Long Short-Term Memory with Bahdanau Attention based Precision Smart Farming

In recent years, Artificial Intelligence (AI) has improved precision agriculture tremendously in recent years by providing remote capabilities and automated decision-making. However, a regression-based approach for traditional agricultural and greenhouse environments supports limited ability to model temporal dependencies or better understand the importance of time series data. To address this, we proposed and developed a hybrid deep learning model pertaining to intelligent greenhouse management using LSTM with Bahdanau Attention (LSTM-BA). Initially, the proposed method uses Internet of Things (IoT)-based sensors to collect multivariate time series data that includes greenhouse variables such as temperature, humidity, CO₂, soil moisture content, and light intensity data. Preprocessing data will then take place using normalization and windowing to develop a pre-processed time series dataset. Afterwards, the LSTM-BA model will be employed in the fog layer to predict actuator control values with more interpretability while also paying attention to relevant time steps. From the results, LSTM-BA model outperformed the other regressors, Support Vector Machine (SVM) in terms of Root Mean Square Error (RMSE) of 0.07 respectively.