Search

CN-121981541-A - Intelligent research platform for oil and gas exploration and well position deployment risk early warning method

CN121981541ACN 121981541 ACN121981541 ACN 121981541ACN-121981541-A

Abstract

The invention discloses an intelligent research platform for oil and gas exploration and a well position deployment risk early warning method, wherein the platform constructs a three-layer intelligent architecture of data-knowledge-decision, and multi-source heterogeneous data is fused with a multi-mode knowledge graph through metadata-driven data management. The method is characterized in that a geological engineering integrated digital twin environment is adopted, a risk early warning model based on a multi-scale feature fusion network and physical mechanism constraint deep learning method is operated, and risk quantitative evaluation of the whole life cycle of well position deployment is realized. The invention innovatively introduces an intelligent decision closed loop of early warning-diagnosis-optimization, and continuously evolves a model through an online incremental learning mechanism, so that the paradigm transition from passive response to active early warning, analysis after the working and optimization in advance is finally realized, and the exploration success rate and the operation safety are obviously improved.

Inventors

  • PENG JING
  • MA YONGXIN
  • PENG HAILONG

Assignees

  • 中海石油(中国)有限公司湛江分公司

Dates

Publication Date
20260505
Application Date
20260120

Claims (10)

  1. 1. An intelligent research platform for oil and gas exploration is characterized in that the platform adopts a layered architecture, comprising, The data intelligent fusion layer adopts a metadata-driven data management frame and a multi-mode knowledge graph for integrating and semantically correlating seismic, logging, drilling, geology and real-time engineering data; The intelligent computing engine layer is used for bearing the geological engineering integrated digital twin and providing a multi-scale feature fusion network and a physical information neural network as a core algorithm model; An application service layer providing risk visualization, interactive analysis and decision suggestion services, and supporting integration with existing business systems through APIs, The data intelligent fusion layer, the intelligent calculation engine layer and the application service layer work cooperatively to form an intelligent decision closed loop from data perception, fusion analysis, risk early warning and decision optimization to feedback learning.
  2. 2. The intelligent research platform for oil and gas exploration according to claim 1, wherein the data intelligent fusion layer comprises a multi-mode knowledge graph construction module, geological entity and relation extraction is carried out by utilizing a pre-training model, unstructured documents and structured data are uniformly represented as triples of entity-relation-entity, embedded learning is carried out by utilizing a graph neural network, and complex semantic query and causal reasoning are supported.
  3. 3. The intelligent research platform for oil and gas exploration according to claim 1 or 2, wherein the multi-scale feature fusion network is used for receiving multi-scale feature input from seismic data, logging data and engineering data while drilling, and generating comprehensive risk feature vectors by introducing a cross-modal attention mechanism and adaptively fusing macroscopic structural features in the seismic data, reservoir features in the logging data and microscopic engineering features in the engineering data while drilling, so as to realize accurate identification of composite risks.
  4. 4. The intelligent research platform for oil and gas exploration according to claim 1 or 2, wherein the physical information neural network is characterized in that based on a traditional deep learning model, a stratum pressure equation and rock mechanics constitutive relation physical rule are used as soft constraint embedding loss functions to form a hybrid model of physical mechanism and data driving fusion so as to improve prediction reliability in a data sparse region.
  5. 5. The intelligent research platform for oil and gas exploration according to claim 4, wherein the loss function formula is as follows, L total =L data +λ 1 L physics +λ 2 L rock Wherein L data is a data fitting loss term, L physics is a physical constraint loss term based on a formation pressure prediction equation, L rock is a physical constraint loss term based on a rock mechanics constitutive relation, and lambda 1 and lambda 2 are super-parameters for balancing data fitting and physical rule constraint.
  6. 6. A well site deployment risk early warning method is characterized by utilizing the intelligent research platform for oil and gas exploration according to any one of claims 1 to 5, comprising the following steps, S1, constructing a multi-mode knowledge graph covering geological desserts and engineering drillability based on a data intelligent fusion layer; S2, extracting comprehensive risk features of the target well site by utilizing a multi-scale feature fusion network at an intelligent computing engine layer; s3, inputting the comprehensive risk characteristics into a physical information neural network, and outputting quantitative probabilities of risks such as well wall instability, drilling tool failure and the like; s4, dynamically updating model parameters by utilizing real-time drilling data through an integrated online incremental learning framework to realize self-evolution of the risk model; And S5, visualizing risk distribution and providing an optimization scheme based on case-based reasoning in an application service layer in a mode of superposition of a three-dimensional geological model and a risk thermodynamic diagram.
  7. 7. The well placement risk early warning method according to claim 6, wherein in the step S1, constructing the multi-modal knowledge graph comprises the following steps: S11, extracting a geological structure-engineering complex-processing measure triplet from a historical well completion report by using a geological text analyzer based on bert-base-chinese model fine tuning; And S12, importing the extracted triples and the seismic data body and logging curve data into a graph database to form a knowledge graph containing tens of thousands of nodes and relations.
  8. 8. The method of claim 6 or 7, wherein in S2, the comprehensive risk features include, but are not limited to, seismic coherence anomalies, abrupt well-logging points, non-uniformity of the ground stress field, and deviation of drilling fluid performance.
  9. 9. The method for well placement risk early warning according to claim 6 or 7, wherein in S4, the online incremental learning framework adopts an elastic weight consolidation technology, and learning of new knowledge and reservation of old knowledge are considered during model updating, so that catastrophic forgetting is avoided.
  10. 10. The method of claim 6 or 7, further comprising an intelligent decision loop for comparing the model prediction result with the actual drilling result, wherein the difference data automatically triggers model retraining and knowledge map updating to form a continuous optimization cycle.

Description

Intelligent research platform for oil and gas exploration and well position deployment risk early warning method Technical Field The invention belongs to the technical field of oil and gas exploration and development information, and particularly relates to an intelligent research platform for oil and gas exploration and a well position deployment risk early warning method. Background Currently, oil and gas exploration is continuously expanded to deep, deep sea and unconventional fields, and the serious challenges of extremely complex geological conditions, high risk of engineering operation and high investment cost are faced. Traditional well placement and risk pre-warning rely primarily on decentralized software tools and expert personal experience, with the following prominent bottlenecks: 1. Data barriers, namely different professional data standards such as geology, geophysical prospecting, drilling and the like, form a data chimney, and are difficult to perform cross-professional collaborative analysis and decision. 2. The model has the limitation that the prediction performance of the existing model based on statistics or pure data driving is rapidly reduced when a new area or complex geological phenomenon (such as a well wall stability problem under a high steep structure) uncovered by training data is faced, and the model lacks of physical common sense and has insufficient reliability. 3. And after decision delay, risk early warning is disjointed with engineering control measures, and only alarm is needed, but prescription cannot be achieved, so that time difference exists from the problem discovery to the execution of optimization measures, and the optimal control time is missed. For example, in the development of shale gas in Sichuan basin, lost circulation and stuck drill accidents caused by unknown faults or abrupt ground stress are frequent. The research of the eastern geophysical prospecting shows that the fracture recognition efficiency can be improved by utilizing the AI technology, but how to convert geological knowledge into engineering risk early warning in real time and give specific operation suggestions is still an industry pain point. Therefore, an integrated intelligent solution capable of deeply fusing multi-source data, embedding domain knowledge mechanisms and realizing early warning and control linkage is urgently needed. Disclosure of Invention The invention aims to solve the problem of providing an intelligent research platform for oil and gas exploration and a well position deployment risk early warning method, which can realize intelligent data fusion, mechanism and data dual-drive, have self-evolution capability and are suitable for popularization and application. In order to solve the technical problems, the technical proposal adopted by the invention is that the intelligent research platform for oil and gas exploration adopts a layered architecture, comprising, The data intelligent fusion layer adopts a metadata-driven data management frame and a multi-mode knowledge graph for integrating and semantically correlating seismic, logging, drilling, geology and real-time engineering data; The intelligent computing engine layer is used for bearing the geological engineering integrated digital twin and providing a multi-scale feature fusion network and a physical information neural network as a core algorithm model; An application service layer providing risk visualization, interactive analysis and decision suggestion services, and supporting integration with existing business systems through APIs, The data intelligent fusion layer, the intelligent calculation engine layer and the application service layer work cooperatively to form an intelligent decision closed loop from data perception, fusion analysis, risk early warning and decision optimization to feedback learning. Furthermore, the data intelligent fusion layer comprises a construction module of a multi-mode knowledge graph, a pre-training model is utilized to extract geological entities and relations, unstructured documents and structured data are uniformly represented as triples of entity-relation-entity, and an image neural network is utilized to conduct embedded learning to support complex semantic query and causal reasoning. Furthermore, the multi-scale feature fusion network is used for receiving multi-scale feature input from the seismic data, the well logging data and the engineering data while drilling, and generating comprehensive risk feature vectors by introducing a cross-modal attention mechanism to adaptively fuse macroscopic structural features in the seismic data, reservoir features in the well logging data and microscopic engineering features in the engineering data while drilling, so as to realize accurate identification of composite risks. Furthermore, the physical information neural network takes a stratum pressure equation and rock mechanics constitutive relation physical rule as soft constraint embed