Error Model and Forecasting Method for Electronic Current Transformers Based on LSTM

Abstract

As an important metering apparatus at the trade settlement gate in intelligent substations, the operating error of electronic current transformers can have an important impact on the electric energy trade settlement, so it is necessary to predict the error state of electronic current transformers. In this paper, a Long Short-Term Memory (LSTM) neural network is used to construct an error prediction model for electronic current transformers, characterizing their errors in the form of multiple input variables and single output variables. In order to reduce the training scale of the LSTM neural network, the partitioning around medoid (PAM) clustering algorithm is used to cluster and analyze the input variables. The analysis results of the algorithm show that the prediction results of the ratio and phase errors meet the requirements of online monitoring and provide information on the change of the error state of electronic current transformers to prevent the risk of electricity trade settlement.