Intelligent control of artificial lighting for controlled environment agriculture

This thesis presents an intelligent lighting control system for indoor horticulture, designed to improve energy efficiency while delivering precise light conditions for optimal crop growth. The system combines a neural network-based predictive model with real-time optimization to adjust the height, angle, and dimming of motorized LED fixtures. A feedforward neural network was trained on experimentally collected data to estimate Photosynthetic Photon Flux Density (PPFD) at multiple tray locations based on lighting parameters. Two optimization methods, Gauss-Newton and L-BFGS-B, were implemented to compute configurations that meet PPFD targets while minimizing energy consumption. L-BFGS-B achieved faster convergence and higher accuracy across all tested scenarios. The system was deployed in a Deep Water Culture (DWC) hydroponic setup using romaine lettuce and operated in real time on a Raspberry Pi. Compared to a fixed lighting setup, the adaptive system reduced energy use by 24.2%, lowered the energy cost per gram of dry mass from 1.81 to 0.65 kWh/g, and nearly doubled the fresh biomass yield. These results demonstrate the potential of machine learning-driven lighting systems to enhance both sustainability and productivity in controlled environment agriculture.