Conference Paper

Proceedings of Building Simulation 2019: 16th Conference of IBPSA

Linear Discriminant Analysis for Classification of a Large Virtual Smart Meter Data Set With Known Building Parameters

Adam Neale, Michaël Kummert, Michel Bernier
Polytechnique Montreal, Canada

DOI: https://doi.org/10.26868/25222708.2019.210568
Abstract: Linear discriminant analysis (LDA) classification is performed on a virtual smart meter (VSM) data set for 40 000 buildings. LDA is used to classify the VSM data according to known building characteristics. The classification accuracy is evaluated based on the number of features and the number of smart meter data profiles used for classification. Some building parameters require a large number of data profiles to distinguish the class categories accurately. In most cases, the classification accuracy reached 90% or higher using 5-fold crossvalidation. For example, the building location is well classified by LDA. However, some parameters such as building rotation and the building’s aspect ratio are not properly discerned by the classification model. The results presented in this paper provide some insight into the effectiveness of LDA to accurately classify building parameters using smart meter data. The paper also describes a general methodology that can be used to apply LDA classification to smart meter data.
Keywords: Building stock modeling, classification algorithms, discriminant analysis, smart meter data, inverse modeling
Pages: 3393 - 3400
Paper: