| Pumping wells are commonly used in most oilfields in China.In the process of oil and gas extraction,the failure of the pumping well will lead to the reduction of oil recovery efficiency and cause economic losses.Fault diagnosis timely of pumping wells can help the group company reduce costs and increase efficiency.Moreover,there is a large amount of knowledge data other than the dynamometer diagram in the field of fault diagnosis of pumping wells.These knowledge data are severely fragmented,loose in organizational structure and limited in standardization.Therefore,it is becoming more and more important to study how to select valuable data for fault diagnosis from massive text data,and to construct a complete knowledge system for pumping well fault diagnosis.As a kind of semantic network,the knowledge graph has strong expressive ability,and can store the domain knowledge of pumping well fault diagnosis in a structured way.The main research content of this paper is the construction of knowledge graph for pumping well fault diagnosis and the realization of question and answer for fault diagnosis.The innovation of this thesis is reflected in the combination of knowledge mapping and fault diagnosis,structured storage of domain knowledge for pumping well fault diagnosis and aided judgment for fault diagnosis.(1)In this thesis,the ontology of pumping well fault diagnosis is first constructed.Firstly,the current domestic and foreign fault diagnosis technology development is summarized.Learn and analyze the field knowledge of pumping well faults,and create a more complete domain ontology library different from the general ontology library.Using ontology expression language OWL and ontology modeling tool protégé to establish domain ontology model of pumping well fault analysis.Among them,in the aspect of pumping well faults,the establishment of ontology first can provide the basis for the hierarchical structure of fault knowledge ontology and realize the repeated use of knowledge.(2)On the basis of ontology construction,compare the traditional information extraction model with the joint entity relationship extraction model.Firstly,unstructured text data is obtained from relevant books and papers,and semi-structured text data is obtained after corpus cleaning.In order to obtain the ontology and relations required to construct the knowledge graph from the relevant corpus of pumping wells,that is named entity recognition and relation extraction.The extraction process of the Bi LSTM-CRF algorithm and the relationship extraction algorithm based on the BERT model is to perform entity extraction first,and then perform relationship extraction,because it has two issues,a.An entity belongs to multiple triplets.b.Multiple entities have overlapping.Therefore,this model and the improved joint entity relationship extraction model are used to verify and compare the method through the constructed named entity recognition and relationship extraction data set in the field of pumping well management.The result is that the improved joint practice relationship extraction model has a better extraction effect.(3)After the joint entity relationship extraction is completed,the knowledge graph can be constructed,and then the visualization of the knowledge map can be completed and the operation of adding,deleting,and checking can be performed.Then,the fault diagnosis of the pumping unit well fault phenomenon and the inference of the fault-related cases can be performed by relying on the interactive platform,and the possible fault diagnosis and the recommendation of measures under the current phenomenon can be obtained.A fault diagnosis method based on knowledge mapping and inverted index is proposed for pumping unit wells.The method infers the three most likely fault results and solves the problem of knowledge graph retrieval results with little priority ranking and measure recommendation.In this section,the method is validated by a fault diagnosis example. |
