S06 - Session O1 - An AI approach for greenhouse control combining models, data and knowledge.

Authors: Jack Verhoosel *, Behrouz Eslami Mossallam, Ellen van Bergen, Michael van Bekkum, Richard Dekker, Paolo de Heer

The aim for a greenhouse grower is to control the climate such that the crop is optimally cultivated against the lowest cost and environmental footprint. The grower uses various control systems that support decision-making for e.g., climate control, irrigation and nutrition schemas. The objectives of these systems can however be contradictory and may require alignment to reach an optimal overall setting. With the number of control systems increasing, this is increasingly challenging. To tackle this challenge, we have developed an AI-based decision support system, GAIA, that acts as a bridge between control systems to find optimal solutions for its settings. In GAIA, the operator can set a performance function that combines objectives of the different control systems. GAIA then makes a prediction of the development of the performance function using a Model Predictive Control (MPC) genetic algorithm. Control setpoints are presented to the operator as an advice accompanied by an explanation, showing predictions of the individual system components. GAIA maintains a knowledge graph with control rules that captures common knowledge about greenhouses, crop and growing strategies. The control rules constrain the solution space of the MPC algorithm and are presented as part of the explanation, showing how they relate to the control setting advice. Preliminary simulation results with a tomato plant model show that the precision of the MPC predictions is very close to targeted optimal values. When adding control rules, MPC can still find an approximately optimal solution under constraints. Future work targets evaluating the GAIA system in a real-life greenhouse compartment with a grower giving feedback on a given advice and optionally providing additional control rules. This feedback and extra knowledge will be used by GAIA to improve its advice. The GAIA system was developed in a national Dutch project with greenhouse construction and control system companies.