| The construction of three-dimensional models of subsurface geological structures(referred to as structural modeling)is one of the most important basic tasks in the exploration and development of oil and gas resources,which can provide support for depth-domain seismic imaging,lithology interpretation,reservoir modeling,and so on,and directly affects the formulation of oil and gas development plans such as reserves calculations,well emplacement,and so on.In recent years,due to the demand for exploration of complex structural oil and gas reservoirs,structural modeling as its supporting technology has become one of the research hotspots of common concern in the fields of geoinformation science and geophysical exploration.Traditional structural modeling methods take the geometric information of geological surfaces(horizons and faults)obtained from seismic interpretation as the main basic data,and build the model through a purely data-driven approach.However,in the context of the increasing difficulty of oil and gas reservoir exploration year by year,the acquisition cost and uncertainty of seismic interpretation have also increased,and the pure data-driven modeling approach has been difficult to meet the application requirements.At this stage,a set of intelligent structural modeling techniques combining data-driven and knowledgedriven are urgently needed to improve the modeling capability.To realize the intelligent structural modeling,firstly,it is necessary to solve two basic artificial intelligence problems: semantic characterization and knowledge graph construction under uncertainty,and secondly,it is necessary to deeply integrate information technology innovation with applications in geology.Therefore,the challenges faced by the intelligent structural modeling can be summarized in the following four aspects:(1)the lack of a method to characterize the semantics of structural models.The existing geological data models are unable to characterize the large amount of expertise generated in the process of seismic interpretation;(2)the lack of a method to construct knowledge graphs of structural models.The existing technical system of knowledge graph construction is not applicable to seismic interpretation;(3)the lack of an integration mechanism between the expertise and modeling process.The existing modeling methods rely on data-driven and are difficult to take advantage of semantic information;(4)the lack of quality control methods for knowledge graph under uncertainty.The existing knowledge graph construction methods are difficult to build a complete knowledge graph under the uncertainty of both basic data and prior knowledge.This dissertation has conducted research in semantic representation and knowledge graph construction,and applied it to intelligent structural modeling with knowledge graph as the core technical tool.The main innovations and research contents of the dissertation include the following four aspects:(1)A semantic characterization method for structural models is proposed: existing geologic data models have been able to accurately describe the morphological information of structural elements(the constituent units of structural models),but lack the characterization of geological features,model topology and tectonic formation processes.To address this problem,this dissertation proposes a semantic characterization method for structural models,which provides a theoretical basis for formalizing the characterization of expertise.The semantic characterization method gives a spatial description of the structural elements from the perspective of geometric topological relations and geological contact relations,and also reveals the physico-logical relationships between structural elements from the perspective of structural evolution process.The ontological modeling of the semantic characterization framework establishes the schema layer of the structural model knowledge graph.(2)A knowledge graph construction method for seismic interpretation is proposed:the existing knowledge graph construction technology mainly serves the semantic processing of natural language and is not suitable for semantic extraction of unstructured spatial geometric data.In response to this issue,this dissertation proposes an automatic construction method for knowledge graph of structural models using seismic interpretation as the source data.The method gradually obtains the semantic relationships between the strcutural elements through perception and knowledge reasoning,and establishes a data layer of strcutural model knowledge graph.On this basis,this dissertation further inferred the structural evolution process based on the semantic relationships between structural elements.(3)A knowledge graph-guided strcutural modeling method is proposed: in the case of significant uncertainty in seismic interpretation,the existing methods try to transform additional data such as outcrops,boreholes,geological maps,etc.into modeling inputs by external means,but the information that can be transformed has limitations,and the modeling effect is still unsatisfactory when facing complex geological background.To address this problem,this dissertation proposes a modeling process guided by the knowledge graph,and the core method is to transform the relationship information of the strcutural elements contained in the knowledge graph into the boundary constraints of the geological surface reconstruction,which fundamentally avoids the instability caused by the interpolation algorithm of the surface reconstruction.(4)A knowledge graph modification and dynamic modeling method based on human-machine knowledge interaction is proposed: in the case of uncertainty in both basic data and prior knowledge,simply using any existing knowledge graph construction route(bottom-up and top-down)cannot obtain a complete knowledge graph.In response to this issue,this dissertation proposes a knowledge graph modification method based on human-machine knowledge interaction,which is combined with research content(2)to form a bottom-up and top-down iterative construction method for knowledge graphs.This dissertation proposes a knowledge graph reliability detector,which can automatically prompt errors and new knowledge in the knowledge graph through graph matching,helping modelers modify seismic interpretation and prior knowledge,and ultimately obtaining a complete knowledge graph.In addition,considering that expertise may constantly change during oil and gas exploration and development,the model needs to be updated accordingly.This dissertation further proposes a dynamic structural modeling method based on knowledge graph.The knowledge graph can identify the structural elements affected by expertise changes from a logical correlation level,and locally reconstruct the model within the most reasonable range.In summary,this dissertation provides a set of knowledge and data driven intelligent structural modeling solutions,which innovate in the basic methods and applications of artificial intelligence.The experiments show that the method proposed has good stability in modeling complex structures under uncertainty. |
