Robust Ensembled Model Based Reinforcement Learning for Better Tomato Yield in Greenhouse

The usage of autonomous greenhouses has become essential in meeting the food demands of the world's expanding population. Finding the right optimization strategy to sustain growth, yield, and profit is one of the major issues with greenhouse production. Addressing this issue effectively requires a combination of advanced technologies, data-driven insights, and innovative management practices. The overarching goal is to maximize crop yield and quality while minimizing input costs and environmental impact. The automated optimizations, which are often implemented using reinforcement learning algorithms, encounter issues with sample efficiency and robustness due to the time-consuming nature of the real-world simulation. Therefore, the goal of this research is to solve these issues by combining the SAC and Q learning algorithms to create an ensemble method. To properly optimize the worst-case inefficient samples, a discrete randomization and dropout module is included. The problem of sample efficiency and resilience is treated as a Mismatch Markov Decision Optimisation problem. The suggested model outperforms the current methods in handling the robustness and sample efficiency issues, according to an experimental evaluation. Additionally, this improvement increased production and maximized net profit.