Task planning based on improved dimensionality reduction strategy

Abstract

This paper addresses the complex and challenging problem of task planning by formulating it as a high-dimensional multi-objective optimization problem and establishing the mathematical model. To address the shortcomings of traditional algorithms, the paper proposes improved algorithms, including an average weighted fitting of redundant objectives and a target dimensionality reduction method based on an improved Aggregation tree, aimed at enhancing the efficiency and robustness of the algorithm. The approach incorporates a Pointer Network and a roulette strategy to generate high-quality initial populations, thereby accelerating algorithm convergence and optimization. Article’s results show that the improved algorithm performs excellently concerning hypervolume and spatial metrics. Its effectiveness is further validated in practical applications, particularly in improving task completion rates, prioritization, time efficiency, and bandwidth utilization.