Outliers detection method using clustering in buildings data

Usman Habib, Gerhard Zucker, Max Blöchle, Florian Judex, Jan Haase

Publikation: Beitrag in Buch oder TagungsbandBeitrag in Tagungsband ohne Präsentation

Abstract

To achieve energy efficiency in buildings, a lot of raw data is recorded, during the operation of buildings. This recorded raw data is further used for the analysis of the performance of buildings and its different components e.g. Heating, Ventilation and Air-Conditioning (HVAC). To save time and energy it is required to ensure resilience of the data by detecting and replacing outliers (i.e. data samples that are not plausible) in the data before detailed analysis. This paper discusses the steps involved for detecting outliers in the data obtained from absorption chiller using their On/Off state information. It also proposes a method for automatic detection of On/Off and/or Missing Data status of the chiller. The technique uses two layer K-Means clustering for detecting On/Off as well as Missing Data state of the chiller. After automatic detection of the chiller On/Off cycle, a method for outlier detection is proposed using Z-Score normalization based on the On/Off cycle state of chillers and clustering outliers by Expectation Maximization clustering algorithm. Moreover, the results of filling the missing values with regression and linear interpolation for short and long periods are elaborated. All proposed methods are applied to real building data and the results are discussed.
OriginalspracheEnglisch
TitelIECON 2015 - 41st Annual Conference of the IEEE 2015
Seitenumfang7
PublikationsstatusVeröffentlicht - 2015

Research Field

  • Ehemaliges Research Field - Energy

Schlagwörter

  • adsorption chillers; physical rules; K-Means Clustering Algorithm; Outliers; Z-Score Normalization; Expectation Maximization Clustering Algorithm (EM); Heating
  • Ventilation and Air-Conditioning(HVAC); Fault detection and diagnosis (FDD);

Fingerprint

Untersuchen Sie die Forschungsthemen von „Outliers detection method using clustering in buildings data“. Zusammen bilden sie einen einzigartigen Fingerprint.

Diese Publikation zitieren