Emulating expert environmental control strategies in strawberry greenhouses using attention-LSTM

• Attention-LSTM learns expert-level greenhouse control strategies from data. • Integrates environmental dynamics and actuator decisions in a unified model. • SHAP analysis confirms alignment between model decisions and grower logic. • Real-world deployment reduces microclimate fluctuations and actuator overuse. • Provides a practical foundation for autonomous, expert-informed greenhouse control. Environmental control in greenhouses is crucial for ensuring stable crop production and quality, particularly for high-value crops such as strawberries. Conventional rule-based methods struggle to adapt to the nonlinear and dynamic interactions between environmental factors and actuator responses. To overcome these limitations, we propose an intelligent greenhouse control framework that emulates expert decision-making using an Attention-LSTM model trained on historical environmental and control data from a high-yield strawberry greenhouse. The proposed model achieved consistently high predictive performance across multiple benchmarks and outperformed conventional LSTM-based approaches. SHAP-based interpretability further confirmed that the model captured expert-level strategies by prioritizing key environmental variables. When deployed in a real greenhouse, the framework reduced unnecessary actuator operations, stabilized internal microclimates, and replicated control patterns observed in expert-managed facilities. In conclusion, this study demonstrates the practical feasibility of data-driven expert-level greenhouse control and provides a solid foundation for developing more autonomous and adaptive smart farming systems.