Daily temperature optimisation in greenhouse by reinforcement learning

Abstract

The goal of this study is to show the usefulness of reinforcement learning (RL) to solve a common greenhouse climate optimisation problem. The problem is to minimise the daily heating cost while achieving simultaneously two agronomic goals, namely maintaining a good crop growth and an appropriate development rate. The complexity of the problem is due to the very different time constants of these two biological processes. First, a simple model for greenhouse roses is presented that simulates the daily crop growth and development. Second, the RL method is presented, in its application to this problem. Finally, optimisation results are presented and discussed. Copyright © 2005 IFAC.