DEEP NEURAL NETWORK BASED ESTIMATION FOR THERMAL COMFORT INDEX

Abstract

Thermal Comfort indicates human interpretation of comfort level of an environment. Predicting Thermal Comfort for a certain future date can have several applications. Predictive Mean Vote (PMV) is one of the most used measure to express thermal comfort index, both indoor and outdoor. Many of the parameters involved are needed to be synthesized which adds to the complexity of it. Many techniques and algorithms to estimate it using only some of the parameters involved have been proposed till date aiming to improve the accuracy. Fuzzy Neural Network (FNN)s have been particularly successful in this scenario generating suitable sets of rules. Improving the accuracy a step further while choosing optimized number of parameters contribute to smoother and expanded applications. Convolutional Neural Network (CNN) is an essential Deep Neural Network (DNN) technique. It is primarily used shrink or convolve large data into smaller versions by keeping essential details intact. These smaller versions are used to classify (or in some case estimate using regression) using softmax layers or ReLU layers. In this work, focus was on combining modified FNN with suitable layers of CNN and/or traditional neural network to estimate PMV by regression with greater accuracy.