Enhanced real time electricity consumption forecasting using a hybrid deep learning model for greenhouse

Accurate energy consumption forecasting is critical for achieving sustainability in smart greenhouse operations. Existing household load forecasting techniques have advanced considerably, yet their direct application to greenhouses remains challenging due to dynamic environmental factors and sensor inaccuracies. These issues often result in inefficient energy scheduling and higher operational costs, emphasizing the need for adaptive and data-driven forecasting models. To address this, the present study proposes a hybrid deep learning framework CBLNet for real time electricity consumption forecasting in greenhouse environments. The model integrates CatBoost, Backpropagation Neural Network (BPNN), and Long Short-Term Memory (LSTM) networks to capture nonlinear feature interactions and temporal dependencies. A multi-time-scale analysis of electricity consumption patterns further enhances robustness and accuracy. Trained and validated on one year of real-world greenhouse data (52,000 samples at 10-minute intervals) and optimized through grid-search cross-validation, CBLNet achieved Mean Absolute Percentage Error of 3.16%, Root Mean Square Error of 0.12, Mean Absolute Error of 0.032, and Mean Bias Deviation of -0.12. Deployment on a Raspberry Pi 3B+ confirmed its feasibility for low-cost edge computing. The results establish CBLNet as an accurate, scalable, and sustainable forecasting solution that advances data-driven energy management in smart greenhouse systems.