Paper Conference

Proceedings of Building Simulation 2021: 17th Conference of IBPSA


Data imputation with a diagnostic purpose in building: Application to an office setting

Houda Najeh, Stéphane Ploix
G-SCOP laboratory, University of Grenoble Alpes,France

Abstract: In a building system, diagnoses are calculated from symptoms resulting from a set of tests; each defined by a bunch of data, behavioral constraints and a set of possible explanations in case of anomaly. To make a test, data are required from different sensors. However, missing data is a common issue in buildings. Because a detection test relies on several variables and that all the data should be available along a given time horizon, performing tests could become an issue. In the literature, many techniques have been proposed to impute missing data. However, detection tests could detect non-existing anomalies with usual types of imputation: it is indeed necessary that the imputed data do not generate alarms. It is all the more important for diagnosis from first principle approach where detection of anomalies has a strong impact on the diagnosis. This paper discusses the issue of imputation with diagnostic purposes i.e generating data satisfying test constraints. The proposed imputation algorithm is based on a genetic algorithm which search iteratively for imputed data satisfying behavioral constraints while preserving as much as possible known statistical properties. The efficiency of the method is evaluated on measurements obtained from a real application: an office at G-SCOP laboratory with a large number of sensors and missing data. Key-words: imputation, missing values, genetic algorithm, building system, diagnosis, false alarms.
Keywords: Imputation, missing values, genetic algorithm, building system, diagnosis
Pages: 2396 - 2403