노지 환경 데이터를 활용한 최적 관수 예측 모델 연구 = (Study on optimal irrigation prediction model using field environment data)

다국어 초록 (Multilingual Abstract)

Precision agriculture using AI technology is currently attracting attention, and this study proposes an AI-based model that predicts the optimal irrigation period of field crops using open-field environmental data. In particular, the prediction performance of irrigation period was analyzed by applying the Random Convolutional Kernel Transform (ROCKET) model optimized for time series analysis. The proposed method is to establish a system for predicting the amount of irrigation by utilizing the weather environment, soil environment, and crop growth data. As a result of the experiment, the ROCKET model recorded better prediction accuracy (RMSE = 2.34, MAE = 1.89, R² = 0.92) and calculation speed than the existing LSTM and Random Forest models, and it was analyzed that soil humidity, rainfall, and external temperature are factors that influence the optimal irrigation. In particular, it was confirmed that the ROCKET model can be quickly learned without large-scale data, so it is highly likely to be applied as a real-time precision irrigation system in a smart farm environment. This study presents the possibility of implementing a precise irrigation system in an AI-based smart farm environment, and plans to further improve model performance through applicability in various environments and optimal hyperparameter exploration in the future.