An AI approach for greenhouse control using models, data, and knowledge

A greenhouse grower uses various systems to control the climate, energy use, irrigation levels, nutrition schemes and labor planning. The objectives of these systems can however be contradictory, for instance optimal climate may require high energy use. Thus, an intermediate solution needs to be found between these objectives. As the number of control systems being used in the greenhouse increases, this becomes more complex for the human grower. Therefore, we have developed the GAIA system that supports the grower in finding this intermediate solution. GAIA takes as input a weighted performance function of the individual control system objectives. As output, GAIA presents a prediction of the best setpoints for each system that together maximize the performance function value. To achieve this, GAIA uses a combination of data-driven AI techniques, a model-based prediction algorithm and a knowledge base with expert control rules. The system supports explainability of its results and active feedback loops with the grower. Preliminary simulation results with a tomato plant model show that the GAIA predictions are very close to the targeted optimal values. Future work targets evaluating the GAIA system in a real-life greenhouse compartment with a grower providing feedback on a given advice and setting additional control rules.