JAUE2024-065: Development of a Random Forest-Based Short-Term Prediction Model for Indoor Mean Radiant Temperature Using Time-Lagged Features

Abstract

This study presents an improved approach for short-term indoor Mean Radiant Temperature (MRT) prediction using a Random Forest model with time-lagged features, distinctly forecasting MRT for core and perimeter zones. Unlike previous studies focused on current MRT estimation, this research enhances proactive HVAC control potential through short-term prediction and spatial differentiation. The model, developed using 5-minute interval data collected over 15 days, incorporates time-lagged features for key environmental variables. Results show high prediction accuracy for both core (R² = 0.989) and perimeter (R² = 0.984) zones, with time-lagged features reducing Mean Absolute Error by 15% in the core zone and 18% in the perimeter zone. This approach effectively captures indoor thermal environment dynamics, potentially optimizing HVAC energy efficiency and occupant comfort. Future research should validate the model across various building types and seasons, and assess its effectiveness in actual HVAC systems.