A novel multi-source data-driven energy consumption prediction model for Venlo-type greenhouses in China

The high energy consumption characteristic of multi-span glass greenhouses significantly limits their widespread adoption. Optimizing energy strategies and implementing predictive models for energy consumption are essential for more efficient management and reduction of greenhouse operational energy costs. Existing methods rely on single-element approaches to predict energy consumption, but this reliance often results in severe performance limitations. Therefore, energy consumption prediction methods that incorporate multi-source data are necessary. To overcome the challenges concerning heterogeneity, redundancy, and interdependence among different data sources, this paper proposed a novel energy consumption method that integrates multi-source data through feature engineering and machine learning techniques, which significantly enhances the efficiency of data utilization and improves prediction accuracy. The final experimental results indicated that the proposed energy consumption prediction method demonstrates excellent performance with high prediction accuracy (R² = 0.9388) and low computational resource consumption (runtime = 926.91s), outperforming other models. Finally, the model was interpreted using SHAP (SHapley Additive exPlanations) values, and ablation experiments were conducted to validate the effectiveness of the proposed method in greenhouse energy consumption prediction, thereby providing strong support for greenhouse management.