Abstract: With the rapid development of industry,a large number of equipment fault case maintenance records have been accumulated in the daily maintenance of power transmission and transformation equipment.Text matching technology mines cases with high similarity with the target fault case from a large number of fault case data,which has important reference value for on-site operation and inspection personnel to quickly judge and make maintenance decisions in case of new faults. Most current text matching methods calculate the similarity between texts by constructing convolutional neural networks（CNN）and long short term memory（LSTM）models,ignoring the potential deep semantic information in massive unlabeled text data.A new text matching model was constructed to transform the matching problem of similar cases into the binary classification problem of sentence pairs.The deep semantic features of sentence pairs were extracted by using the improved pre training language model.Moreover,the semantic similarity of sentence pairs was captured by accessing the classification model. Experiments show that the proposed method has higher matching accuracy than CNN and LSTM on similar fault case data.