Comfort Management Based on Predictive PMV in Smart Homes

Traditional heating, ventilation, and air conditioning (HVAC) systems typically function on predetermined schedules, disregarding variations in occupancy or external environmental conditions. However, with the advent of smart homes, there is a growing demand among home managers and owners for solutions that can reduce energy consumption while ensuring occupant comfort. The Predicted Mean Vote (PMV) is a widely used metric that quantifies occupant thermal comfort based on factors like temperature, humidity, airspeed, and personal factors such as clothing and activity level. In this paper, we propose a comfort management based on predictive PMV approach for providing indoor thermal comfort in smart homes. The system is composed of three essential modules: a temperature prediction model, a PMV prediction model, and a recommendation engine. The temperature prediction module utilizes a Convolutional Neural Network - Gated Recurrent Unit (CNN-GRU) architecture to forecast indoor conditions by capturing both spatial and temporal dependencies from historical data, ensuring precise predictions that contribute to maintaining optimal thermal comfort. The PMV prediction module employs a two-stage approach, combining models such as Extreme Gradient Boosting (XGBoost), random-forest, and Deep Neural Network (DNN), followed by a secondary DNN for enhanced accuracy. Lastly, the recommendation engine provides actionable suggestions for controlling home systems to improve energy efficiency while maintaining occupant comfort.