Research On Reliability Of Smart Energy Meter Based On Data Driven CK-GPR

The massive smart energy meters in the power grid are the nerve endings of ubiquitous power Internet.It is of great significance to ensure their safe and reliable operation under the actual on-site network operation conditions.Based on the power grid and driven by the data flow brought by the power flow,the application of new artificial intelligence methods can further improve the holographic perception ability of the smart grid in many aspects,such as operation,transmission,transformation,distribution and adjustment.The state evaluation and orderly replacement of smart energy meters,measurement anomaly detection and diagnosis,and on-line error monitoring of power metering devices have increasingly become hot spots.The basic error data is an important basis for the verification and replacement of smart energy meters.Combined with the field data of typical environment and trend evaluation algorithm,the research on the change characteristics and trend of the basic error data in the actual operating environment is of great significance to the reliability evaluation of smart energy meters.This thesis first describes the background of the research on the reliability of smart energy meters,introduces the research methods of predicting the life of devices based on data-driven,then analyses and summarizes the research status of reliable life prediction of smart energy meters at home and abroad,summarizes the characteristics and shortcomings of existing methods,introduces the Gauss process regression algorithm used in this paper,and introduces its application status.Secondly,the generation mechanism of basic error data is revealed through the principles of energy measurement and smart energy meter,and the effect of environmental stress on basic error is analyzed.To explore the true existence of degradation under the typical field operation conditions the accelerating factors such as temperature,humidity and air pressure under the actual field operation conditions in Xinjiang are calculated based on IEC62059 standard,while introducing the test configuration and the source of basic error data of the typical environmental test base in Turpan,Xinjiang.Based on the degeneration model of the random process,a probability life prediction model for smart energy meters based on the random degeneration track called CK-GPR is established.Based on the large amount of environmental data and basic error data produced by typical environmental field operation tests,the status dataset suitable for researching the lifetime of smart energy meters with high reliability is obtained by data filling,data downsampling and abnormal detected,and the influence degree of natural environmental stress on basic error data is analyzed by correlation analysis.A Bayesian cross-validation method is proposed to obtain a maximum synergy-minimum redundancy combination of stress input variables in multidimensional variable selection.To fuse the periodicity,trend and irregular variation of basic error data in typical environments,a Gaussian process regression model with combined kernel functions is established to track and predict the trend.Finally,the data from the field operation test in the typical environment of ”high dry heat” are analyzed and predicted using the algorithm named CK-GPR proposed in this thesis,and the separate prediction of smart energy meters of different manufacturers and the comprehensive prediction of smart energy meters of different manufacturers in the same region of Xinjiang are compared.The results show that,compared with other artificial intelligence models,the data-driven combined kernel Gaussian process regression model can predict the probability life distribution of smart energy meters under field operation conditions,and can provide guidance for the formulation of maintenance rotation strategies such as condition evaluation and dynamic cycle verification based on the prediction results.