Publications and deliverables
|Profiling and pattern detection report|
The aim of this document is to present the results of the work related to the profiling and pattern detection performed within work package 3.
For the purpose of profiling and pattern detection, test data from existing residential and commercial buildings were collected and refined. This includes set points, indoor temperature, humidity, and CO2 concentration. The additional semantics were modeled in the ontology described in deliverable 2.1. Furthermore, the building context is also part of the ontology and linked to the created data services.
State-of-the-art clustering algorithms were analyzed and promising candidates were tested on collected datasets. Implemented pattern detection algorithms, such as occupancy and set point detection algorithms, are able to discover interesting features in time series. Correlation analysis is performed over collected daily feature data in order to determine the correlations between features. The findings are utilized during prediction model training for reducing the feature set. After consumption clusters are identified, other relevant data points are reduced to daily points to be used for prediction model training. The focus is on short-term energy consumption forecasting based on various available data (e.g., calendars, building location, and orientation) or other forecasted inputs (e.g., weather conditions, occupancy). This forecast is used by the semantic-aware energy optimization in work package 4.
Privacy concerns and usage of personal data in the context of smart buildings and smart metering were addressed. A selection of privacy enhancing techniques was made. Some of these techniques were also applied here for the data preparation process of the profiling and pattern detection.