Authors - Khalid BOUALI, Abderrahim BAJIT, Hamza BENZZINE, Ibtissam EL AOUNI, Hicham ESSAMRI, Siham BELOUALID, Rachid EL BOUAYADI Abstract - The transition towards intelligent greenhouse agriculture is driven by the need to optimize resource efficiency and manage climate variability. Classical control methods, such as PID and open-loop control systems, have shown limi-tations in precision and in handling the nonlinearities. To address these chal-lenges, advanced automation and predictive control strategies like Model Predic-tive Control (MPC) have gained attention. However, achieving accurate forecasts of meteorological parameters remains critical for maintaining optimal growing conditions and reducing energy consumption. This paper proposes a hybrid deep learning model that integrates Long Short-Term Memory (LSTM) networks with Transformer architecture within an MPC framework to improve forecasting of key meteorological parameters, including temperature, humidity, and solar irra-diance. By combining memory and attention mechanisms, the model captures both short- and long-term dependencies in time series data. Experiments were conducted with varying input window sizes, and performance was evaluated us-ing mean absolute error (MAE), root mean squared error (RMSE), and symmetric mean absolute percentage error (SMAPE). Results show that the proposed hybrid model outperforms standard LSTM models in both accuracy and inference time, highlighting its potential to support intelligent greenhouse climate management and promote sustainable agriculture.
