关键词: 门控循环单元神经网络, 极限学习机, 灰狼优化算法, 电力线路, 故障诊断, 机器学习

Abstract:

To improve the detection precision and classification accuracy of power line faults， a power line fault diagnosis system based on machine learning is designed and implemented， whose core modules are Gated Recurrent Unit （GRU）neural network—A classical algorithm of machine learning， and Kernel Extreme Learning Machine（KELM）. First，GRU is used to separate fault data from normal data accurately in electrical fault diagnosis. Then，the kernel parameters and penalty factors of KELM are optimized by Grey Wolf Optimization（GWO）. The KELM with optimal parameters can successfully distinguish different types of faults. Proven by experimental data， GRU's accuracy in dataset classification is as high as 98%，and the KELM with the optimal parameters has an accuracy of 99%. Comparing the accuracies of the algorithms obtained by Simulated Annealing（SA），the superiority of GWO can be confirmed. The voltage and current data are visualized，presenting the dataset concisely and intuitively. This article provides a practical and effective method for diagnosing power line faults.

Key words: gated recurrent unit, extreme learning machine, grey wolf optimizer, power lines, fault diagnosis, machine learning