Efficient energy management and temperature control of a high-tech greenhouse using an improved data-driven model predictive control

Abstract

Greenhouses in arid climates require advanced control systems to maintain the microclimate and reduce energy utilization, ensuring economic viability. To address these challenges, model predictive control is an effective method that forecasts the system’s future state and adjusts control variables accordingly. However, deterministic model predictive control does not account for system uncertainties, leading to performance degradation. Therefore, this study proposes an improved model predictive control framework that utilizes an artificial neural network developed from historical greenhouse data. This method uses a double layer approach, where the primary controller provides the nominal trajectory, and an ancillary controller adjusts for uncertainties. The double layer predictive control framework was assessed under varying conditions to evaluate the performance in terms of temperature control and energy utilization. Results illustrated that, despite system uncertainties, the double layer model predictive control framework outperformed the existing greenhouse climate system, deterministic and robust model predictive control approaches. It demonstrated mean absolute errors of 0.09 °C in winter and 0.10 °C in summer, with corresponding root mean squared errors of 0.19 °C and 0.36 °C, respectively. Moreover, the double layer model predictive control method reduced energy utilization by 20.01 % in winter and 13.34 % in summer compared to the existing control system over a 4 d simulation period.