Study and Development of an Adaptive Vertical Farm

The world’s population is projected to approach 10 billion by 2050, driving an expected rise in food demand due to population growth, economic development, and urbanization. To meet this demand sustainably, greenhouse systems, particularly vertical farming, have emerged as a promising solution, offering high crop yields per unit of cultivation area. The Adaptive Vertical Farm (AVF), is an innovative industrial vertical greenhouse that dynamically adjusts the distance between its stacked shelves, optimizing growing conditions as plants progress through their growth stages. This adaptive principle overcomes the traditional conflict between maintaining optimal conditions and minimizing energy consumption. This thesis presents two main research axes contributing to the development of the AVF. The first axis focuses on developing a data-driven, black-box growth model for crops cultivated in AVFs. Given the dynamic adaptation to plant growth, an accurate crop growth model is essential for optimizing shelf movement and system control. While traditional dynamic growth models often rely on numerous parameters and are specific to certain crop types, we propose a black-box approach using feedforward neural networks to predict plant height at each time step. This model is adaptable to various crop types, computationally efficient after training, and particularly suitable for innovative vertical farming systems like the AVF. The effectiveness of the model is demonstrated through synthetic and real-world datasets, showcasing its potential in optimizing crop production. The second research axis focuses on automating the AVF using Unmanned Aerial Vehicles (UAVs) within the context of Precision Agriculture. UAVs support applications such as crop health monitoring, automatic pollination, spraying, and irrigation, optimizing farm operations across stacked shelves and complementing the existing stationary sensors. To support this automation, we introduce observer designs for nonlinear systems using Linear Matrix Inequalities (LMIs) to provide accurate state estimation for UAVs, ensuring exponential convergence of the observer. Two key contributions are presented: first, a new, less conservative LMI condition applied to solve the H∞ circle criterion design, and second, a nonlinear observer design based on output dynamic extension. This method minimizes the impact of measurement noise and guarantees the Input-to-State Stability (ISS) property of the estimation error via a novel LMI condition.