Modeling a Thermal Area for Energy Consumption Estimation using Artificial Neural Networks

Abstract

This document describes an Artificial Neural Network (ANN) approach to modeling a thermal area to predict energy consumption. We propose the architecture for the ANN and explore the optimal configuration which yields the best results in terms of accuracy. Additionally, we validate our approach by using 3 months' data from the Passive House from the campus of University “Politehnica” of Bucharest (UPB). We implemented neural networks with various thermal energy consumption modeling (black box, grey-box) to compare the results achieved between them, but also with state of the art purely mathematically models. The best results achieved a total error of only 1.87% or 4.3 RMSE - the model and optimization of the ANN are discussed in this paper