The Remaining Useful Life Prediction of Smart Meter Based on Improved Gaussian Process Regression

The prediction of the remaining lifespan of smart meters holds significant importance for the healthy operation of modern power grids. In this study, we propose a Gaussian process regression model based on a fusion kernel structure to forecast the remaining lifespan of smart meters. Preliminary analysis of field-collected data during the operation of the smart meters is conducted using Spearman correlation analysis and significance testing. This analysis aids in selecting natural environmental factors with higher coefficients of correlation to measurement errors as inputs for the model, thereby reducing parameter dimensions. Furthermore, a data recovery algorithm based on the Fourier Series Approximation (FSA) is devised to address the issue of missing values in the samples. To capture the diversity of different input features, multiple kernel functions are integrated to effectively estimate the remaining lifespan of smart meters under various pressure conditions. The model hyperparameters are solved using Bayesian theory. Taking the smart meters in the Turpan region of Xinjiang as an example, experimental results indicate that after about 2009 days, the measurement error of Type I meters will reach its upper limit, indicating an inability to meet usage requirements. Similarly, after about 1541 days, Type II meters will also fail to meet usage standards, necessitating maintenance and repairs.