Reinforcement Learning-Based Intelligent Resource Optimization for Sustainable Indoor Agriculture

Efficient control of indoor agricultural environments requires continuous regulatio of temperature, humidity, carbon dioxide concentration, and soil moisture. Traditional
greenhouse systems operate using fixed thresholds and rule-based strategies, which are unable to adapt to dynamic environmental changes. In this work, a reinforcement learning based intelligent control framework is developed using a Deep Q-Network (DQN) to autonomously optimize
irrigation, cooling, and CO₂ supply. A simulated greenhouse environment is modeled using four state variables, and the RL agent learns optimal actions through interaction with the environment based on a reward function that reflects plant comfort conditions. The model is trained for 50 episodes, each consisting of 30 control steps. Experimental results show that the total reward increases from very low values in early episodes to above 30 in later episodes, indicating that the agent successfully learns to maintain stable and optimal environmental conditions. The proposed system demonstrates effective autonomous control and improved environmental stability,making it suitable for sustainable indoor agriculture applications.By optimizing irrigation,cooling, and CO₂ injection through intelligent learning, the proposed system significantly reduces unnecessary water and energy usage, thereby minimizing the carbon footprint of indoor farming operations. This AI-driven control framework supports sustainable digital agriculture by enabling efficient, resource-aware greenhouse management.