Evaluating Deep Learning Models for Prediction of Climate Conditions in Greenhouse

This study evaluates the performance of four representative deep learning models-1D CNN (Convolutional Neural Network), LSTM (Long-Short Term Memory), DNN (Deep Neural Network), and Transformer-for environmental prediction in smart greenhouses using 13 sensing data. Data collected at one-minute intervals over nine months were used to compare the prediction performance for temperature, humidity, and CO<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> levels at three different time intervals (5 minutes, 30 minutes, and 60 minutes). The results showed that prediction errors increased as the time interval grew larger. Predictions within 30 minutes were deemed applicable for complex environmental control in smart farm systems, while predictions for 60 minutes ahead could be used to forecast general trends. This approach can support more precise environmental control through proactive predictive management in smart greenhouses.