Optimizing Controlled Environmental Agriculture for Strawberry Cultivation Using RL-Informer Model

Abstract

Controlled Environmental Agriculture (CEA) has gained a lot of attention in recent years, mainly because of its ability to overcome extreme weather problems and ensure food safety. CEA can meet the full growth state monitoring of the crop period; however, the optimization of the growing environment is still limited by the algorithm defects. In this article, we present an optimization method of growing environment based on reinforcement learning, Q-learning and the time-series prediction model Informer. This approach is demonstrated for the first time as far as we know. By employing Informer, we predicted the growth of strawberries based on current environmental variables and plant status. The prediction results served as a reward to motivate Q-learning, guiding it to make optimal modifications to the environment in real-time. This approach aimed to achieve the optimal cultivation environment continuously. Two groups of validation experiments were conducted based on different cultivation objectives: “obtaining the most stolons” and “obtaining the highest fruit count”. Compared to the empirically planted groups, the experimental groups using the RL-Informer model achieved yield increases of 17.81% and 20.78%, respectively. These experiments highlight the outstanding performance of the proposed RL-Informer model in real-time prediction and modification of environmental variables.