Research On Downhole Operation Question Answering System Based On Cognitive Graph

At present,"artificial intelligence" has begun to move towards the laboratory gradually,and began to be commercialized and applied on a large scale.Smart oil fields based on hightech technologies such as artificial intelligence,Internet of things and big data are the development direction of the oil field at present.Petroleum related enterprises and organizations pay special attention to it.Petroleum enterprises make full use of the advantages of new technology and widely use it in oil and gas exploitation,staff training,downhole operation and so on.On this basis,the cognitive graph came into being.Cognitive graph is a new cognitive engine integrating cognitive psychology,brain science and other disciplines.It uses new theories and methods such as cognitive representation,extraction,reasoning and calculation to form an understandable third-generation artificial intelligence engine.Downhole operation is a knowledge intensive specialty integrating geology,oil production,drilling,tools and other disciplines,in which minor repair operation occupies a very important part and requires extremely rich knowledge.The training of minor repair operators is an important part of minor repair operation,and the training quality directly affects the level of minor repair operation.Strengthening the training of minor repair operators is an important factor to improve the quality of minor repair operations.At present,the knowledge in minor repair operation is relatively scattered,and there is no unified knowledge system,which brings great difficulty to the training work.Q&A systems are mostly trained based on labeled text,that is,the number of tags is limited,and there is nothing to do with the knowledge base of large-scale complex annotation.And can only answer the questions of a single entity,not the questions that need to identify multiple entities.Based on this situation,the conceptual knowledge system and relationship are modeled,and the downhole operation event knowledge expression model is constructed.Then,according to the constructed event knowledge graph mode layer,the downhole operation minor repair domain knowledge system is constructed by extracting the knowledge from the downhole operation minor repair domain technical report.For the question and answer system,the dual process theory is introduced to construct the downhole operation question and answer system.A question answering system in the field of minor repair of downhole operation is developed.Using this research,the difficult problem of downhole operation minor repair staff training is solved.The training and work efficiency of downhole operation minor repair personnel are improved.So that they can quickly obtain operational knowledge and realize knowledge sharing.The research focuses on:Firstly,for the event knowledge graph in the field of downhole operation minor repair,this thesis analyzes the knowledge in the field of downhole operation minor repair,obtains the relevant knowledge of downhole operation through standard specifications,professional books,news information,data and other channels,and explains it clearly and in detail.The modeling method of downhole operation knowledge base based on object state is established,which takes the events in the field of downhole operation minor repair as the core.The model of conceptual knowledge system and relationship is established.The technical report in the field of downhole operation minor repair is extracted into knowledge,and the event knowledge graph in the field of downhole operation minor repair is constructed.Secondly,aiming at the problem that the number of question answering systems in the field of downhole operation is small and can not carry out multi-entity question answering,this thesis introduces the concept of cognitive graph and studies it from the perspective of cognition.Through the research of double process theory,this thesis establishes the downhole operation question answering system,divides the problems input by users into simple problems and complex problems,and gives the corresponding solutions.For simple problems,Bert algorithm is used to solve the problem of entity recognition,and Bert plus logistic regression algorithm is used to solve the problem of entity screening within two hops.The model architecture of Bert and logistic regression algorithm is designed.For complex problems,formulate corresponding reasoning rules to solve the problems that cannot be answered by multiple entities.Finally,the accuracy of the query results of user input questions is 94.5%.The question answering system makes up for the blank of the question answering system in the field of downhole operation.Finally,through the above research,a downhole operation Q&A system based on cognitive graph is developed.The visual architecture of each module is designed.Provide corresponding knowledge resources.It solves the actual question and answer needs of users,and designs a visual interface for displaying and viewing relevant knowledge.Provide employees with appropriate answers and realize intelligent Q&A.In addition,the system is intuitively displayed in the form of graph nodes,so that employees can clearly understand all knowledge.