Horticultural Science and Technology 원예과학기술지 ISSN:1226-8763(Print) 2465-8588(Online)

Greenhouse growers are spending a lot of money on energy management, such as for heating, cooling and CO₂ enrichment. To date, many studies have been conducted on energy-consumption prediction models in greenhouses. However, no study has examined ventilation controls for energy saving for a given geographical location. The objective of this study was to use the predicted internal temperature from an Artificial Neural Network (ANN) model to control the ventilation system and to reduce energy costs in greenhouses. For developing the model, we carried out the preprocessing of collected data. First, to detect and eliminate the noise from sensors, we used the Kalman filter algorithm. Then, the dimensions of these data were reduced using Pearson Correlation Coefficient analysis to enhance the accuracy of the model. The ANN model was developed using a backpropagation algorithm, which is a supervised learning method for calculating the weight of nodes. The Levenberg-Marqardt method was used as a learning algorithm. Hyperbolic Tangent was also used as an active function for continuous differentiation of weights of the ANN. This study found that the root mean square errors of the ANN, Multiple Regression Model and Recurrent Neural Network were 1.723, 1.834 and 1.971 respectively. Therefore, the ANN predicted value was more accurate than other prediction models. The predicted greenhouse temperature was used to control greenhouse ventilation. Ventilation of windward and leeward sides was controlled separately by the P-band. The control of ventilation can be performed by different ranges of the P-band for different seasons. Applying predicted temperature data for the P-band’s range based on the ANN to control ventilation can minimize energy loss by opening and closing the window in advance.