Efficient Greenhouse Temperature Control with Data-Driven Robust Model Predictive

Appropriate greenhouse temperature should be maintained to ensure crop production while minimizing energy consumption. Even though weather forecasts could provide a certain amount of information to improve control performance, it is not perfect and forecast error may cause the temperature to deviate from the acceptable range. To inherent uncertainty in weather that affects control accuracy, this paper develops a data-driven robust model predictive control (DDRMPC) approach for greenhouse temperature control. The dynamic model is obtained from thermal resistance-capacitance modeling derived by the Building Resistance-Capacitance Modeling (BRCM) toolbox. Uncertainty sets of ambient temperature and solar radiation are captured by support vector clustering technique, and they are further tuned for better quality by training-calibration procedure. A case study shows that the DDRMPC has better control performance compared to rule-based control, certainty equivalent MPC, and robust MPC. DDRMPC approach ends up with 12% less total energy consumption than rule-based control strategy.