Greenhouse temperature regulation in the presence of uncertainties using data-driven robust model predictive control

Closed environment agriculture is gaining popularity due to the exponentially increasing demand for food, a growing shortage of water, urbanization, reduction in arable land, and degrading soil quality. Greenhouses are closed environment agriculture that provides a viable solution to these problems by maintaining favorable growing conditions. However, maintaining optimum conditions inside a greenhouse is a resource-intensive process, especially in hot and arid climates. Microclimate management, especially the temperature, requires a model-based systematic approach for consistent performance. Therefore, model predictive control is an effective method of managing greenhouse temperature; however, it requires a detailed system model. Moreover, the model predictive control strategy considers the perfect knowledge of the system while not accounting for uncertainties and disturbances. This leads to sub-optimal temperatures inside the greenhouse in the presence of uncertainties. Therefore, this study proposes a robust model predictive control framework to regulate the greenhouse temperature in the presence of external disturbances/uncertainties. An artificial neural network represents the nonlinear and dynamic greenhouse system. The model inputs are solar irradiance, ambient temperature, fan speed, and HVAC control, while the greenhouse temperature is the output. Results illustrate that the robust model predictive control algorithm has superior temperature control compared to a basic model predictive control and the existing greenhouse climate management system with a root mean squared error of 0.25 °C.