Biclustering of Smart Building Electric Energy Consumption Data

Abstract

Nowadays, smart buildings can collect data regarding the electric energy consumption, which can then be analyzed to gain insights or to predict or identify abnormal energy consumption. Numerous models are applied to face this problem but they are based on a global point of view and cannot detect local patterns of abnormal consumption. This work lies in the former option, as we propose a way to analyze energy consumption data from smart buildings. In particular, we use energy consumption data collected by various buildings over a five-year period. These data were analyzed to gain insight into the functioning of the considered buildings, with the aim of detecting anomalous situations, which could indicate that some energy usage policy should be changed or that there is a fault in the sensor network. In particular, we propose an approach based on biclustering, which allows obtaining subgroups of buildings that show a similar behaviour over a specific period of time. To the best of our knowledge, this is the first application of biclustering to energy consumption data analysis. Results confirm that the proposed approach can help policy makers in detecting irregular situations, which can provide hints on how to improve the efficiency of buildings.

Parallel Evolutionary Biclustering of Short-term Electric Energy Consumption

“…In [18], a biclustering method identifies the most suitable group of user friends in social network datasets. In [19], a multi-objective biclustering algorithm was applied for the first time to the time series of the electricity consumption in smart buildings.…”

“…The energy consumption data can be usefully treated as time series data over a period. These time-series data are extensively used in different applications such as science, engineering, finance, economics, communications, control, health care, government, among others [22,21]. Clustering time series, which can be defined as identifying the homogeneous groups of timeseries data based on their similarity [22], is an important technique that can provide knowledge from raw energy data [23].…”

“…These time-series data are extensively used in different applications such as science, engineering, finance, economics, communications, control, health care, government, among others [22,21]. Clustering time series, which can be defined as identifying the homogeneous groups of timeseries data based on their similarity [22], is an important technique that can provide knowledge from raw energy data [23]. Traditional clustering techniques are not always enough in all applications because these techniques usually only look for similarities in the entire time series.…”

“…Good customer segmentation can improve the performance of demand response programs [24]. Biclustering methods could fit to study concrete periods on energy consumption time series [22]. Biclustering consists of simultaneous partitioning of samples and their attributes (features) into subsets (classes).…”

“…Biclustering consists of simultaneous partitioning of samples and their attributes (features) into subsets (classes). Samples and features classified together are supposed to have high relevance to each other [19].…”