Adaptive Reallocation and Optimization Methods for Emerging Dynamic Tasks of UAV Inspection

In UAV smart grid inspection, UAVs initially adhere to a predefined planning scheme that specifies the number of tasks and their execution order for routine inspections. However, unexpected task scenarios may arise in the inspection environment, necessitating dynamic readjustment of the planning scheme. To address this, a two-stage adaptive task optimization scheduling method integrating Deep Reinforcement Learning (DRL) and a Multi-objective Genetic Algorithm is proposed. Firstly, based on the current task state and task priority, UAVs are selected for emerging task reallocation through a DRL-based UAV selection decision module. Secondly, the remaining task execution sequence for UAVs is optimized by an NSGA-II-based sequential optimization module. This module enhances the algorithm's effectiveness by refining the crossover and mutation operators as well as the crowding degree calculation formula. Simulation experiments demonstrate that the proposed method reduces the average task execution time by 16.84% compared to the Ant Colony Optimization and Distance Priority Sorting Algorithm, thereby significantly improving the adaptability of UAV smart grid inspection in handling additional task reallocation.