Research and Application of Text Segmentation and Entity Data Extraction on Operation Ticket

The unstructured operation ticket text data generated during the operation and maintenance process of the power system has a large inventory, high value density, and strong timeliness. Studying the data information contained in it is of great significance for reducing the probability of operation accidents and achieving substation safety risk control. However, traditional unstructured processing models are difficult to effectively utilize it, which poses many challenges to the mining of operation ticket text data and leads to waste of operation ticket text data. At the same time, in terms of substation safety control, traditional video surveillance can only achieve post alarm and requires manual intervention, and combining video surveillance with operation ticket text mining poses certain difficulties. In response to the above issues, a method for extracting and mining operation ticket text data is proposed by combining relevant technologies of natural language processing. This method compares the segmentation effects of operation ticket text based on statistical and dictionary methods, proposes a pre-processing method for operation ticket text segmentation with dictionary method as the main method and statistical method as the auxiliary method, and then compares the operation ticket text represented by discrete representation and distributed representation. The distributed method is used to represent the operation ticket text, and the word vector is obtained through training. Finally, based on the analysis of the advantages and disadvantages of the long short-term memory network model and the conditional random field model, the BiLSTM CRF algorithm combining the two is used to achieve named entity recognition of operation ticket text, with an accuracy of over 90%. At the same time, the traditional entity relationship extraction method based on feature vectors is targeted. The recognition accuracy is low Unable to effectively utilize contextual information and other drawbacks, a BiLSTM based entity relation extraction method is proposed with an accuracy rate of 86%.