Energy-efficient greenhouse climate control using Gaussian process-based stochastic model predictive control

Abstract

This paper proposes a Gaussian process-based stochastic model predictive control (GP-SMPC) framework for energy-efficient greenhouse climate control. In greenhouse systems, uncertainties arise from variations in crop growth rates and fluctuations in outdoor weather conditions, leading to suboptimal energy usage and increased operational costs. By incorporating a Gaussian process regression (GPR) model, the framework probabilistically captures uncertainties arising from crop growth variations and fluctuating outdoor weather conditions, enhancing robustness and efficiency. An online learning algorithm further improves the generalizability of the GPR model by capturing real-time observations, preventing overfitting problems. Numerical experiments using real-world greenhouse data demonstrate the significant energy-saving potential of the proposed framework. Compared to nonlinear MPC, the GP-SMPC framework achieves tracking error reductions of up to 67 % during the winter and 48 % in spring. Moreover, it reduces energy and CO₂ costs by up to 51.4 % during the winter season and 40 % during the spring season, minimizing resource wastage and operational inefficiencies. By optimizing resource usage while maintaining optimal growing conditions, the GP-SMPC framework provides a robust and sustainable solution for greenhouse climate control. This enhances the economic viability of high-tech food production systems.