Sensor-Driven Architecture for Resource-Efficient Hydroponic NFT Vertical Farming

Abstract

Urbanization and freshwater stress are intensifying pressure on conventional food systems, motivating closed-environment production. This Short Communication specifies a sensor-driven architecture for hydroponic nutrient film technique (NFT) leafy-green production, comprising an in-situ sensing layer, a three-tier Internet of Things (IoT) data path, and a gradient-boosted decision-tree (GBDT) scheduler. The architecture is evaluated in a coupled lettuce-growth and room-climate simulator built on the Van Henten crop model. Three controllers were compared over a single 14-day trajectory under one fixed seed. The GBDT scheduler holds the room inside the target envelope 75.8 percent of the time at 91.3 kWh per day, against a rule-based hysteresis baseline (15.6 percent, 140.1 kWh per day) and a Ziegler-Nichols-tuned proportional-integral-derivative (PID) reference (77.9 percent, 96.8 kWh per day). On a combined cost J = 10(1 - a) + 0.05E, the scheduler reduces the hysteresis cost by 55 percent, while it and the tuned PID differ by 1.0 percent (6.98 against 7.05), a gap a single replicate cannot resolve. Reported predictor coefficient-of-determination values describe in-distribution quality on simulator-generated data, not generalization to physical sensors. Avoiding a per-installation relay auto-tune step is the design rationale and is not tested here.