An Improved Neural Network Algorithm and its Application in Fault Diagnosis

Abstract:

Aiming at BP neural network algorithms limitation such as falling into local minimum easily and low convergence speed, an improved BP algorithm with two times adaptive adjust of training parameters (TA-BP algorithm) was proposed. Besides the adaptive adjust of training rate and momentum factor, this algorithm can gain appropriate permitted convergence error by adaptive adjust in the course of training. TA-BP algorithm was applied in fault diagnosis of power transformer. A fault diagnosis model for power transformer was founded based on neural network. The illustrational results show that this algorithm is better than traditional BP algorithm in both convergence speed and precision. We can realize a fast and accurate diagnosis for power transformer fault by this algorithm.

• Periodical: Advanced Materials Research (Volumes 765-767)
• Pages: 2355-2358
• DOI: https://doi.org/10.4028/www.scientific.net/AMR.765-767.2355
• Online since: September 2013
• Authors: Tai Shan Yan, Guan Qi Guo, Wu Li, Wei He
• Keywords: Adaptive Adjust of Training Parameters, BP Algorithm, Fault Diagnosis, Neural Network (NN), Power Transformer

References

[1] ZHANG Yue－qin, LIU Xiang, SUN Xian-yang. An Imporved Algorithm of BP Neural Network and Its Application. Computer Technology and Development, 2012, 22(8): 163-166.

[2] Li Ming, Yi Shen-min, Yang Shu-zi. Research on Neural Network's Adaptive Training. Systems Engineering and Electronics, 2007, 5(3): 29-36.

[3] Gao Da-qi. Research on the architecture of forward three-layer neural network for linear basis function. Journal of Computers, 2008, 2l(1): 80-85.

[4] Wang Wen-jian. Anefficient training algorithm of input driving BP network. Systems Engineering and Electronics, 2000, 20(1): 99-101.

[5] Bai Lian-ping, Chen Xiu-zhen. An adaptive BP neural network algorithm with high training speed. Journal of System Simulation, 2001, 8(13)9: 7-98.

[6] Yam Y F，Chow W S． A weight initialization method for improving training speed in feed －forward neural network. Neu- rocomputing, 2000(30): 219-232.

DOI: 10.1016/s0925-2312(99)00127-7

[7] Hou Jian-Min, Research on transformer fault diagnosis based on neural network. Nanjing Meteorology Institute, Nanjing, (2003).

[8] Lan Guo-Liang, Guan Gen-Zhi. Study on Insulation On-line Monitoring Expert System in Substation Based on Neural Network. High Voltage Apparatus, Vol. 2, pp.99-101, (2006).