Intelligent control and monitoring for energy-efficient horticultural lighting

This thesis presents an intelligent supplemental lighting control and monitoring system for controlled environment agriculture (CEA). The intensity and quality of light are essential factors in plant growth and morphology. Thus, maintaining light in the photosynthesis spectrum with specific intensities, color ratios, and light-dark cycles called light recipes, is of utmost importance. To this end, our research objective is to address challenges in three main areas of lighting related to control, sensing, and integration, for the purpose of achieving cost, energy, and resource-efficient CEA. We propose a learning neural network controller for a multi-input, multi-output (MIMO) light- ing system consisting of light sensors and dimmable horticultural LEDs. In the first step, a proof- of-concept platform was built with emulated sunlight to verify the proposed scheme. To improve performance in the presence of nonlinearities, we extend our design to a nonlinear neural network controller with input saturation compensation. A Lyapunov analysis is presented to guarantee convergence of the errors to an arbitrarily small ultimate bound. A predictor feedback delay compensation scheme is also developed to counteract the destabilizing effect of communication delays. In the second step, to avoid the need for expensive quantum sensors commonly used in horticulture, we formulated Photosynthetic Photon Flux Density (PPFD) parameters as functions of multi- spectral sensor readings and utilized machine learning (ML) algorithms to estimate the mapping. Furthermore, a crop monitoring system with inexpensive imaging arrays and image recognition algorithms is presented, consisting of modified Naive Bayes classifiers that estimate key color and geometrical plant features to identify plant types. The image-acquisition framework is later used to create a database and post-process image data with ML algorithms to monitor plant health and growth. In the last step, an IoT-based system is implemented consisting of the neural network controller, ML-based PPFD sensors, and plant monitoring system to optimize light intensity, duration, and spectral distribution based on plant requirements. The system delivers desired Daily Light Integrals (DLIs) to the plant canopy with minimum energy usage and is model-free and highly scalable. Plant experiments conducted in a research greenhouse indicate that the proposed solution reduces energy consumption per unit dry mass of lettuce by 28% compared to conventional time scheduling techniques and prevents plant health issues due to excessive light exposure.