基于预训练语言模型的电力领域设备缺陷检测

电力设备缺陷种类繁多，部分缺陷会引发设备故障，及时检测电力设备存在的缺陷是防止发生设备故障的重要手段。设备缺陷检测旨在从文本中识别触发词并且将文本划分对应的设备缺陷类型。针对电力领域缺陷数据集标注不足，以及由于文本中包含大量专业术语造成语义理解难等问题，研究基于深度学习的设备缺陷检测方法，设计电力领域设备缺陷检测预训练语言模型，利用事件三元组知识。文中，构建一个电力设备缺陷检测数据集，在模型进行缺陷检测任务之前，通过事件三元组预训练的方式提高语言模型的表征能力。实验表明，基于现场设备案例数据经过预训练的模型在缺陷检测任务上具有更好的表现效果，可以有效实现对电力领域缺陷报告文本的缺陷检测。

Pre-trained defect event detection in electric power field

There are many kinds of defects in power equipment, some of which will lead to equipment fault. Defects detection timely of in power equipment is an essential means to prevent equipment. It aims to identify trigger words from the text and divide the text into corresponding device defect types. Aiming at the problems of insufficient annotation of defect data set and difficulty in semantic understanding due to numerous professional terms in the text, device defect detection methods based on deep learning are studied. We design a pre-training language model for device defect detection in the power field, utilizing event triplet knowledge. In addition, we construct a power equipment defect detection data set. The representation ability of the language model is improved by event triplet pre-training before the defect detection task. Experimental results show that the pre-trained model performs better in the event detection task and can effectively realize the defect detection.