A meta-analysis on the use of statistical machine learning tools for accurate quantitative estimation of the energy performance of residential buildings

Shahab Aldin Siahposht¹

1) PhD in Architectural Engineering, Islamic Azad University, South Tehran Branch, Tehran, Iran,

IOI Code : XHFC-KCKCF
Publication : 6th.International Conference & 7th.National Conference on Civil Engineering, Architecture, Art and Urban Design(7iccacs.ir)
Abstract :
    Machine learning (in English: Machine learning) or ML (abbreviation ML) is the study of algorithms and statistical models used by computer systems that use patterns and inferences to perform tasks instead of using clear instructions. Machine learning is the science that makes computers learn about a specific subject without the need for an explicit program. As a subset of artificial intelligence, machine learning algorithms create a mathematical model based on sample data or training data in order to make predictions or make decisions without overt programming. Buildings consume 40% of primary energy in the world and emit approximately one third of greenhouse gases (GHG). Forecasting the amount of energy consumption in the future is very important for the optimal management of energy supply and consumption in this area. In addition to preparation and planning for energy demand management, factors affecting energy consumption can be identified and architectural solutions can be used to reduce these factors in residential buildings. Energy consumption forecasting is very challenging because it often follows complex non-linear patterns and also has various influencing variables from building physics details, occupant behavior and heating and cooling equipment, hence traditional statistical forecasting techniques to achieve Accurate energy forecasts perform very poorly. In recent years, the power and accuracy of complex calculations and real data analysis have been improved using machine learning. Many machine learning techniques are capable of analyzing non-linear patterns of data, which provides a suitable model for predicting energy consumption. We developed a statistical machine learning framework to investigate the effect of eight input variables (relative compression, surface area, wall surface, ceiling area, overall height, orientation, wall area, wall area distribution) on two output variables, i.e. heating load. (HL) and cooling load (CL) of residential buildings have been developed. We systematically examine the strength of the relationship between each input variable and each of the output variables using a variety of classical and non-parametric statistical analysis tools in order to identify highly correlated input variables. Then, we compare a classical linear regression approach against a powerful state of the art non-parametric non-linear method, Random Forest, to estimate HL and CL. Extensive simulations on 768 diverse residential buildings show that we can predict HL and CL with low mean absolute error deviation from ground verification established using Ecotect (0.51 and 1.42, respectively). The results of this study support the possibility of using machine learning tools to estimate building parameters as a suitable and accurate method, as long as the questions asked are related to the data that was actually used in the first place to train the mathematical model.

Keywords : smart meter building energy assessment heating load cooling load non parametric statistics statistical machine learning