| Long-distance water transmission project has many river crossings,long routes,different geological environments along the route,and wide distribution of risk factors during operation.In order to find all kinds of safety hazards in time,the project management unit arranges daily inspection staff to inspect along the route,so a large amount of inspection text data will be generated during the daily operation of the water transmission project.As most of the inspectors are villagers from villages along the water transmission line,they lack knowledge of water conservancy and often have repeated expressions of problems,colloquial descriptions of problems and unclear expressions of problems in the process of recording inspection texts,which makes it impossible for the inspection texts of water transmission projects to be managed in a structured way in project management.At the same time,because the inspection text relies on the manual classification of management personnel,inefficient and prone to subjective problem classification errors.If the problems reflected in the inspection text are not judged in time,the efficiency of the water transmission project may be affected by the danger that cannot be handled in time,and even threaten the safety of people’s property along the coast.In summary,in view of the problems of structured management of water transmission project inspection text and the current situation of low efficiency due to the need for manual judgment of the serious level of inspection problems,we now conduct research on named entity identification and text classification.The research contents are as follows:(1)A combined word vector BiLSTM-CRF water transmission project inspection text named entity recognition model is proposed,which combines word vector and entity recognition and uses BiLSTM model to learn semantic features of inspection text.Since there is no open source thesaurus for water resources,a new word separation model is trained for water resources proper nouns in inspection text to improve the recognition rate of water resources proper nouns,and then Word2 Vec is used to train word vector and word vector fusion as input to the named entity recognition model.Finally,the state transfer matrix of CRF is used to solve the logical sequential problem of entity labeling output.The experimental results show that the word vector BiLSTMCRF model can correctly identify the target entities from the water transmission project inspection text,so as to structure the water transmission project inspection text.(2)A BERT-BiLSTM inspection text classification and grading model is proposed.The BERT model is used to generate character-level feature vectors,the BiLSTM model is used to learn the semantic features of inspection texts,and finally the Softmax classifier is used to classify and quickly determine the severity level to which the inspection texts belong.The experiments show that the model can classify the inspection text of water transmission project quickly and intelligently.(3)To verify whether the model proposed in the article can bring practical application value to the management and operation and maintenance personnel in the water transmission project work scenario.We propose to design a prototype system for water transmission project inspection text to display project inspection data,and at the same time,to identify named entities and classify text for inspection data.In this way,it helps managers to handle engineering inspection texts more easily.This research work can help the management and operation and maintenance personnel to structure the large amount of engineering inspection text data,and according to the inspection text content intelligent judgment of the serious level of the problem,can greatly reduce the work pressure of the management personnel,effectively improve the efficiency of the project operation safety hidden danger is found,help the management personnel to make reasonable disposal faster. |
PDF Full Text Request