Greenhouses have become a pivotal solution in modern agriculture, providing controlled environments that address various challenges such as climate variability, pest infestations, and resource inefficiencies. However, efficient management requires advanced technological solutions. Recent advancements in the Internet of Things (IoT), computational intelligence (CI), and cloud computing have revolutionized greenhouse monitoring and optimized productivity. Despite these benefits, real-time processing of data and the high energy consumption of Wireless Sensor Networks (WSN) present ongoing challenges. This paper presents a novel IoT-based system integrating a hybrid edge, fog, and cloud architecture to enable real-time, localized sensor data processing and reduce dependance on cloud resources. Furthermore, we propose an AI-driven algorithm operating in a dynamic ON/OFF mode to optimize WSN activation and sensor data aggregation. By leveraging data aggregation techniques and advanced time series forecasting models, our approach enhances energy efficiency while ensuring accurate microclimate monitoring. The models employed in this research study are as follows: Long Short-Term Memory (LSTM), Transformer, and the hybrid LSTM-Transformer model. These models were trained to predict internal greenhouse conditions using historical variations and real-time external climate inputs. The evaluation process incorporated the use of R2 score, MAE, MSE, and RMSE. Experimental results demonstrate that the LSTM-Transformer model achieved superior forecasting performance, with an R2 score average of 99.94%, followed by LSTM and Transformer. By minimizing unnecessary sensor activation and improving data aggregation efficiency, our proposed approach enhances the sustainability and energy efficiency of intelligent greenhouse management systems.