A power meter fault classification method and device based on transfer learning

Abstract

The application discloses a kind of based on transfer learning's electric energy meter fault classification method and device.Therein, method includes: collecting the historical fault data sample set of electric energy meter;Respectively traverse each fault category sample in historical fault data sample set, all samples under this fault category are as minority class sample set, the sample of remaining each fault category is as majority class sample set, generates multiple two-class data sets;According to the pre-trained transfer data selector and transfer task supervisor, respectively to multiple two-class data sets are iterated against, generates multiple transfer data sets;Respectively multiple transfer data sets are input in at least number class sample generation model, generates multiple balanced sample sets respectively training classifier, generates multiple fault category classifier;Real-time acquisition is measured fault data respectively input to multiple fault category classifiers, outputs multiple fault category probability, and selects maximum value in multiple fault category probability as the fault category of measured fault data.

Images