Estimation of primary current in saturated current transformer using flexible neural network

Abstract

Current transformer saturation causes a distorted secondary current, which produces many problems in applications of protection and measurements. This paper proposes a compensating algorithm to calculate the accurate primary current by measuring the secondary current in real time. A new flexible neural network with the activation function of two changeable parameters, where the flexible neurons will flexibly change the shape of each unit to adapt its role in the learning process, is used to realize estimation of magnetizing current. This greatly simplifies the network with fewer neurons and reduces iterative learning epochs. The primary current is obtained by adding the estimated magnetizing current to the secondary current. The estimator can ensure high precision regardless of the remanent flux level and load characteristics. Data from a 900:5A current transformer are applied to train the neural network. The simulated operation in the saturated current transformer EMTP model verifies the proposed algorithm.