Reinforcement learning control method for greenhouse vegetable irrigation driven by dynamic clipping and negative incentive mechanism

The irrigation control is a critical aspect of greenhouse vegetable production. However, existing agricultural irrigation studies face limitations such as high equipment requirements, overly complex systems and difficulty in configuring algorithm parameters. This study proposes a greenhouse vegetable irrigation prediction method based on an improved Proximal Policy Optimization (PPO) algorithm. By integrating various greenhouse environmental factors and reinforcement learning algorithms, the study establishes a reinforcement learning framework to simulate vegetable growth. To address the challenges of continuous action space and high-dimensional state space, this study introduces the PPO algorithm to enhance convergence efficiency, thereby proposing an enhanced reinforcement learning algorithm (ENPPO). Experimental results demonstrate that the ENPPO algorithm outperforms two other methods in irrigation control. By utilizing real-time environmental data and historical irrigation records, the ENPPO algorithm predicts reasonable irrigation amounts, achieving precise irrigation control to enhance vegetable growth efficiency. The study explicitly distinguishes between irrigation prediction and control methods, providing a comprehensive technical approach to improving water resource utilization and reducing agricultural production costs.