Research On Differential Protection Of Transformers Based On Artificial Neural Network

A large-scale power transformer is one of the most important equipment of a power system, whose operating states will affect the security and stability of the power system directly. Once failures occur, a serious harm will be for the power system and users. Therefore, it has an important practical significance to monitor transformer operating states effectively and to diagnose and predict transformer faults. The traditional transformer condition identification processes may fail in a practical diagnosing process, and it can not tell the exact faults of transformers. Hence, a further exploration of new theories, which can identify field flows and inernal fault currents of a transformer rapidly and precisely, is quite necessary for the improvement of the differential protection of transformers.In recent years, with the development of the intelligent control technology, it has been applied into the differential protection of transformers. In order to compare the performance of different neural networks in the differential protection of a transformer, two differential protection methods based on the probabilistic neural network and the GA-BP network which can identify operating states of transformers are proposed in the paper based on simulating model of a transformer and data preprocessing. Network structures and principles of two networks are introduced and affects on training effects and generanizing abilities of a network with different parameters are described in detail by using MATLAB programming language.For a better analysis of intelligence thechnologies in applications of transformer differential protections, a special kind of neural network-SVM is also used in the experiment, which uses the multi-classification ability of support vector machine to establish a new differential protection for operating state identification of transformers.Meantime, according to the comparation of different simulation results, a propor experimental design can be achieved. A helpful attempt and exploration is carred on for SVM in the differential protection of transformers in the paper.