Fault detection based on Bayesian network and missing data imputation for building energy systems. (5th January 2021)

• Record Type: Journal Article
• Authors: Wang, Zhanwei; Wang, Lin; Tan, Yingying; Yuan, Junfei
• Is Part Of: Applied thermal engineering. Volume 182(2021)
• Publication Date: 2021-01-05
• Subjects: Bayesian network -- Building energy system -- Energy efficiency -- Fault detection -- Missing data imputation; Heat engineering -- Periodicals; Heating -- Equipment and supplies -- Periodicals; Periodicals; 621.40205
• DOI: 10.1016/j.applthermaleng.2020.116051
• Languages: English
• ISSNs: 1359-4311

Abstract

Highlights: FD for building energy systems in the presence of missing data is discussed. An enhanced FD method based on combination of EM and BN is proposed. The developed model reduces complexity and improves computational efficiency. FD performance is improved vastly, especially under missing multivariate data. The proposed EM-BN method is proven to be effective by using experimental data. Abstract: Fault detection (FD) for building energy systems in the presence of missing data is discussed in this paper. The purpose is to propose an enhanced FD method with higher accuracies under both missing univariate data and multivariate data. The solution is to develop an effective missing data imputation model with low complexity and high computational efficiency to impute the missing values. An FD method based on expectation–maximization (EM) algorithm and Bayesian network (BN), which is called EM-BN method, is presented. The EM algorithm is utilized to impute the missing data, thus to keep the information hidden by the missing data. The imputed complete data sets are addressed with maximum likelihood estimation to obtain the parameters of BN. The presented method is evaluated using the experimental data. Test results show that (i) compared with the method discarding the missing data, the proposed EM-BN method significantly improves the FD accuracies from 55.9% to 96.3% at most (for refrigerant overcharge at severity level 3); (ii) compared with the method using back-propagation neural network (BPNN) to impute the missing data, the proposed EM-BN method significantly reduces the model complexity and improves computational efficiency, particularly under the missing multivariate data.