Study on Fault Line Detection Based on Genetic Artificial Neural Network in Compensated Distribution System

The faulty line detection of single phase to earth fault in power system with neutral grounding via arc suppression coil has not been well solved. The commonly used single faulty line detection methods, such as wavelet transform method, the fifth harmonic current method and zero sequence current active components method, etc., can only process partial fault information, so their reliability of faulty line detection is not satisfied. By means of constructing both relative fault measurement function and confirmable fault measurement function the fault measurement function of each faulty line detection method is determined, then using genetic neural network the intelligent fusion of practical fault measurements of those faulty line detection methods is conducted, thereby the faulty line detection result with higher reliability can be obtained. Simulation results by EMTP show that the faulty line detection result by the proposed method is more precise and possesses stronger robustness

I. Introduction

The basic requirements for modern cities to MV network are as follows: high reliability of power supply; high standard of human safety; high standard of equipment safety; low amount of maintenance work; electromagnetic compatibility, coexistence with telecommunication system; reasonable overall technical and economical considerations, etc [1]. So it is very important that the fault line is selected quickly when a fault occurs on a line. But the problem on fault line detection of single-phase-to-earth fault has not been well settled for a long time [2]. A lot of works have been done and many fault line detection methods based on different principle have been brought forward [3]–[7]. But these results of fault line detection aren't reliable. One of main reasons is that some fault characteristics are clear and some are fuzzy when a single-phase-to-earth fault takes place. The other is that the interference has effect on some fault characteristic but has very little effect on others. So, the single fault line detection method based on a kind of fault characteristic often judge healthy lines as fault lines and judge fault lines as healthy lines in compensated distribution system, there are three current fault line detection methods: wavelet transform method, the fifth harmonic current method and zero sequence current active components method. However none of these methods can ensure that the result of fault line detection is correct to any fault type. To improve the accuracy of the fault line detection, complementarities among the techniques of line detection can be taken use of and multifold techniques of line detection be amalgamated in an intelligentized way. So the reliability of fault line detection is improved [8]– [11]. The paper proposes that the information fusion can be carried out by means of genetic neural networks.