| The distribution of sedimentary facies is closely related to the formation of oil and gas storage structures.Therefore,as a specific reflection unit of sedimentary facies,the identification of seismic facies is an important basis for determining the potential for oil and gas exploration.The method of manual interpretation is closely related to the quality of results and the professional level of explanatory experts,and its efficiency is not high.There are also certain limitations in the research.Faced with the massive amount of data resources in the field of geosciences,manual interpretation has become inadequate.With the rapid development of the computer field,deep learning technology has also been applied in seismic facies analysis research,greatly promoting the efficiency of oil and gas exploration.However,facing the challenge of improving recognition accuracy in situations where exploration requirements are not high.In tasks with higher precision requirements,it is difficult to identify seismic microfacies due to weak data features.To solve the above two challenges,it is necessary to rely on expert guidance,and the development of knowledge graphs quantifies expert knowledge and combines it with deep learning.This article is based on deep learning networks in seismic facies recognition,and gradually utilizes knowledge graphs to improve existing models and improve the accuracy of recognition results.In this thesis,a set of seismic facies knowledge graph are constructed.Based on the knowledge graph construction technology,combined with the prior knowledge of geological experts and the concepts,data,and regular information provided by existing actual data,the pattern layer and data layer are sequentially constructed in a top-down manner to achieve rule reasoning and ultimately form a set of seismic facies knowledge graph system.A seismic facies recognition method based on knowledge graph embedding,which initially utilizes the knowledge graph data layer to effectively improve the accuracy of seismic facies recognition,is proposed in the article.In order to break through the barriers of existing deep learning technology to seismic facies recognition and solve the problems of edge ambiguity,random noise,and unreasonable distribution in seismic facies recognition,this article introduces point entity constraints and sub surface entities from the constructed knowledge graph instantiation data layer,effectively mitigating the noise appearing in the results,and improves the accuracy of the edge of the recognition results by controlling the integrity of the target seismic facies.At the same time,an entity relationship loss constraint method is also proposed,which uses the knowledge graph to obtain the dependency relationships of relevant seismic facies.From the perspective of geological knowledge,it further controls the network’s distribution learning of seismic facies,enhancing the interpretability and reliability of the identification results.In this article,a knowledge graph guided seismic microfacies identification method is raised,which further utilizes the constructed knowledge graph data layer and inference rules to solve the problem of insufficient recognition accuracy of seismic microfacies under small sample conditions.Because of the sophistication of seismic microfacies interpretation of seismic data,it requires a large amount of manpower.This method fully utilizes the original information and geoscientific inference information of geological data under small sample conditions.The problem of misclassification caused by weak reflection characteristics of seismic microfacies and category balance has been preliminarily resolved,and the accuracy of seismic facies identification has also been effectively improved,and good results have been achieved in actual work area data.The method proposed in this article fully integrates geological knowledge with geological data,breaks through the improvement of accuracy in general seismic facies recognition using deep learning technology,effectively solves the problem of limited network ability to extract seismic reflection features in seismic microfacies recognition,and realizes the progressive utilization process of seismic facies knowledge graph from data layer to rule inference.The research method in this article achieves the target effect on seismic facies identification results on actual work area data and public data,which strongly demonstrates the feasibility and effectiveness of the method in practical application. |
