A reservoir intelligent inversion method and related apparatus
By constructing a reservoir knowledge graph and using the thinking chain technology for task decomposition and collaborative execution, the reservoir inversion process is optimized, solving the problems of multiple solutions and uncertainty in traditional reservoir inversion methods, and achieving efficient and accurate intelligent reservoir inversion.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- INSTITUTE OF GEOLOGY AND GEOPHYSICS CHINESE ACADEMY OF SCIENCES
- Filing Date
- 2026-03-24
- Publication Date
- 2026-06-23
Smart Images
Figure CN122260437A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of reservoir intelligent inversion technology, and in particular to a reservoir intelligent inversion method and related apparatus. Background Technology
[0002] Global energy demand continues to grow, with oil and natural gas playing a crucial role in the global energy mix. Reservoir geological modeling is essentially a process of accurately predicting the spatial distribution of underground reservoirs by integrating various geological information, including wellbore, seismic, and geological models. It is key to oil and gas reserve evaluation, drilling design, and adjustments to oil and gas field development strategies. However, as easily exploitable oil and gas resources gradually deplete, the difficulty and cost of oil and gas exploration continue to increase, making traditional exploration reservoir geological modeling techniques insufficient to meet the demands for high efficiency and accuracy. Many undeveloped oil and gas resources are located in deep and ultra-deep layers, with unconventional oil and gas reservoirs accounting for a significant proportion. Due to more complex geological conditions, limitations in modeling methods, and incompleteness of exploration data, oil and gas reservoir inversion exhibits multiple solutions and strong uncertainties.
[0003] Currently, traditional reservoir inversion methods, such as geostatistical inversion methods and CNN (convolutional neural network)-based inversion methods, struggle to effectively integrate multi-source heterogeneous data and complex geological prior knowledge. This results in highly ambiguous and uncertain inversion results, leading to low accuracy in reservoir inversion. Furthermore, existing deep learning models exhibit poor generalization ability, showing significant performance degradation when faced with new reservoir types that differ greatly from the training sample distribution (such as the transition from sandstone to shale). They lack effective knowledge transfer and online adaptation mechanisms. Combined with insufficient automation and intelligence in the reservoir modeling process, which heavily relies on expert experience for data preparation, parameter adjustment, and result evaluation, this ultimately results in low inversion efficiency.
[0004] In summary, how to provide a smart reservoir inversion method to improve the accuracy and efficiency of reservoir inversion has become a technical problem that urgently needs to be solved in this field. Summary of the Invention
[0005] The purpose of this application is to provide a reservoir intelligent inversion method and related apparatus, which can improve the accuracy and efficiency of reservoir inversion.
[0006] To achieve the above objectives, this application provides the following solution.
[0007] In a first aspect, this application provides a reservoir intelligent inversion method, which includes the following steps.
[0008] Acquire multi-source heterogeneous data of the target region and perform preprocessing.
[0009] An initial geological model generator is constructed based on a conditional diffusion model, and the implicit space parameters of the initial geological model generator are optimized by seismic wave forward modeling based on preprocessed multi-source heterogeneous data.
[0010] A reservoir knowledge graph is constructed, and the initial geological model generator is adaptively pre-trained and transferred online to obtain an optimized geological model generator and optimized geological model generator parameters.
[0011] Based on the thinking chain technology and language large model technology, the task requirements of reservoir modeling are analyzed, the task is decomposed and dynamically coordinated, and the latent space parameters are obtained after iterative optimization.
[0012] The iteratively optimized latent space parameters are input into the optimized geological model generator, which outputs a three-dimensional distribution model containing porosity, permeability and oil saturation, and extracts confidence interval features.
[0013] Based on the three-dimensional distribution model, the confidence interval features, and the optimized geological model generator parameters, the reservoir knowledge graph is incrementally updated, and a reservoir model that conforms to the geological model features and matches the seismic waveform is determined, thereby realizing intelligent reservoir inversion.
[0014] Secondly, this application provides a reservoir intelligent inversion device, which includes the following functional modules.
[0015] The data acquisition and preprocessing module is used to acquire multi-source heterogeneous data from the target area and perform preprocessing.
[0016] The seismic wave forward modeling optimization module is used to construct an initial geological model generator based on a conditional diffusion model, and to optimize the implicit space parameters of the initial geological model generator by means of seismic wave forward modeling based on preprocessed multi-source heterogeneous data.
[0017] An adaptive online pre-training and model transfer module is used to construct a reservoir knowledge graph and perform adaptive online pre-training and model transfer on the initial geological model generator to obtain an optimized geological model generator and optimized geological model generator parameters.
[0018] The task decomposition and collaborative execution module is used to analyze task requirements, decompose tasks, and dynamically and collaboratively execute tasks for reservoir modeling based on thinking chain technology and language large model technology, so as to obtain the latent space parameters after iterative optimization.
[0019] The confidence interval feature extraction module is used to input the iteratively optimized latent space parameters into the optimized geological model generator, output a three-dimensional distribution model including porosity, permeability and oil saturation, and extract confidence interval features.
[0020] The reservoir model determination module is used to incrementally update the reservoir knowledge graph based on the three-dimensional distribution model, the confidence interval features, and the optimized geological model generator parameters, and to determine the reservoir model that conforms to the geological model features and matches the seismic waveform, thereby realizing intelligent reservoir inversion.
[0021] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the reservoir intelligent inversion method described in any one of the above.
[0022] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the reservoir intelligent inversion method described above.
[0023] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the reservoir intelligent inversion method described above.
[0024] According to the specific embodiments provided in this application, this application has the following technical effects.
[0025] This application provides a reservoir intelligent inversion method and related apparatus. On one hand, this application acquires multi-source heterogeneous data of the target area and constructs an initial geological model generator based on a conditional diffusion model. It then optimizes the latent space parameters of the initial geological model generator using seismic wave forward modeling, thus completing a seismic wave generative inversion process driven by both data and physics. This process is validated through the design of the initial geological model generator and seismic wave forward modeling, integrating the modeled geological model with multi-source heterogeneous data to reduce the uncertainty of reservoir inversion and improve its accuracy. On the other hand, this application constructs a reservoir knowledge graph and adaptively performs online pre-training and model transfer, combined with incremental updates to the reservoir knowledge graph, thus completing a knowledge-increment-driven adaptive online pre-training process for the reservoir. This process uses the reservoir knowledge graph to store, retrieve, utilize, and update the encoded latent space information obtained from the pre-training of the geological model generator, mimicking the process of geological experts' understanding and updating of geological model knowledge. This significantly improves the transferability and portability of the geological model generator across different geological conditions, solving the generalization problem of reservoir generative models when facing complex geological conditions. On the other hand, this application uses the thinking chain technology and language large model technology to analyze the task requirements, decompose the task, and dynamically and collaboratively execute the reservoir modeling task. The iteratively optimized latent space parameters are input into the optimized geological model generator, which outputs a three-dimensional distribution model containing porosity, permeability, and oil saturation to provide rich confidence interval features. This facilitates incremental updates of the reservoir knowledge graph, thereby obtaining a reservoir model that conforms to the geological model characteristics and matches the seismic waveform. This completes the process of automated reservoir modeling based on the thinking chain, realizes online automated reservoir inversion modeling, and can effectively improve the efficiency of reservoir inversion. Attached Figure Description
[0026] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0027] Figure 1 This is an application environment diagram of a reservoir intelligent inversion method provided in an embodiment of this application.
[0028] Figure 2 This is a schematic flowchart of a reservoir intelligent inversion method provided in an embodiment of this application.
[0029] Figure 3 This is a technical roadmap for seismic wave generative inversion based on data-physics dual-drive, provided in one embodiment of this application.
[0030] Figure 4 This is a technical roadmap for knowledge increment-driven adaptive online pre-training of reservoirs provided in one embodiment of this application.
[0031] Figure 5 This is a technical roadmap for automated reservoir modeling based on thought chain, provided as an embodiment of this application.
[0032] Figure 6 This is a schematic diagram of the structure of a reservoir intelligent inversion device provided in an embodiment of this application.
[0033] Figure 7 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation
[0034] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0035] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0036] The reservoir intelligent inversion method provided in this application embodiment can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on another server. Terminal 102 can send multi-source heterogeneous data of the target area to server 104. After receiving the multi-source heterogeneous data of the target area, server 104 performs preprocessing on the multi-source heterogeneous data of the target area; constructs an initial geological model generator based on the conditional diffusion model, and optimizes the latent space parameters using seismic wave forward modeling; constructs a reservoir knowledge graph and adaptively pre-trains and transfers models; performs task requirement analysis, task decomposition, and dynamic collaborative execution of the reservoir modeling task based on Chain-of-Thought (CoT) technology and Language Large Model (LLM) technology to obtain iteratively optimized latent space parameters; inputs the iteratively optimized latent space parameters into the optimized geological model generator, outputs a three-dimensional distribution model, and extracts confidence interval features; incrementally updates the reservoir knowledge graph to determine reservoir models that conform to geological model characteristics and match seismic waveforms, realizing intelligent reservoir inversion. Server 104 can feed back the obtained reservoir models that conform to geological model characteristics and match seismic waveforms to terminal 102. In addition, in some embodiments, the reservoir intelligent inversion method can also be implemented by the server 104 or the terminal 102 alone. For example, the terminal 102 can directly perform reservoir intelligent inversion processing on the multi-source heterogeneous data of the target area, or the server 104 can obtain the multi-source heterogeneous data of the target area from the data storage system and perform reservoir intelligent inversion processing on the multi-source heterogeneous data of the target area.
[0037] The terminal 102 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, and IoT devices. The server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers, or it can be a cloud server.
[0038] In one exemplary embodiment, such as Figure 2 As shown, a smart reservoir inversion method is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps S1 to S6.
[0039] S1: Acquire multi-source heterogeneous data of the target area and preprocess it to obtain preprocessed multi-source heterogeneous data.
[0040] S2: Construct an initial geological model generator based on the conditional diffusion model, and optimize the implicit space parameters of the initial geological model generator by using seismic wave forward modeling based on the preprocessed multi-source heterogeneous data.
[0041] S3: Construct a reservoir knowledge graph and perform adaptive online pre-training and model transfer on the initial geological model generator to obtain an optimized geological model generator and optimized geological model generator parameters.
[0042] S4: Based on the thinking chain technology and language large model technology, the task requirements of reservoir modeling are analyzed, the task is decomposed and dynamically coordinated, and the latent space parameters are obtained after iterative optimization.
[0043] S5: Input the iteratively optimized latent space parameters into the optimized geological model generator, output a three-dimensional distribution model containing porosity, permeability and oil saturation, and extract confidence interval features.
[0044] S6: Based on the three-dimensional distribution model, the confidence interval features, and the optimized geological model generator parameters, the reservoir knowledge graph is incrementally updated, and a reservoir model that conforms to the geological model features and matches the seismic waveform is determined to achieve intelligent reservoir inversion.
[0045] By implementing steps S1 to S6 above, this application can realize seismic wave generative inversion based on data-physics dual-drive, reservoir adaptive online pre-training based on knowledge increment drive, and reservoir automated modeling based on thought chain. This can not only improve the accuracy and efficiency of reservoir inversion, but also improve the transferability and portability of geological model generators between different geological conditions, and solve the generalization problem of reservoir generative models when facing complex geological conditions.
[0046] In this embodiment, in step S1, the multi-source heterogeneous data includes seismic data, well logging data, well logging data, core data, and historical reservoir parameters. The preprocessing includes standardization, noise reduction, and cross-scale fusion.
[0047] In this embodiment, step S1 specifically includes the following steps.
[0048] S11: Obtain multi-source heterogeneous data for the target region.
[0049] S12: Standardize the multi-source heterogeneous data to obtain standardized multi-source heterogeneous data.
[0050] S13: Perform noise reduction processing on the standardized multi-source heterogeneous data to obtain noise-reduced multi-source heterogeneous data.
[0051] S14: Perform cross-scale fusion processing on the denoised multi-source heterogeneous data to obtain preprocessed multi-source heterogeneous data.
[0052] In this embodiment, step S2 specifically includes the following steps.
[0053] S21: Construct an initial geological model generator based on the conditional diffusion model.
[0054] S22: Obtain a large amount of geological-geophysical samples, and use the large amount of geological-geophysical samples to pre-train the initial geological model generator to extract latent space parameters.
[0055] S23: Input the preprocessed multi-source heterogeneous data as a condition constraint into the initial geological model generator, and output the simulated synthetic data.
[0056] S24: Forward modeling is performed using seismic wave forward modeling. The simulated composite data is compared with the measured data, and the hidden space parameters are optimized using error backpropagation.
[0057] In this embodiment, step S3 specifically includes the following steps.
[0058] S31: Construct a reservoir knowledge graph. The reservoir knowledge graph contains reservoir entity attributes, model entity attributes, and relationships between entities in the form of triples. The reservoir entity attributes include sedimentary facies, reservoir thickness, lithological distribution, elastic distribution, physical property distribution, and fluid distribution. The model entity attributes include model network parameters and applicable geological conditions.
[0059] S32: Based on the reservoir knowledge graph, retrieve similar model entities in the reservoir knowledge graph according to the target reservoir type, and use transfer learning to initialize the parameters of the initial geological model generator and load the latent space distribution prior.
[0060] S33: Using some labeled samples from the preprocessed multi-source heterogeneous data, the initial geological model generator is fine-tuned online. Combined with the seismic waveform matching loss optimization model, the optimized geological model generator and its parameters are obtained.
[0061] In this embodiment, step S4 mainly analyzes the reservoir modeling task requirements based on the thinking chain technology and language large model technology, decomposes the reservoir modeling task into a knowledge retrieval subtask, a model generation subtask, and a forward modeling verification subtask, and dynamically coordinates the reservoir knowledge graph module, the geological model generator, and the forward modeling simulation module to collaboratively execute the knowledge retrieval subtask, the model generation subtask, and the forward modeling verification subtask to obtain the iteratively optimized latent space parameters. The reservoir knowledge graph module refers to the module responsible for constructing and managing the reservoir knowledge graph, and the forward modeling simulation module refers to the module responsible for seismic wave forward modeling simulation. Specifically, it includes the following steps.
[0062] S41: Use language large model technology to analyze reservoir modeling tasks and determine modeling objectives and constraints.
[0063] S42: Based on the modeling objectives and constraints, the reservoir modeling task is decomposed using the thinking chain technique to obtain the knowledge retrieval subtask, the model generation subtask, and the forward verification subtask.
[0064] S43: Based on the knowledge retrieval subtask, the model generation subtask, and the forward modeling verification subtask, dynamically call the reservoir knowledge graph module to retrieve entity attributes and constraint rules related to the current task, and use the entity attributes and constraint rules related to the current task as input conditions to activate the geological model generator to generate candidate reservoir models that meet the constraint rules.
[0065] S44: Continuously monitor the execution status and results of the knowledge retrieval subtask, the model generation subtask, and the forward verification subtask; the execution status and results include data matching degree indicators.
[0066] S45: Based on the execution status and results, when the data matching degree index is less than a preset threshold (e.g., the preset threshold is 90%), the sampling weight strategy of the latent space vector is dynamically adjusted to obtain the iteratively optimized latent space parameters. Simultaneously, the incremental update process of the reservoir knowledge graph is triggered, realizing dynamic collaboration and closed-loop optimization between the reservoir knowledge graph module, the geological model generator, and the forward simulation module. When the data matching degree index is greater than or equal to the preset threshold, no operation is required.
[0067] In this embodiment, the constraint rules include geological constraint rules, physical constraint rules, and data constraint rules. Specifically, the geological constraint rules are applicable conditions defined based on the reservoir entity attributes in the reservoir knowledge graph; the physical constraint rules are physical laws verified by seismic wave forward modeling; and the data constraint rules are standardization requirements for the input data.
[0068] In this embodiment, the adaptive online pre-training process of the initial geological model generator includes an encoding stage and a decoding stage. The encoding stage stores abstract geological model knowledge by encoding latent space parameters, and the decoding stage integrates reservoir parameter prior information, well logging information, and well logging information to generate a geological model that meets the constraints.
[0069] In this embodiment, step S6 specifically includes the following steps.
[0070] S61: Based on the three-dimensional distribution model, the confidence interval features, and the optimized geological model generator parameters, the reservoir knowledge graph is incrementally updated to obtain the updated reservoir knowledge graph.
[0071] S62: Based on the updated reservoir knowledge graph, determine the reservoir model that conforms to the geological model characteristics and matches the seismic waveform, and realize intelligent reservoir inversion.
[0072] To make the technical solution of this application clearer, the specific implementation process of the technical solution of this embodiment will be described in detail below by way of examples. Specifically, the implementation steps are as follows.
[0073] Step 1: Obtain multi-source heterogeneous data through massive geological and geophysical sample collection and preprocessing.
[0074] This embodiment acquires multi-source heterogeneous data such as seismic data, well logging data, well logging data, core data, and historical reservoir parameters of the target area, covering reservoir types such as clastic rocks, carbonate rocks, and shale, and performs standardization, noise reduction, and cross-scale fusion processing on the data.
[0075] First, a large-scale, systematic geological and geophysical sample collection effort is undertaken for the target area to obtain datasets from diverse sources and with varying structures. Specifically, this involves collecting seismic exploration data (reflecting subsurface impedance structure), well logging data (providing high-resolution vertical information at well points), core analysis data (providing direct measurements of rock physical properties), and reservoir parameter data accumulated during historical exploration and development (such as porosity and permeability). This data covers major reservoir types, including clastic rocks (such as sandstone), carbonate rocks (such as limestone and dolomite), and unconventional reservoirs such as shale. Subsequently, necessary preprocessing operations are performed on these multi-source, heterogeneous data, including data format unification and standardization (ensuring data comparability), noise filtering (improving data quality), and crucial cross-scale fusion processing (effectively correlating and integrating data information at different scales and resolutions to provide a consistent, high-quality data foundation for subsequent modeling).
[0076] Step 2: Construct an initial geological model generator based on the conditional diffusion model, and optimize the hidden space parameters through seismic wave forward modeling.
[0077] Building upon traditional diffusion models, conditional diffusion models introduce additional conditional information to guide the generation process, thereby generating images or data that better meet specific needs. Pre-trained on massive geological and geophysical samples, the initial geological model generator based on the conditional diffusion model learns potential geological patterns and uses encoded latent space parameters to store abstract geological pattern knowledge. In any similar geological region, using multi-source heterogeneous data such as reservoir parameter prior information, well logging information, and well logging data as constraints, these data fully participate in the decoding process of the diffusion model, enabling the generation of geological models that comprehensively meet the conditional constraints. If a large-scale geological model generator for various reservoir types is constructed, only the basic geological knowledge and conditions of the target reservoir need to be input to generate a large number of potential geological model samples in a short time.
[0078] Based on this, by performing seismic wave forward modeling on the reservoir geological model, the simulated synthetic data is compared with the actual data. Through error backpropagation, the vectors of the latent space can be sampled and optimized, thereby updating the control geological model generation process. This satisfies both the conditional constraints and matches the direction of the seismic waveform, achieving dual constraints of the geological model and the seismic waveform.
[0079] Step 3: Construct a reservoir knowledge graph and associate the attributes of reservoir entities with those of model entities.
[0080] In step 3, the reservoir knowledge graph stores reservoir entity attributes, model entity attributes, and relationships between entities in the form of triples. Attributes of reservoir sample entities include sedimentary facies, reservoir thickness, lithological distribution, elasticity distribution, physical property distribution, and fluid distribution. Relationships between these entities are established based on the level of abstraction. Attributes of geological model generator entities include model network parameters and typical geological conditions applicable to the model, such as different types of oil and gas reservoirs like clastic rocks, carbonate rocks, and shale.
[0081] This embodiment defines reservoir entity attributes (sedimentary facies, lithology, physical properties, fluid distribution) and model entity attributes (network parameters, applicable geological conditions); relationships between entities are established through triples, such as "clastic reservoir, applicable to ResNet (deep learning neural network) geological model generator". Furthermore, it supports incremental updates and transfer learning of the reservoir knowledge graph.
[0082] Step 4: Implement adaptive online pre-training and model transfer based on reservoir knowledge graph.
[0083] In step 4, a geological model generator adapted to the current geological conditions and reservoir type is obtained by retrieving relevant entities and reasoning based on relevance. During knowledge retrieval, similar geological model generators are retrieved for a given geological model category based on relevance. The retrieved models and their parameters are used to assist subsequent geological model generation tasks. Similar model entities in the reservoir knowledge graph are retrieved based on the target reservoir type. During knowledge application, transfer learning is used to initialize the parameters of the geological model generator, loading latent space distribution priors to reduce the online training time for further pre-training and modeling. Online fine-tuning is performed using a small number of labeled samples in the target area, combined with seismic waveform matching loss to optimize the model. During knowledge updating, relevant knowledge in the reservoir knowledge graph is updated based on the results of the knowledge application steps. Through this interactive process, the geological model generator acquires transferability and can be applied efficiently and accurately to different geological and oil and gas reservoir types.
[0084] Step 5: Decompose the modeling task based on the thinking chain technology and optimize multiple modules in a coordinated manner.
[0085] In this embodiment, the core of step 5 lies in utilizing the thought chain technology combined with the powerful parsing and reasoning capabilities of the language big model to intelligently decompose and schedule complex reservoir modeling tasks. The specific execution process is as follows: First, the language big model is responsible for parsing the task requirement description input by the user or system, gaining a deep understanding of the modeling objectives and constraints. Then, based on thought chain reasoning, the language big model logically decomposes the overall task into a series of executable sub-task units, such as: retrieving relevant reservoir knowledge graph entries according to the task objectives (knowledge retrieval sub-task), calling and activating the conditional diffusion model to generate preliminary reservoir model candidate schemes (model generation sub-task), and using forward simulation to verify the matching degree between the generated model and actual data (forward verification sub-task), etc. During task execution, the system dynamically calls the reservoir knowledge graph module to retrieve information such as entity attributes and constraint rules related to the current task, and uses this knowledge as conditional input to activate the geological model generator (i.e., the conditional diffusion model) to generate candidate reservoir models that meet the constraints. The system continuously monitors the execution status and results of each subtask, especially the data matching degree index. When the matching degree is detected to be less than the preset threshold, the system will dynamically adjust the sampling weight strategy of the latent space vector and may trigger the update process of the reservoir knowledge graph, thereby realizing dynamic collaboration and closed-loop optimization among core modules such as reservoir knowledge graph, geological model generator, and forward modeling verification.
[0086] In this embodiment, the constraint rules include the following:
[0087] (1) Geological constraint rules: applicable conditions based on the reservoir entity attributes (such as sedimentary facies, lithology, physical properties) defined in the reservoir knowledge graph, such as "clastic reservoirs must use the ResNet geological model generator model".
[0088] (2) Physical constraint rules: based on the physical laws verified by seismic wave forward modeling, such as "the generated geological model must match the seismic waveform error threshold ≤ 5%".
[0089] (3) Data constraint rules: Standardization requirements for input data, such as "well logging data must be denoised and fused across scales". For example, if the task involves shale reservoirs, the constraint rule may be "prioritize the retrieval of shale-applicable model parameters and load fluid distribution constraints to optimize the generation process".
[0090] Step 6: Automated generation of multi-scale fused reservoir geological models.
[0091] In this embodiment, the latent space parameters after iterative optimization are input into the geological model generator, which outputs a three-dimensional distribution model of reservoir porosity, permeability, and oil saturation. Combined with an intelligent agent, the results from multiple rounds are integrated to extract confidence interval features, which refer to the spatial distribution characteristics of the confidence interval.
[0092] The intelligent agent framework based on thinking chain technology and language big model technology can provide developers with pre-set components, abstract concepts and tools to simplify the development of complex artificial intelligence systems. It plays a vital role in promoting the development and application of big model technology, and helps to accelerate the development of application scenarios, standardize business processes and enhance the scalability and accessibility of functions.
[0093] In this embodiment, the components of the intelligent agent framework typically include the following.
[0094] 1) Intelligent agent architecture, used to define the structure of the internal organization, including its decision-making process, memory system and interaction capabilities.
[0095] 2) Environment interface, which connects the intelligent agent to its operating environment.
[0096] 3) Task management, used to define, assign and track the completion status of agent tasks.
[0097] 4) Communication protocols enable interaction between intelligent agents and between intelligent agents and humans.
[0098] 5) Integration tools connect intelligent agents with external data sources and application interfaces.
[0099] 6) Monitoring and debugging, allowing developers to observe agent behavior, track performance, and identify problems.
[0100] Step 7: Incrementally update the reservoir knowledge graph to improve the system knowledge base.
[0101] This embodiment writes the attributes of the new reservoir model and the parameters of the geological model generator into the reservoir knowledge graph, updates the relationships between entities and the tags of applicable geological conditions, and completes online knowledge iteration.
[0102] For automated reservoir modeling scenarios, an intelligent agent framework based on the thought chain is employed. Leveraging the powerful reasoning capabilities of the thought chain and a large language model, and constrained by business scenario rules, it achieves the decomposition of complex geological modeling tasks, accurate invocation of intelligent algorithms for oil and gas reservoir geological model generators, and utilization and updating of reservoir knowledge graphs. Ultimately, this results in the efficient generation of geological models and online updates based on geological changes. By designing and optimizing agent roles and communication protocols, system-level task coordination and optimization are achieved, enabling intelligent and automated processing of geological modeling tasks. Finally, a knowledge-, data-, and physical multi-driven reservoir generative intelligent inversion system is constructed.
[0103] Step 8: Through the collaborative efforts of data, knowledge, and physics-driven mechanisms from Steps 1 to 7, intelligent reservoir inversion is ultimately achieved. The final output is a high-precision reservoir model that conforms to geological model characteristics and matches seismic waveforms, thereby reducing ambiguity and supporting dynamic reconstruction.
[0104] In this embodiment, step 8 is the result of the synergistic effect of all the aforementioned technical steps, marking the final completion of the intelligent reservoir inversion process. Through the deep integration and synergistic optimization mechanism of "data-driven" (using multi-source heterogeneous actual observation data), "knowledge-driven" (using structured geological and model knowledge stored in the reservoir knowledge graph), and "physical-driven" (using physical constraints such as seismic wave forward modeling), constructed and executed in steps 1 to 7, the system can finally generate a high-precision three-dimensional geological model of the target reservoir. This output model has two key characteristics: first, it strictly conforms to the geological pattern laws (such as sedimentary facies distribution, lithological assemblage characteristics, etc.) learned from massive samples and the reservoir knowledge graph; second, the seismic waveforms generated by its forward modeling have a high degree of matching with the actual acquired seismic data. This modeling method, under the dual strong constraints of data and physical laws and incorporating prior geological knowledge, significantly reduces the inherent multiple-solution problem in traditional reservoir inversion. Meanwhile, the system design supports the dynamic reconstruction capability of the model, which means that when new data inputs are obtained or the reservoir knowledge graph is updated, the optimization process can be efficiently re-executed to output an updated reservoir model, thereby adapting to the needs of the ever-deepening understanding during the exploration and development process.
[0105] The reservoir intelligent inversion method proposed in this embodiment mainly includes: seismic wave generative inversion based on data-physics dual-drive, reservoir adaptive online pre-training based on knowledge increment drive, and reservoir automated modeling based on thought chain.
[0106] Figure 3 This is a technical roadmap for data-physical dual-driven seismic wave generation inversion. The data-physical dual-driven seismic wave generation inversion includes two processes: geological model generator (imagination) and physical simulation (empirical verification). The main steps are as follows: (1) Using a large number of geological-geophysical samples including sedimentary facies, reservoir thickness, lithological distribution, elastic distribution, physical property distribution and fluid distribution, a geological model generator is pre-trained and encoded to extract the geological model parameters of the latent space; (2) Input multi-source condition data (well logging, seismic, etc.) of the target reservoir, and generate a preliminary geological model through conditional diffusion generation decoding; (3) Through forward modeling of the seismic wave equation, the forward modeling results are matched with the actual seismic waves, and the geological model parameters of the latent space are optimized by error backpropagation and sampling optimization of latent vectors; (4) Iteratively generate a geological model that meets the conditional constraints and matches the seismic waveform.
[0107] Figure 4 This is a technical roadmap for knowledge-increment-driven adaptive online pre-training of reservoirs. This roadmap includes a geological model generator and a reservoir knowledge graph. The transfer learning and online pre-training of the geological model generator and the reservoir knowledge graph module are linked (e.g., knowledge retrieval, transfer parameters, and online fine-tuning). Its focus is limited to the synergy between the reservoir knowledge graph and the geological model generator. Through knowledge extraction, retrieval, application, and updating of historical data, new conditions and data are added to a vast database of geological and geophysical samples, including sedimentary facies, reservoir thickness, lithological distribution, elasticity distribution, physical property distribution, and fluid distribution. This enables pre-training and encoding of the latent space and generates new model knowledge, thereby improving the transferability and generalization of the geological model generator under different geological conditions.
[0108] The process of knowledge increment-driven reservoir adaptive online pre-training mainly includes the following steps: (1) constructing reservoir knowledge graph triples (entity, relation, attribute); (2) retrieving similar model parameters according to the current geological conditions and transferring them to the target task; (3) updating the model entities and reservoir sample entities in the reservoir knowledge graph online.
[0109] Figure 5This is a technical roadmap for automated reservoir modeling based on the thinking chain. The process of automated reservoir modeling based on the thinking chain mainly includes the following steps: (1) Analyze the geological modeling task requirements through the language big model. Specifically, generate the reservoir model according to the user instructions, and extract, retrieve, utilize and update the reservoir knowledge graph using the language big model; (2) Decompose and automatically execute the task after thinking and planning using the language big model. Decompose the task into sub-processes (data call, model selection, constraint loading), and generate the geological model generator agent by calling tools and calling data (multi-type data of oil and gas fields); (3) Coordinate the agent to perform knowledge retrieval, model generation, inversion optimization and result verification, and determine whether the requirements are solved. If so, output the result; otherwise, return to the language big model to repeat the task decomposition and automatic execution.
[0110] This embodiment expands upon cutting-edge technologies such as conditional diffusion neural network models, seismic waveform forward modeling constraint optimization, geological knowledge fusion and embedding, and large-scale model thinking chains through core work including data-physical dual-driven seismic wave generative inversion, knowledge increment-driven reservoir adaptive online pre-training, and thought chain-based automated reservoir modeling. By integrating knowledge-data-physical multi-driven approaches, it improves the accuracy of intelligent reservoir inversion, explores the automation level of knowledge flow, data flow, and model flow in intelligent systems, and contributes to the strategic needs of new productivity and intelligent transformation of oil and gas.
[0111] This embodiment reduces the uncertainty of reservoir inversion in the data-physics dual-driven seismic wave generative inversion process by designing a generative neural network to fuse modeled geological models and exploration data. It mainly includes a geological model generator based on a conditional diffusion model – an imaginative step – and a forward modeling matching optimizer based on seismic wave equations – an empirical step. By iteratively coupling the imaginative and empirical steps, reservoir parameters are obtained that satisfy both prior knowledge and well logging constraints, and match the physical processes of seismic wave propagation, thereby reducing the uncertainty and ambiguity of reservoir modeling.
[0112] This embodiment addresses the generalization problem of reservoir generative models under complex geological conditions during knowledge-increment-driven adaptive online pre-training. The geological model generator stores abstract geological model knowledge in the encoding latent space and constraint knowledge such as well logging core samples in the decoding latent space, using neural network parameters. By constructing a reservoir knowledge graph to store, retrieve, utilize, and update the encoded latent space information obtained from the pre-training of the geological model generator, it mimics the process of geological experts recognizing and updating geological model knowledge, significantly improving the transferability and portability of the geological model generator across different geological conditions.
[0113] In this embodiment, to improve the automation level of the entire system and realize online reservoir inversion modeling based on the thinking chain, a multi-agent framework for reservoir inversion is proposed. The reservoir knowledge graph is incrementally coupled to the geological model generator, roles are designed for relevant functional modules, and communication protocols between agents are set. This enables the reservoir modeling system to dynamically adjust its behavior according to changes in geological conditions, ensuring efficient allocation and consistency with the target.
[0114] This embodiment conducts a quantitative comparative experiment with existing technologies (such as purely data-driven deep learning inversion, traditional geostatistics-based inversion, or generative models without reservoir knowledge graphs and CoT collaboration). This embodiment utilizes a knowledge-data-physics multi-drive approach, combined with adaptive transfer and collaborative optimization, to form an overall multi-drive framework. While ensuring model applicability, it effectively improves the accuracy of intelligent reservoir inversion and reduces the uncertainty and ambiguity of reservoir modeling. Transfer learning reduces training costs, and automated processes reduce manual labor, improving the efficiency of intelligent reservoir inversion. CoT task decomposition and scheduling, along with a closed-loop process, enhance the automation level of reservoir modeling and intelligent inversion. The structured representation and transfer mechanism of the reservoir knowledge graph, along with incremental updates, enhances the generalization ability of reservoir modeling and inversion.
[0115] In one exemplary embodiment, a reservoir intelligent inversion system is provided. This system aims to achieve data-knowledge-physics multi-driven collaborative intelligent reservoir inversion, outputting a high-precision reservoir model that supports online adaptive and dynamic reconstruction. The system includes a geological model generator, a reservoir knowledge graph module, and an intelligent task scheduling and collaborative optimization central module. The geological model generator is used for high-fidelity geological model generation (constrained by both geological knowledge and physical laws). Based on a conditional diffusion model, it stores geological model knowledge through pre-trained latent space parameters and optimizes the generation process by combining seismic wave forward modeling and error backpropagation. The reservoir knowledge graph module is used for structured knowledge storage, retrieval, and incremental updates. It includes reservoir entity attributes (sedimentary facies, lithology, physical properties, etc.) and model entity attributes (network parameters, applicable geological conditions), supporting incremental updates and transfer learning. The intelligent task scheduling and collaborative optimization central module is mainly used for intelligent task parsing, sub-task scheduling, multi-module collaborative optimization, and automated modeling process driving. Based on the reasoning ability of the language big model, it parses user task requirements, intelligently decomposes them into sub-tasks (such as knowledge retrieval sub-tasks, model generation sub-tasks, and forward modeling verification sub-tasks), and dynamically coordinates the reservoir knowledge graph module, geological model generator, and forward modeling module for collaborative execution and closed-loop optimization.
[0116] Based on the same inventive concept, this application also provides a reservoir intelligent inversion device for implementing the reservoir intelligent inversion method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more reservoir intelligent inversion device embodiments provided below can be found in the limitations of the reservoir intelligent inversion method described above, and will not be repeated here.
[0117] In one exemplary embodiment, such as Figure 6 As shown, a reservoir intelligent inversion device is provided, which includes the following functional modules.
[0118] The data acquisition and preprocessing module is used to acquire multi-source heterogeneous data from the target area and perform preprocessing.
[0119] The seismic wave forward modeling optimization module is used to construct an initial geological model generator based on a conditional diffusion model, and to optimize the implicit space parameters of the initial geological model generator by means of seismic wave forward modeling based on preprocessed multi-source heterogeneous data.
[0120] An adaptive online pre-training and model transfer module is used to construct a reservoir knowledge graph and perform adaptive online pre-training and model transfer on the initial geological model generator to obtain an optimized geological model generator and optimized geological model generator parameters.
[0121] The task decomposition and collaborative execution module is used to analyze task requirements, decompose tasks, and dynamically and collaboratively execute tasks for reservoir modeling based on thinking chain technology and language large model technology, so as to obtain the latent space parameters after iterative optimization.
[0122] The confidence interval feature extraction module is used to input the iteratively optimized latent space parameters into the optimized geological model generator, output a three-dimensional distribution model including porosity, permeability and oil saturation, and extract confidence interval features.
[0123] The reservoir model determination module is used to incrementally update the reservoir knowledge graph based on the three-dimensional distribution model, the confidence interval features, and the optimized geological model generator parameters, and to determine the reservoir model that conforms to the geological model features and matches the seismic waveform, thereby realizing intelligent reservoir inversion.
[0124] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 7As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores multi-source heterogeneous data. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a reservoir intelligent inversion method.
[0125] Those skilled in the art will understand that Figure 7 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.
[0126] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.
[0127] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.
[0128] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).
[0129] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.
[0130] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0131] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A smart reservoir inversion method, characterized in that, The reservoir intelligent inversion method includes: Acquire multi-source heterogeneous data of the target area and perform preprocessing; An initial geological model generator is constructed based on a conditional diffusion model, and the latent space parameters of the initial geological model generator are optimized by seismic wave forward modeling based on preprocessed multi-source heterogeneous data. A reservoir knowledge graph is constructed, and the initial geological model generator is adaptively pre-trained and model transferred to obtain an optimized geological model generator and optimized geological model generator parameters. Based on the thinking chain technology and language large model technology, the task requirements analysis, task decomposition and dynamic collaborative execution of reservoir modeling tasks are carried out to obtain the latent space parameters after iterative optimization. The iteratively optimized latent space parameters are input into the optimized geological model generator, which outputs a three-dimensional distribution model containing porosity, permeability and oil saturation, and extracts confidence interval features. Based on the three-dimensional distribution model, the confidence interval features, and the optimized geological model generator parameters, the reservoir knowledge graph is incrementally updated, and a reservoir model that conforms to the geological model features and matches the seismic waveform is determined, thereby realizing intelligent reservoir inversion.
2. The reservoir intelligent inversion method according to claim 1, characterized in that, An initial geological model generator is constructed based on a conditional diffusion model. Then, based on preprocessed multi-source heterogeneous data, the latent space parameters of the initial geological model generator are optimized using seismic wave forward modeling. Specifically, this includes: An initial geological model generator is constructed based on the conditional diffusion model. A large amount of geological and geophysical samples are acquired, and the initial geological model generator is pre-trained using the large amount of geological and geophysical samples to extract latent space parameters. The preprocessed multi-source heterogeneous data is used as a conditional constraint input into the initial geological model generator, which outputs simulated synthetic data. Forward modeling was performed using seismic wave forward modeling. The simulated composite data was compared with the measured data, and the hidden space parameters were optimized using error backpropagation.
3. The reservoir intelligent inversion method according to claim 1, characterized in that, A reservoir knowledge graph is constructed, and the initial geological model generator is adaptively pre-trained and transferred online to obtain an optimized geological model generator and its parameters, specifically including: Construct a reservoir knowledge graph; the reservoir knowledge graph contains reservoir entity attributes, model entity attributes and relationships between entities in the form of triples. The reservoir entity attributes include sedimentary facies, reservoir thickness, lithological distribution, elasticity distribution, physical property distribution and fluid distribution. The model entity attributes include model network parameters and applicable geological conditions. Based on the reservoir knowledge graph, similar model entities in the reservoir knowledge graph are retrieved according to the target reservoir type, and the parameters of the initial geological model generator are initialized and the latent space distribution prior is loaded using transfer learning. Using some labeled samples from the preprocessed multi-source heterogeneous data, the initial geological model generator is fine-tuned online. Combined with the seismic waveform matching loss optimization model, the optimized geological model generator and its parameters are obtained.
4. The reservoir intelligent inversion method according to claim 1, characterized in that, Based on the thinking chain technology and language large model technology, the task requirements analysis, task decomposition, and dynamic collaborative execution of reservoir modeling tasks are performed to obtain the iteratively optimized latent space parameters, specifically including: The reservoir modeling task was analyzed using language large modeling technology to determine the modeling objectives and constraints; Based on the modeling objectives and constraints, the reservoir modeling task is decomposed using the thinking chain technique to obtain the knowledge retrieval subtask, the model generation subtask, and the forward verification subtask. Based on the knowledge retrieval subtask, the model generation subtask, and the forward modeling verification subtask, the reservoir knowledge graph module is dynamically invoked to retrieve entity attributes and constraint rules related to the current task. The entity attributes and constraint rules related to the current task are then used as input conditions to activate the geological model generator to generate candidate reservoir models that conform to the constraint rules. Continuously monitor the execution status and results of the knowledge retrieval subtask, the model generation subtask, and the forward verification subtask; the execution status and results include data matching degree indicators; Based on the execution status and results, when the data matching index is less than a preset threshold, the sampling weight strategy of the latent space vector is dynamically adjusted to obtain the iteratively optimized latent space parameters. At the same time, the incremental update process of the reservoir knowledge graph is triggered to realize dynamic collaboration and closed-loop optimization between the reservoir knowledge graph module, the geological model generator, and the forward simulation module.
5. The reservoir intelligent inversion method according to claim 4, characterized in that, The constraint rules include geological constraint rules, physical constraint rules, and data constraint rules; the geological constraint rules are applicable conditions defined based on the reservoir entity attributes in the reservoir knowledge graph; the physical constraint rules are physical laws verified by seismic wave forward modeling; and the data constraint rules are standardization requirements for the input data.
6. The reservoir intelligent inversion method according to claim 1, characterized in that, The adaptive online pre-training process of the initial geological model generator includes an encoding stage and a decoding stage. The encoding stage stores abstract geological model knowledge by encoding latent space parameters, and the decoding stage integrates reservoir parameter prior information, well logging information, and well logging information to generate a geological model that meets the constraints.
7. A reservoir intelligent inversion device, characterized in that, The reservoir intelligent inversion device includes: The data acquisition and preprocessing module is used to acquire multi-source heterogeneous data from the target area and perform preprocessing. The seismic wave forward modeling optimization module is used to construct an initial geological model generator based on a conditional diffusion model, and to optimize the implicit space parameters of the initial geological model generator by using seismic wave forward modeling based on preprocessed multi-source heterogeneous data. An adaptive online pre-training and model transfer module is used to construct a reservoir knowledge graph and perform adaptive online pre-training and model transfer on the initial geological model generator to obtain an optimized geological model generator and optimized geological model generator parameters. The task decomposition and collaborative execution module is used to perform task requirement analysis, task decomposition and dynamic collaborative execution of reservoir modeling tasks based on thinking chain technology and language large model technology, and to obtain the latent space parameters after iterative optimization. The confidence interval feature extraction module is used to input the iteratively optimized latent space parameters into the optimized geological model generator, output a three-dimensional distribution model containing porosity, permeability and oil saturation, and extract confidence interval features. The reservoir model determination module is used to incrementally update the reservoir knowledge graph based on the three-dimensional distribution model, the confidence interval features, and the optimized geological model generator parameters, and to determine the reservoir model that conforms to the geological model features and matches the seismic waveform, thereby realizing intelligent reservoir inversion.
8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that the processor executes the computer program to implement the reservoir intelligent inversion method according to any one of claims 1-6.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the reservoir intelligent inversion method according to any one of claims 1-6.
10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the reservoir intelligent inversion method according to any one of claims 1-6.