Real-Time VR-in-the-Loop Model Predictive Control for Greenhouse Temperature Regulation with LSTM Forecasts

Abstract

This paper presents a real-time greenhouse temperature control system integrating Model Predictive Control (MPC) with deep learning forecasts and a Virtual Reality (VR) human-in-the-loop interface. The indoor climate is regulated by an MPC algorithm that optimizes heating and ventilation actions using predictive models of the greenhouse and upcoming weather conditions. Long Short-Term Memory (LSTM) neural networks provide short-term forecasts of external temperature, solar radiation, and wind, enabling the MPC to anticipate disturbances. A VR interface allows expert users to immerse in a virtual greenhouse environment and dynamically adjust the temperature setpoints in real time. By combining automated predictive control with human expertise, the system aims to improve climate regulation, energy efficiency, and adaptability to unforeseen conditions. Simulation case studies compare a baseline MPC-driven climate control against scenarios where a human operator in VR overrides setpoints during critical events. The results indicate that VRbased setpoint adjustment can enhance temperature stability and constraint compliance, with minor trade-offs in energy consumption. This work demonstrates the potential of integrating advanced control, AI-based forecasting, and immersive VR technology for next-generation smart greenhouse management.