Loading...
Multi-Regime GRU-Scheduled PID Control for Greenhouse VPD Regulation with Real-Time PLC Deployment
Author(s)
Wanna, Natthanan
Nabudda, Kriengkrai
Sitthaphanit, Suphasit
Thasnas, Natakorn
Date Issued
January 1, 2026
Type
Article
Abstract
Vapor pressure deficit is a physiologically meaningful climate variable governing plant transpiration and water transport in greenhouse cultivation. However, accurate regulation in evaporative-cooled greenhouses is challenging due to nonlinear temperature-humidity coupling, time-varying dynamics, and rapidly changing solar disturbances. Conventional fixed-gain proportional-integral-derivative controllers, tuned for a single operating condition, often exhibit performance degradation under regime transitions. This study proposes a multi-regime admissible-region-constrained gain-scheduled control framework in which a gated recurrent unit network serves as a supervisory scheduler to generate smooth, near-optimal controller gains for real-time vapor pressure deficit regulation. Control-oriented first-order-plus-dead-time models are identified under representative operating regimes, and an interval-bounded plant ensemble is constructed to capture dynamic variability. Regime-wise optimal gains are obtained offline using hybrid evolutionary optimization and used as supervisory training labels. To support safe deployment, all scheduled gains are constrained within a pre-validated admissible stability region, such that each deployed gain configuration remains inside an offline-validated frozen-time stabilizing set. The controller is implemented on an industrial programmable logic controller and experimentally validated in a full-scale evaporative greenhouse under real environmental disturbances. Comparative multi-day experiments demonstrate improved tracking accuracy, reduced overshoot, smoother actuator behavior, and enhanced disturbance attenuation relative to fixed-gain strategies, while maintaining deterministic real-time execution. The results establish an industrially deployable learning-based gain-scheduling framework for practical greenhouse climate control.
Citation
IEEE Access, 14, 69612-69632, 2026