Analysis Of Power Equipment Defects Based On Text Data Mining

Various types of equipment in the power grid generate a large amount of data during production and operation,which can accurately and timely reflect the operating status of the equipment.Therefore,how to efficiently process and analyze data will be a top priority in ensuring the reliable operation of equipment,and also a key link in enhancing the digitization and intelligence of modern power grids.Among them,the text data of power equipment defects mainly records the defect status of the equipment during operation.Conducting data analysis on it is beneficial for timely understanding of the health status of the equipment.However,currently,the mining of defect text information in power grid enterprises is mostly done manually,which not only requires a large amount of work and takes a long time,but also makes it difficult to ensure the accuracy of information mining due to different manual experiences.Secondly,as unstructured data,its text structure is highly arbitrary,concealing key information present in the data,which is not conducive to rapid analysis and timely positioning of defect locations by operation and maintenance personnel.In response to the above situation,this thesis takes transformer defect text data as the research object,and analyzes power equipment defects from two aspects:classification of defect levels and accurate identification of defect information.The main work points are as follows:(1)As unstructured data,defective text cannot be directly recognized by the model and requires preprocessing.On the basis of analyzing the characteristics of defect text data,this thesis summarizes the power equipment defect dictionary and stop use vocabulary.And it carries out word segmentation,de stop words,vectorization representation and similar word processing.(2)Using a hybrid neural network based on attention mechanism,a defect level classification model for power equipment is constructed to achieve automatic classification of equipment defect levels.Firstly,the preprocessed data is input to CNN and BiLSTM to extract local and contextual features of the text,respectively;Then,the extracted information is fused using a fully connected layer;Finally,the Attention mechanism is used to allocate feature weights and enhance the impact of key features on classification performance.And compared with other models from multiple evaluation dimensions,the effectiveness of this model was verified.(3)A precise identification method for defect information based on improved semantic framework and dependency syntax analysis technology has been proposed.First,according to the power grid defect terminology standard and Chinese text preprocessing process,the initial semantic framework of defect text is constructed,and unstructured text data is converted into structured data;Then,dependency syntax analysis technology is introduced to analyze the internal structure of the text from the perspective of grammatical relationships,and construct a dependency syntax tree;On this basis,a text similarity calculation method based on dependency syntax tree is proposed,which combines semantic similarity and dependency structure similarity to match actual defect text with standard text;Finally,a tree matching algorithm based on dependency relationships was proposed to improve the quality of power equipment defect text data.By using specific cases,the output effects of each link of the model were demonstrated,demonstrating its feasibility and effectiveness.At the same time,the defect status of the transformer was analyzed based on the results.