UAV Group Distribution Route Optimization Under Time-Varying Weather Network

Abstract

The rapid advancement in unmanned aerial vehicle (UAV) technology has marked a transformative shift in various industries, with logistics distribution service being one of the prime sectors reaping the benefits. UAVs offer substantial benefits in speed, cost, and reach, promising to revolutionize logistics, especially in remote areas. On the one hand, they are poised to meet demands for quick and versatile delivery options. On the other hand, their deployment comes with challenges. Weather variabilities such as rainfall, wind speed, and the need for safe take-off intervals can compromise UAV safety and operation. Conventional route optimization often overlooks these dynamic factors, resulting in inefficient or unworkable delivery routes. The repeated time-consuming calculations are caused by repeated trials when making UAV group distribution plans. Recognizing these gaps, this study proposes a data representation to effectively transform the flight flyable area of UAVs into a time-varying network that maintains spatiotemporal connectivity and establishes a mathematical model that represents the complexities of UAV group distribution. Then, a multistage dynamic optimization algorithm specifically tailored for large-scale time-varying network distribution route search is designed to obtain the stable and optimal solution. Subsequent experimental validations on actual case datasets have confirmed the correctness, effectiveness, and adaptability of the algorithm. Benchmarking against traditional CPLEX methods demonstrated that the algorithm not only rivals the best solutions but does so with a 38.8 times increase in computational speed. When pitted against the shortest path Dijkstra and A^∗ algorithms, the method consistently outperformed, delivering solutions up to 3.5 times faster in large-scale applications. Moreover, the parameter sensitivity analysis is performed on the algorithm by adjusting the safe flight thresholds of rainfall and wind speed parameters and revealed that the performance of the algorithm has a strong positive correlation with the size of the time-varying network.