Optimizing irrigation scheduling using Deep Reinforcement Learning and crop growth model
Jiamei Liu, Jiahong Yang, Xinyi Jie, Fangle Chang, Longhua Ma, Hongye Su · 2025
Abstract. As the main area of water resource consumption, agricultural water use faces the challenges of low efficiency, and requires precise irrigation scheduling to reduce the waste. In this study, we developed a smart irrigation system integrating Simulation data with predictive models to optimize irrigation scheduling. An interactive environment was generated by integrating deep reinforcement learning (DRL) with the crop growth simulator DSSAT to optimize the agent's decision-making ability. Simulation data were collected and analyzed to provide dynamic feedback to the simulator. Meanwhile, the Long Short-Term Memory (LSTM) network was developed to predict the soil moisture changes and crop water demand for the following day. Additionally, the Soft Actor-Critic (SAC) algorithm was developed to manage continuous action spaces and precisely control irrigation amounts. Over-irrigation and under-irrigation were successfully avoided by utilizing SAC's maximum entropy mechanism to optimize both water use efficiency and crop yields.