Research on collaborative scheduling strategies of multi-agent agricultural machinery groups

Abstract

Addressing the challenges of high scheduling costs and low efficiency in the collaborative operations of agricultural machinery across multiple dispatch centers, this paper develops a scheduling model designed to minimize total costs. It introduces a Multi-Center and Multi-Machine Path Planning Algorithm Based on Deep Reinforcement Learning (MCMPP-DRL). Firstly, a deep reinforcement learning environment is established, and a policy network utilizing the attention mechanism is designed. Subsequently, the network model is trained to improve its solution capabilities using the policy gradient algorithm (REINFORCE). Finally, the solution is optimized through a local search strategy. In this study, three dispatch centers were selected within the maize growing area of Hebei Province, and comparative analyses were conducted for 20, 40, 50, 100 and 120 farmlands, respectively.The results indicate that the MCMPP-DRL algorithm achieves a reduction in total scheduling costs of at least 9.66%, 14.34% and 24.41% compared to Ant Colony Optimization (ACO), Simulated Annealing (SA) and Genetic Algorithms(GA), respectively.The significant optimization in scheduling costs demonstrates that the MCMPP-DRL algorithm establishes a robust theoretical foundation and offers technical support for addressing complex scheduling problems involving multiple dispatch centers and multiple.