A data-driven improved genetic algorithm for agile earth observation satellite scheduling with time-dependent transition time

The agile earth observation satellite (AEOS) task scheduling problem has been proven to be NP-hard. The traditional meta-heuristics is easy to converge too early or too late, and difficult to ensure the quality of the final solution. To address the AEOS task scheduling problem more effectively, a data-driven improved genetic algorithm (DDIGA) is proposed, which is composed of a traditional genetic algorithm, an artificial neural network(ANN), a frequent pattern-based new solutions construction procedure, and competition-based adaptive local adjustment strategy. In DDIGA, the data from the real-world or the history of the search is used to train the ANN model, and then the initial population is built by the trained ANN model. Next, some high-quality solutions created by selection, crossover, mutation operator are gathered to mine the frequent patterns, and some new solutions are constructed based on the chosen patterns. Finally, the new solutions are further improved by an optimization procedure, and competition-based adaptive local adjustment strategy is worked on these solutions with high similarity. Some scenarios are designed to verify the validity of the proposed approach. Extensive experiments on the satellite instances demonstrate that the DDIGA algorithm outperforms the state-of-the-art algorithms in solution quality and computation time. • We study the AEOS scheduling with the time-dependent transition time. • A Data-Driven Improved Genetic Algorithm is proposed to solve the problem. • A novel similarity detection method is proposed to test the population. • A data generation method for ANN training is proposed.