Intelligent Algorithms For Learning Occupants' Thermal Comfort And Their Application On HVAC System

Heating,ventilation and air conditioning(HVAC)system is designed as a facility to guar-entee the thermal comfort of indoor occupants.Traditional operational settings of HVAC sys-tems(e.g.26~?C),however,are neither comfort-driven nor energy efficient.Furthermore,the concept of thermal comfort is multi-variable,strong-nonlinear and personalized,which means that it correlates not only to indoor and outdoor air temperature but also to occupants'gender,age,clothes,location and rate of activity.That's why even the operation of HVAC system based on PMV-PPD index cannot reach the proportion of 80%thermal satisfaction.What's more,it highlights that we should figure out how to build the zone level energy consumption model of HVAC system and extend the proposed method to the existing VAV control loop,which has been widely used all over the world.Based on the basic concept of thermal comfort and the brief principle of HVAC system,this dissertation focuses on learning personalized thermal comfort profiles based on data col-lected from occupants'votes and improving the energy efficiency of HVAC system.The main contributions of this paper are as follows:Firstly,a well-designed human-machine interface,incorporating occupants'thermal sensa-tion,preference and expectation,is presented to collect their thermal vote under different indoor thermal conditions.An approach called dynamic evolving fuzzy inference system is proposed to predict the expected indoor air temperature under various personalized thermal perception index.Considering the fact that there should be more than one person in one shared thermal zone under most circumstances,we utilize the machine learning model named softmax regression to convert the thermal vote of occupants into a probability distribution under different indoor temperature based on the pre-processed data.After that,we propose a method to formulate the model of multi-occupants'thermal profile to obtain the range of most comfortable temperature.And we build an optimization problem with the zone-level HVAC system energy consumption function as the object function,and the learned comfortable temperature range as constraints.The main contribution is that we transform the traditional multi-object optimization problem for improving both thermal comfort and energy efficiency into a single-object optimization prob-lem.Last and not the least,in order to verify the feasibility of the proposed method,we adopted both online simulation and offline implementation.A professional building energy analysis tool called EnergyPlus is utilized to co-simulate with matlab using the building settings of my lab.Furthermore,we bridge the gap between Matlab and BACnet instruments of VAV system through the middleware named BCVTB.The data collected during the operation and simulation demonstrates the satisfying performance of the propoed method.