Optimization Algorithm of Logistics Distribution Path Based on Deep Learning

The path planning problem is a very classic problem, which is widely used in many fields. In recent years, solving the path optimization problem through deep reinforcement learning technology has attracted the attention of scholars, and has become a hot direction of the path planning problem. The purpose of this paper is to study the logistics distribution route optimization algorithm, solve the continuous route planning problem with deep reinforcement learning technology, and conduct experiments in the Miniworld maze environment. A DDPG algorithm based on reward shaping is proposed, which dynamically optimizes the reward function by representing the reward function with a neural network. Through the comparative analysis of the DDPG algorithm, the genetic algorithm and the historical average method, the results show that the historical average method is simple and easy to calculate, the accuracy rate does not change much with the increase of data, and the accuracy rate is small and not ideal. The accuracy of the genetic algorithm decreases as the training size increases, and the accuracy is around 70%. With the increase of the training system, the training level of the deep learning model is higher, and the prediction accuracy rate is also improved accordingly, and finally stabilized at about 83%.