PLROP), and the reward value is redesigned and defined. Several different scales of numerical computations are performed, results of three algorithms are compared, and the experiment results show that the constructed random model can control the tardiness risk effectively and the presented algorithms can obtain satisfactory solutions quickly according to the customer's different confidence levels.