Abstract Electric power is increasingly being used in the globalized day-to-day and keeping the electric power system running is necessary. Insulators are important components of the electric power system. In case of failure in these components, there may be disconnections and, consequently, no electricity. Contaminated insulators can develop irreversible failures if they are not inspected. One equipment used for the inspection of the electric power system is the ultrasound, which generates an audible noise based on a time series that is used to identify possible failures. The time series forecast can be used for possible prediction of the development of failure. In this paper, a hybrid method that uses Wavelet Energy Coefficient (WEC) for feature extraction and Group Method of Data Handling (GMDH) for time series prediction is proposed, being defined as Wavelet GMDH. For comparison and validation of the proposed method, a benchmark is made with well-established algorithms such as Long Short-Term Memory (LSTM) and Adaptive Neuro-Fuzzy Inference System (ANFIS). For a fairer analysis, these algorithms are also evaluated based on the same data extraction with WEC. The proposed method proved to have good accuracy comparing with LSTM and ANFIS, and is much faster than the compared methods.

中文翻译：

---

电力绝缘子故障预测数据处理的小波群方法

摘要 电力在全球化的日常生活中得到越来越多的使用，保持电力系统正常运行是必要的。绝缘体是电力系统的重要组成部分。如果这些组件出现故障，可能会断开连接，从而无法通电。如果不进行检查，被污染的绝缘体可能会发生不可逆转的故障。用于电力系统检查的一种设备是超声波，它根据时间序列产生可听噪声，用于识别可能的故障。时间序列预测可用于故障发展的可能预测。在本文中，提出了一种使用小波能量系数（WEC）进行特征提取和数据处理组方法（GMDH）进行时间序列预测的混合方法，被定义为小波 GMDH。为了比较和验证所提出的方法，使用成熟的算法（例如长短期记忆（LSTM）和自适应神经模糊推理系统（ANFIS））进行基准测试。为了更公平的分析，这些算法也基于与 WEC 相同的数据提取进行评估。与LSTM和ANFIS相比，所提出的方法被证明具有良好的准确性，并且比比较方法快得多。