Nueva metodología para la detección de anomalías usando técnicas de minería de datos y Bootstrap. Caso de aplicación eficiencia energética.

Abstract

A new methodology (LOCI-BOOTSTRAP algorithm) is proposed for the detection of outliers in HVAC (heating, ventilation, and air conditioning) systems, through Local Correlation Integral (LOCI) algorithm and using Bootstrap techniques in order to obtain a rule considering the score distribution used by the LOCI method and to improve the classification of the observations. This methodology was applied to the case of the facilities of a clothing store located in Panama, where 24 daily readings were recorded for 434 days. In each reading, 15 variables measuring thermal comfort, air quality and energy efficiency were monitored. For algorithm training and evaluation, anomalous events recorded by HVAC system operators are considered. In the training stage, the parameters that best fit the data are estimated, creating an index using the LOCI method to obtain a score for each of the observations and study their distribution by applying Bootstrap techniques to perform the classifications. For the evaluation of the performance of the algorithm, cross validation is used and from these results it is compared with those obtained in previous studies.