Field-based implementation of machine learning forecasting for energy-efficient temperature control in a greenhouse system

In this study, we developed a simulation model for the internal environment of greenhouses using artificial intelligence techniques. This study focuses on the practical application of machine learning (ML)-based forecasting for greenhouse temperature control and its potential to improve energy efficiency under real-world operating conditions. This study was conducted to optimize energy consumption through smart agriculture applications. A prediction model was established using a time-series ML approach, followed by the development of forecasting models through gap-labeling feature analysis. Data were collected from a greenhouse equipped with a pellet boiler-based heating system over 55 d under real operating conditions. Iterative learning was conducted using 7 d of training data and 1 d of validation data, resulting in 48 models (average r2 = 0.77, RMSE = 1.43). Based on the forecasting results, a data-driven control strategy was applied to adjust the heating operation in advance. The reduction in heating operation time demonstrates improved energy efficiency of the greenhouse system under practical operating conditions by minimizing unnecessary heating. Implementing data-driven predictive control using the developed forecasting model can save 5 %–15% of the energy. Statistical analysis using the Wilcoxon signed-rank test indicated that the performance differences among the models were not statistically significant (p > 0.05). Therefore, this study emphasizes the practical implementation of machine learning-based forecasting for real-time greenhouse control under actual operating conditions.