Research on Vehicle GPS Dynamic Sampling Based on Deep Reinforcement Learning

This article addresses the issues of considering the long-term impact of a single sampling decision, the difficulty in determining the relationship between vehicle state movement characteristics and trajectory accuracy, and the difficulty in balancing the number of sampling points in dynamic adjustment of GPS sampling rate. A deep reinforcement learning method is designed to learn the GPS sampling strategy network, define the problem of vehicle GPS trajectory data, establish discrete rules for vehicle movement state generation, establish a model, and train using publicly available datasets and network crawled data. Through experiments, the GPS dynamic sampling results are compared to estimate the weight of vehicle movement state characteristics, and generate a reinforcement learning model reward function based on this. The experimental results show that using this method to sample trajectories can effectively reduce the number of GPS points, while the accuracy of vehicle GPS trajectories decreases slightly.