Machine learning modeling and optimizing of greenhouse climate conditions

Enhancing food security and increasing production efficiency can be achieved by adjusting various operating parameters in greenhouses that influence plant growth. These parameters include temperature, irrigation scheduling, humidity, supplemental carbon dioxide, sunlight exposure, and the use of artificial lighting on cloudy or low-light days. This study presents the results of an Artificial Neural Network–accelerated dynamic greenhouse model, which, for the first time, considers all of these parameters together and enables optimization of operational conditions based on minimizing the cost per kilogram of tomato yield over the cultivation period. The selected design variables are daytime temperature (for both day and night with artificial lighting), nighttime temperature (during the dark period), relative humidity, and carbon dioxide concentration. The use of the ANN reduced computation time considerably. Compared to typical greenhouse environmental settings (day time temperature = 22 °C, night time temperature = 18 °C, relative humidity = 80 %, CO₂ concentration = 800  ppm), the optimized environmental settings reduced production cost by 6 %, lowering it from 0.50 $.kg⁻¹ to 0.47 $.kg⁻¹. The optimum values of the objective function and design variables are 0.47 $.kg⁻¹, 23.7 °C for daytime temperature, 16 °C for nighttime temperature, 68.2 % RH, and 627.7  ppm CO₂ concentration.