AI-enabled cyber-physical-biological systems for smart energy management and sustainable food production in a plant factory

Abstract

The advancement of controlled-environment agriculture, particularly in plant factories, offers an innovative solution to address the rising demand for food due to global population growth and urbanization. These controlled environments provide consistent and predictable crop yields, irrespective of external weather conditions, and can be tailored to achieve optimal plant growth. However, the intensive energy requirements of these systems have raised sustainability concerns. In plant factories, which provide regulated environments for sustainable food production, it remains essential to minimize energy consumption while maintaining operational efficiency. This study introduces a novel cyber-physical-biological system (CPBS) for managing energy and crop production in plant factories. The CPBS accurately captures plant biological dynamics, such as temperature, humidity, lighting, and CO2 levels, optimizes control variables, and predicts crop growth within these controlled environments. To achieve these outcomes, we leverage physics-informed deep learning (PIDL) techniques to develop high-fidelity and computationally efficient digital twins for the plant factory's internal microclimate and crop states. PIDL enables us to capture complex relationships between environmental factors and crop growth, thereby improving accuracy and decision-making in control. Using the CPBS, we optimize energy usage and resource expenses to ensure sustainable crop production rates under different daylight scenarios in the plant factory. Simulation results from a full growth cycle demonstrate that our proposed CPBS, compared to a certainty equivalent model predictive control (MPC), reduces violation cases by 84.53%. Additionally, it achieves a reduction of 13.41% and 13.04% in energy and resource usage, respectively, compared to a traditional robust MPC that considers a box-shaped uncertainty set.