Cooperative Scheduling of Heterogeneous Farm Machines for Farmlands with Multi-Entry/Exit

Proper scheduling strategies for farm machinery are crucial for reducing farm operation time. However, existing methods are difficult to effectively represent, capture and utilize complex dependencies between tasks and resources, such as resource competition, region matching and path selection dependencies, which are important for optimizing farm machine scheduling. In this paper, we propose a deep reinforcement learning-based task-resource dependency management network model for managing complex dependencies between tasks and resources and optimizing scheduling strategies. The model consists of a task policy network, a resource policy network and a parameter optimization module. The task policy network captures and manages the dynamic dependencies between tasks and resources in real time through the Dynamic Undirected Disjunction Graph and Dynamic Bottleneck Optimization modules to optimize the task execution order. The resource policy network allocates appropriate resources for tasks. The parameter optimization module, on the other hand, uses the Multi-Proximal Policy Optimization algorithm to optimize the policy network to further enhance the scheduling efficiency. The experimental results show that the model reduces the farm job completion time by an average of 49.51 %,35.54%, and 32.51 % compared with the rule-based heuristic algorithm, meta-heuristic algorithm, and deep reinforcement learning algorithm on the two publicly available datasets, significantly outperforming the existing methods.