Transfer Optimization for Heterogeneous Drone Delivery and Pickup Problem

Xupeng Wen, Guohua Wu, Jiao Liu, Yew-Soon Ong

Abstract: Recently, with the enhancement of drone's payload and endurance capability, drone logistics has received extensive attention from giant logistics enterprises. This research delves into the Heterogeneous multi-Drone Delivery Pickup Problem (HDDPP) through a novel distribution mode. Here, a large drone ferries multiple smaller drones to a specified sub-region. These smaller drones oversee parcel delivery and pickup tasks before returning to automated airports for recycling. Then we formulate it into an open two-echelon multi-objective routing optimization problem, addressing pickup and delivery problems. To enhance optimization efficiency, relying on conventional optimization algorithms initiates the optimization process from an initial state, neglecting pertinent empirical knowledge embedded in optimized instances. This oversight results in a considerable squandering of computational resources. Based on the above motivation, we propose a Transfer Optimization with Heuristic Operators (TOHO) method. TOHO adopts insights from previously optimized task models, ensuring quick convergence for new tasks. We also present a mapping method rooted in topological alignment, enabling effective alignment between source and target instances, which fosters faster optimization convergence. Furthermore, we design a voting-mechanism-based ensemble genetic optimization strategy employing heuristic operators to select the Pareto frontier with high-quality diversity and convergence. Extensive experimental results on extended benchmark instances validate the efficacy of TOHO, and underscore its capability to speed up the optimization while ensuring superior knowledge transfer compared to baselines.

External IDs:dblp:journals/tetci/WenWLO25