This paper proposes a new machine learning approach for fault detection and classification tasks in electrical power transmission networks. This method exploits the temporal sequence of the power system's operational data and develops an ‘end to end’ model employing Long Short-Term Memory (LSTM) units working directly on the operational data instead of features. The temporal sequences are different in the case of normal and faulty conditions. End to end learning simplifies the decision-making process and eliminates the need for complex feature extraction process by learning directly from the labelled datasets. The method is rigorously tested for all types of faults, which are further subjected to a range of fault resistance, distance, loading conditions, system parameters and noise levels. The proposed method can also work under power swing conditions. The method is also tested on WSCC 9 bus system. The proposed method has shown fast response in terms of time performance and is resilient towards operational conditions.

本文提出了一种新的机器学习方法，用于电力传输网络中的故障检测和分类任务。该方法利用电力系统运行数据的时间序列，并使用长短期记忆 (LSTM) 单元直接处理运行数据而不是特征来开发“端到端”模型。在正常和故障条件的情况下，时间序列是不同的。端到端学习简化了决策过程，并通过直接从标记数据集中学习来消除复杂特征提取过程的需要。该方法针对所有类型的故障进行了严格测试，这些故障还受到一系列故障电阻、距离、负载条件、系统参数和噪声水平的影响。所提出的方法也可以在功率摆动条件下工作。该方法还在 WSCC 9 总线系统上进行了测试。所提出的方法在时间性能方面显示出快速响应，并且对操作条件具有弹性。