Failure rate prediction of electrical meters based on weighted hierarchical Bayesian

Abstract

Accurate lifetime and failure rate predictive analysis of electrical meters is critical for establishing reliability requirements in the field of energy metering. The unavoidable outliers and the characteristics of small sample dataset have an adverse effect on the performance of the conventional prediction model. In this paper, we introduce a weighted hierarchical Bayesian (WHB) approach for failure rate prediction by combining different environmental characteristics. Firstly, the local outlier factor (LOF) approach is introduced to adjust the weights of the failure rate data, in which the potential outliers are assigned lower weight values. The weights of environmental attributes are determined by utilizing correlation analysis. Next, a robust threshold setting method is identified to assign different weights to the failure rate data based on Grubbs. Finally, the optimized prior selection is incorporated to set partial hyperparameters of the WHB. The proposed WHB is validated using a real sample data set of electrical meters. The experimental results show the effectiveness of the WHB compared with some typical data-based methods.