Explainable column-generation-based genetic algorithm for knapsack-like energy aware nanosatellite task scheduling

Abstract

The Offline Nanosatellite Task Scheduling (ONTS) problem poses a complex optimization challenge, focused on maximizing the number of tasks executed by a satellite in orbit while adhering to Quality of Service constraints such as priority, execution time-frames, and resource management. Based on mixed integer programming, existing methods rely on branch-and-bound aided algorithms and can struggle to achieve satisfactory computational time performance. In order to avoid the computational burden of branch-and-bound, this work introduces the Column-Generation-based Genetic Algorithm (CGbGA) as a heuristic approach to the ONTS problem. The method, based on branch-and-price principles, combines Genetic Algorithm (GA) and Dynamic Programming (DP) to solve the problem of interest efficiently. We generate solution vectors for each job using DP and adapt mutation and crossover operators to work on a column-wide scale. This ensures that every solution is valid for the given job. Also, a novel pseudo-shadow pricing strategy is employed to mimic the pricing procedure of the branch-and-price algorithm. To better understand the impact of the number of available columns on the incumbent solution, we employ Local Interpretable Model-Agnostic Explanations (LIME). Our results, based on a set of representative literature instances, demonstrate the potential of CGbGA in terms of solution value and computational solving time compared to commercially available solvers.