Profile control and flooding well optimization method based on mapping knowledge domain

Abstract

The invention discloses a profile control and flooding well optimization method based on a knowledge graph, and belongs to the technical field of profile control and flooding. The method comprises the following steps: acquiring and processing structured data and unstructured data related to oilfield development; carrying out entity recognition and relation extraction on the unstructured document by utilizing a natural language processing technology; according to a processing result, constructing a profile control and flooding well field knowledge graph; a preferable profile control and flooding well position is screened out through multi-dimensional correlation analysis of the knowledge graph; and generating a profile control and flooding optimization scheme and a visual report in combination with well location characteristics. Through combination of natural language processing and a knowledge graph technology, deep mining and comprehensive analysis of oil field data are realized, so that scientificity and accuracy of profile control and flooding well optimization are improved, and time cost of manual analysis is remarkably reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of profile control and displacement driving, and particularly to a method for optimizing profile control and displacement driving wells based on a knowledge graph. Background Art

[0002] Profile control and displacement driving technology is an important means to improve oil recovery in the late stage of oilfield development. It mainly improves the utilization of remaining oil by adjusting the seepage field of the formation or the displacement process. However, due to the complexity of reservoir geological characteristics, the uneven layout of injection-production well patterns during the development process, and the uncertainty of fluid displacement, the optimization of profile control and displacement driving wells has always been a challenging task. The quality of the optimized wells is directly related to the implementation effect of profile control and displacement driving and the overall economic benefits of the oilfield. Therefore, it is of great significance to scientifically and reasonably select profile control and displacement driving wells.

[0003] In practical applications, the optimization of profile control and displacement driving wells needs to comprehensively consider various factors, such as the physical properties of the formation (porosity, permeability), reservoir fluid properties, injection-production well pattern relationships, historical injection-production data, and reservoir connectivity. The analysis of these factors often relies on a large amount of heterogeneous data, including structured data (such as production data, geological model parameters, etc.) and unstructured data (such as profile control and displacement driving operation reports, expert review opinions, etc.). However, existing analysis methods usually focus on the processing of a single data type, especially the analysis of structured data. For example, common methods include fitting analysis of injection-production curves, numerical simulation, and single-well parameter statistics. Although this method can reveal the key laws of profile control and displacement driving to a certain extent, it has important limitations.

[0004] On the one hand, unstructured data plays an important role in the optimization of profile control and displacement driving wells, but the existing methods have extremely low utilization rate for it. A large number of unstructured documents such as technical reports, meeting records, and profile control operation summaries during the oilfield development process contain rich practical experience and potential laws. However, due to the lack of standardization of these document data, traditional methods are difficult to directly process them and are often ignored or only used as manual references, which obviously cannot fully exploit their value. On the other hand, the multi-factor complexity of profile control and displacement driving well optimization also makes it difficult for single data analysis to comprehensively cover all influencing factors. Especially in complex reservoirs, the potential correlations and dynamic changes between various data are crucial for the optimization results, but traditional methods are difficult to achieve in-depth integration and correlation analysis of multi-source data.

[0005] Therefore, to address these challenges, there is an urgent need for an intelligent method that can integrate multi-source heterogeneous data. This method should have the ability to process both structured and unstructured data simultaneously, and reveal the key factors affecting profile control and displacement driving effects through multi-dimensional correlation analysis. Summary of the Invention

[0006] In view of this, the present invention aims to propose a method for optimizing profile control and displacement wells based on a knowledge graph, which can achieve efficient integration and analysis of multi-source heterogeneous data in oilfields, effectively solve problems such as low efficiency of manual analysis and low utilization rate of unstructured data in traditional methods, and improve the scientificity and accuracy of the process of optimizing profile control and displacement wells. The present invention can extract potential knowledge from unstructured texts by combining natural language processing technology, and use knowledge graph technology to conduct correlation analysis on multi-dimensional data, providing intelligent decision-making support for oilfield development.

[0007] To achieve the above object, the technical solution of the present invention is realized as follows: A method for optimizing profile control and displacement wells based on a knowledge graph includes the following steps:

[0008] Step 1: Collect multi-source heterogeneous data of oilfield development, including structured data and unstructured data; clean and standardize the collected structured data; uniformly store the processed data in a database;

[0009] Step 2: Use natural language processing technology to process unstructured documents, and convert the processing results into structured triple data to provide input for knowledge graph construction;

[0010] Step 3: Based on the above-extracted structured triple data, construct a knowledge graph related to profile control and displacement; use a graph database to store the knowledge graph to achieve efficient query and analysis of data;

[0011] Step 4: Use the knowledge graph for intelligent analysis, comprehensively consider reservoir connectivity, injection-production relationship and historical effects through graph algorithms to identify preferred well positions suitable for profile control and displacement; generate a well position recommendation list according to the analysis results, and propose an optimization plan in combination with technical parameters related to profile control and displacement; finally generate a visual report to provide decision-making support for oilfield development.

[0012] Further, Step 2 includes:

[0013] Step 2.1: Before processing unstructured documents, it is necessary to preprocess the unstructured document data;

[0014] Step 2.2: Use entity recognition technology to identify and extract key information in unstructured documents;

[0015] Step 2.3: Apply relationship extraction technology to extract implicit relationships between entities in unstructured documents;

[0016] Step 2.4: Convert the extracted key information and implicit relationships between entities into structured triple data to provide input for subsequent knowledge graph construction.

[0017] Further, in step 2.1, preprocess the unstructured documents, including:

[0018] Step 2.1.1, collect unstructured data related to profile control and displacement, and unify the format;

[0019] Step 2.1.2, clean redundant information in the documents, and remove duplicate paragraphs and blank documents;

[0020] Step 2.1.3, use the domain dictionary and Chinese word segmentation tool to segment the documents.

[0021] Further, step 2.2 includes:

[0022] Step 2.2.1, domain entity definition and data annotation;

[0023] Step 2.2.2, use the BERT module and BiLSTM module to capture the context and form the input representation;

[0024] Step 2.2.3, use the pytorch framework to implement an entity recognition model based on word pair relationships;

[0025] Among them, step 2.2.3 includes:

[0026] Construction of word pair relationship representation;

[0027] Generation and optimization of the relationship matrix;

[0028] For the optimized relationship matrix, jointly use double affine transformation and multi-layer perceptron for relationship classification prediction;

[0029] Use the predefined relationship categories for decoding, and extract entities from the relationship matrix.

[0030] Further, step 2.3 includes:

[0031] Step 2.3.1, construction of the label matrix;

[0032] Step 2.3.2, generation of the label matrix;

[0033] Step 2.3.3, triple decoding;

[0034] Among them, step 2.3.3 includes:

[0035] Entity extraction;

[0036] Subject-object relationship decoding;

[0037] Relationship verification.

[0038] Further, step 3 includes:

[0039] Step 3.1: Identify the key entities and relationship types involved in the knowledge graph, and design the overall structural framework of the knowledge graph to provide basic support for the standardization and systematicness of knowledge representation;

[0040] Step 3.2: Through the data preparation of the knowledge graph in Step 2, convert structured and unstructured data into the input data required for the knowledge graph to provide basic support for graph construction;

[0041] Step 3.3: Based on the data preparation in Step 3.2, use graph database technology to implement the construction and storage of the knowledge graph, and improve the efficiency of query and access through index optimization;

[0042] Step 3.4: Verify the accuracy and practicality of the knowledge graph through visualization tools and actual data, and optimize its reasoning logic and query capabilities.

[0043] Furthermore, in Step 3.1, designing the overall structural framework of the knowledge graph includes,

[0044] Definition of core elements;

[0045] Structure design;

[0046] In Step 3.3, implementing the storage of the knowledge graph includes,

[0047] Use tools to generate structured representations of nodes and edges, and import the cleaned entities and relationships into the graph database;

[0048] Adopt distributed storage and index optimization technologies to support the efficient access and intelligent analysis of large-scale data;

[0049] In Step 3.4, verifying the accuracy and practicality of the knowledge graph through visualization tools and actual data includes,

[0050] Provide an intuitive interactive interface to display the structure and entity relationships of the knowledge graph;

[0051] Compare the reasoning results of the knowledge graph with historical data to ensure its reliability, and adjust and optimize according to expert suggestions.

[0052] Furthermore, Step 4 includes:

[0053] Step 4.1: Use the knowledge graph for intelligent analysis;

[0054] Step 4.2: Based on the above analysis results, classify and sort potential preferred well positions, and generate a well position recommendation list;

[0055] Step 4.3: According to the generated list of preferred well positions, propose an optimized implementation plan in combination with relevant parameters of profile control and flooding;

[0056] Step 4.4: Generate a visualization report.

[0057] Further, Step 4.1 includes:

[0058] Utilize path search and influence analysis algorithms;

[0059] Conduct connectivity analysis;

[0060] Conduct historical effect analysis;

[0061] In Step 4.2, generate a well location recommendation list, including:

[0062] Combined with geological and production actual situations, set the core parameters for well location optimization, including connectivity score, injection-production relationship strength, and historical profile control effect;

[0063] Score and rate the well locations to ensure that the recommended well locations have good adaptability under different optimization conditions;

[0064] Screen out the optimized well location set suitable for the current reservoir development strategy.

[0065] Further, Step 4.3 includes:

[0066] Optimize injection-production parameters;

[0067] Formulate a construction plan;

[0068] Step 4.4 includes:

[0069] Through the knowledge graph visualization function, generate a dynamic graph including well location association relationships, profile control effect distribution, and injection-production network layout;

[0070] Generate a comprehensive report including a well location optimization list, profile control parameter suggestions, and expected effect evaluation;

[0071] Provide a flexible data interaction interface to support development management personnel to customize the report content according to requirements.

[0072] Compared with the prior art, the profile control and drive well optimization method based on a knowledge graph of the present invention has the following advantages: The present invention proposes a profile control and drive well optimization method based on a knowledge graph. This method makes full use of natural language processing technology to mine potential knowledge in unstructured documents and realizes the efficient integration and correlation analysis of multi-source heterogeneous data through the knowledge graph, thereby significantly improving the scientificity and efficiency of well location optimization. Through automated optimization analysis and plan generation, the risk of manual intervention and decision-making errors is reduced, providing a reliable and intelligent solution for the development of complex reservoirs, helping to improve the recovery rate and reduce the development cost.

[0073] (1) The present invention can effectively identify domain entities in complex documents, including nested and discontinuous structures; by combining deep learning and semantic rules, it can significantly improve the recognition accuracy and efficiency; provide high-quality entity inputs for the knowledge graph, and enhance the reliability of subsequent data analysis;

[0074] (2) The present invention can solve the overlapping problem and avoid the cascading errors in traditional methods; through a unified matrix marking scheme, it greatly improves the decoding calculation efficiency; the design of the joint marking matrix significantly improves the extraction accuracy;

[0075] (3) The present invention constructs a domain knowledge graph with comprehensive functions and clear structure, realizes the efficient storage, analysis and intelligent reasoning of multi-dimensional data for profile control and flooding control, significantly improves the scientificity and efficiency of the optimization process, and provides reliable technical support for oilfield development;

[0076] (4) The present invention can quickly and accurately select well positions suitable for profile control and flooding control from complex reservoir multi-source data, and provide scientific and reasonable optimization schemes;

[0077] (5) The present invention integrates multi-source data correlation analysis and intelligent optimization strategies, and through the deep integration of knowledge graph and natural language processing technologies, solves the problems of well position selection in traditional reservoir development relying on manual experience, low data correlation, and insufficient analysis efficiency; to improve the adaptability and accuracy of the system, the present invention introduces multi-dimensional graph algorithms, natural language processing modules and automated analysis tools for the problems of heterogeneity, dynamics and uneven data distribution in reservoir data, ensuring that the well position selection process is more scientific and comprehensive;

[0078] (6) The present invention constructs a dynamic knowledge graph, and through knowledge graph related technologies, ensures the accuracy of well position selection; for the possible noise in the data and the ambiguity of historical records, it integrates entity recognition and relationship extraction technologies based on various feature information, thereby improving the reliability and practicability of the knowledge graph; these technological innovations significantly improve the efficiency and accuracy of well selection for profile control and flooding control wells, and provide solid technical support for oilfield development decisions;

[0079] (7) Compared with traditional manual screening and empirical judgment methods, the present invention effectively overcomes the problems of high labor cost, long analysis cycle and poor result stability. Using natural language processing technology to extract and structure key information in unstructured documents, it solves the pain points of high information redundancy and low processing efficiency in traditional text analysis. This comprehensive optimization method integrating natural language processing and knowledge graph technologies demonstrates excellent reliability and application value in the recommendation of oilfield development well positions, bringing significant economic and management benefits to actual production. Description of the Drawings

[0080] The accompanying drawings, which form a part of the present invention, are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the accompanying drawings:

[0081] Figure 1 is the overall flowchart of the method for optimizing profile control and displacement wells based on a knowledge graph according to an embodiment of the present invention;

[0082] Figure 2 is the logical block diagram of the method for optimizing profile control and displacement wells based on a knowledge graph according to an embodiment of the present invention;

[0083] Figure 3 is the entity recognition model structure diagram;

[0084] Figure 4 is the relationship extraction model structure diagram;

[0085] Figure 5 is the flowchart for constructing the knowledge graph of the profile control and displacement well field;

[0086] Figure 6 is the flowchart of the optimization and application of profile control and displacement wells based on a knowledge graph. Detailed implementation manners

[0087] It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments may be combined with each other.

[0088] The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

[0089] The present invention relates to a method for optimizing profile control and displacement wells based on a knowledge graph. By using natural language processing technology, key information can be automatically extracted from unstructured documents, and in combination with knowledge graph technology, an association network between data can be established, thereby providing new technical support for the optimization of profile control and displacement wells. This method can not only improve the data processing efficiency, but also convert manual experience into available knowledge resources, provide a scientific basis for well location optimization, and further improve the implementation effect of profile control and displacement and the overall development efficiency of the oilfield.

[0090] Embodiment 1, as Figure 1 shown, a method for optimizing profile control and displacement wells based on a knowledge graph includes the following steps:

[0091] Step 1: Collect and process multi-source data for oilfield development, specifically including:

[0092] Step 1.1: Collect multi-source heterogeneous data for oilfield development, including structured data and unstructured data. Structured data mainly comes from numerical data such as injection-production data, pressure distribution, reservoir characteristics, and lithology distribution during the reservoir production process; unstructured data includes text or image data such as technical documents related to profile control and flooding control, construction operation reports, and oilfield geological analysis reports.

[0093] Step 1.2: Clean and standardize the collected structured data, including removing missing values and outliers, filling in null values, and unifying data formats and units to ensure data consistency and usability; at the same time, preprocess unstructured data using natural language processing techniques, such as word segmentation, denoising, and format conversion, to provide support for subsequent knowledge extraction and analysis.

[0094] Step 1.3: Establish a unified database system, store the processed structured data and unstructured data in the central database, and lay a data foundation for subsequent analysis. The database can support efficient query and multi-dimensional data access, ensuring fast and reliable data retrieval during subsequent analysis.

[0095] It should be noted that through the systematic collection and standardization of multi-source data, this step not only lays a solid data foundation for knowledge extraction and graph construction, but also improves the accuracy and integrity of the data, providing reliable support for the optimization of profile control and flooding control well positions in oilfield development.

[0096] Step 2: Use natural language processing techniques to process unstructured documents related to profile control and flooding control, including extracting entity information related to profile control and flooding control in the documents through entity recognition technology and identifying implicit relationships between entities in the documents using relationship extraction technology; convert the above processing results into structured triple data required for knowledge graph construction;

[0097] Step 3: Based on the structured triple data extracted in Step 2, complete the construction of the profile control and flooding control knowledge graph;

[0098] Step 4: Based on the knowledge graph constructed in Step 3, conduct intelligent analysis to support the optimization decision-making of profile control and flooding control. According to the generated optimal analysis, recommend profile control and flooding control well positions, combine with the technical parameter optimization plan, and support oilfield development decision-making through visual reports.

[0099] Embodiment 2: This embodiment further limits a method for optimizing profile control and flooding control wells based on a knowledge graph described in Embodiment 1. In this embodiment, the content described in Step 2 is refined, as Figure 2 shown, and specifically includes the following steps:

[0100] Step 2.1: Before processing unstructured documents, it is necessary to preprocess the document data to improve the accuracy and efficiency of subsequent natural language processing steps.

[0101] Step 2.1.1: Collect unstructured data related to profile control and displacement, including technical documents, operation reports, expert evaluation reports, and geological research materials. Unify the formats of the collected documents (such as converting formats like PDF and Word into text format) for subsequent processing.

[0102] Step 2.1.2: Use techniques such as regular expressions to clean irrelevant symbols, HTML tags, and noise characters in the documents. Remove duplicate paragraphs and redundant annotations in the documents to ensure that the text content is concise and clear.

[0103] Step 2.1.3: Adopt Chinese word segmentation tools and combine with domain dictionaries for word segmentation to accurately segment professional terms such as well names and reservoir names. Use sentence segmentation algorithms to divide the document content into sentence paragraphs for subsequent semantic analysis.

[0104] Step 2.2: Entity recognition is the core technology for extracting key information (such as well names, reservoir names, parameter values) from unstructured documents. To adapt to the complex entity types in the profile control and displacement scenarios, the present invention adopts a unified entity recognition technology based on word features, formalizes the entity recognition problem as a classification task of word pair relationships, to ensure the accuracy and diversity of the recognition results. As Figure 3 shown, it specifically includes the following contents:

[0105] Step 2.2.1: Domain entity definition and data annotation

[0106] According to the business requirements of the profile control and displacement scenarios, clarify the domain entity categories, and use the BIOHD tag scheme (B - Begin, I - Inside, O - Outside, H - Head, D - Discontinuous) to handle the complex structures of domain entities.

[0107] Step 2.2.2: Input representation

[0108] The input text sequence X = {x1, x2, …, x N}, and each word x i is represented by word embedding . Use the pre - trained language model BERT to obtain context embeddings H = {h1, h2, …, h N}, and the formula is as follows:

[0109] h i = BERT(x i ) + BiLSTM(h i-1 , h i+1 )

[0110] Among them, BFRT(·) captures global context features and further enhances the context feature capture ability in combination with BiLSTM(·).

[0111] Step 2.2.3: Model implementation

[0112] Step 2.2.3.1: Construction of word pair relationship representation

[0113] To better capture the semantic relationship between words, the present invention adopts a word pair relationship matrix to represent the relationships of all word pairs:

[0114] R ij = Concat(h i .h j , d ij , f ij )

[0115] Among them, h i and h j are the context representations of words x i and x j respectively; d ij is the relative distance embedding, representing the distance between words x i and x j ; f ij represents additional semantic features.

[0116] Step 2.2.3.2: Optimization of the relationship matrix

[0117] Use a multi-granularity convolutional network to extract features from the relationship matrix R and capture the semantic connection between word pairs:

[0118] Q l = σ(DConv(R, l))

[0119] Q = Concat(Q1, Q2,..., Q L )

[0120] Among them, l is the dilation rate, and different dilation rates capture short-distance and long-distance relationships; σ is the activation function; DConv is the multi-granularity convolutional network; R is the relationship matrix; Q l represents the convolutional output with a dilation rate of l. The final output Q represents the optimized relationship matrix.

[0121] Step 2.2.3.3: Relationship classification prediction

[0122] On the optimized Q, calculate the relationship distribution of each word pair as follows:

[0123] P(R ij) = Softmax(W biaffine ·Q ij +W MLP ·Q ij )

[0124] where \(W\ biaffine \) and \(W\ MLP \) are the parameter matrices of the bi - affine transformation and the multi - layer perceptron respectively. softmax(·) is used to normalize the probability distribution, and \(Q\ ij \) represents the relationship matrix of the word pair \((x\ i , x\ j ). Therefore, the bi - affine transformation and the multi - layer perceptron jointly complete the joint prediction of relationships to improve the prediction accuracy.

[0125] Step 2.2.3.4: Decoding

[0126] To uniformly solve the complex entity types in the profile control and flooding control scenarios, the present invention adopts two categories, namely the Next Neighborhood Word (NNW) and the Tail Head Word (THW), in the word pair relationship classification task. Among them, NNW means that two words belong to the same entity and are adjacent to each other; THW means that a word is the head or tail boundary of an entity. Specifically, the THW relationship is used to mark the hierarchy, and the discontinuous segments are connected through the NNW relationship. Therefore, combining the above two word pair relationships, entities are extracted from the relationship matrix.

[0127] Entity = {Path(x\ i , x\ j ) | R\ ij ∈ NNW}

[0128] Through the above process, Step 2.2 shows significant advantages in the following aspects: effectively identifying domain entities in complex documents, including nested and discontinuous structures; combining deep learning and semantic rules to significantly improve the recognition accuracy and efficiency; providing high - quality entity inputs for the knowledge graph and enhancing the reliability of subsequent data analysis.

[0129] Step 2.3: Relationship extraction aims to identify entity pairs and their semantic relationships from unstructured text and is one of the important links in knowledge graph construction. This step uses the method of decoding the tag matrix to formalize the relationship extraction problem as a word pair linking task, which can efficiently solve the problems of Single Entity Overlap (SEO) and Entity Pair Overlap (EPO), and achieve high - precision modeling of multi - relationship extraction. As Figure 4 shown, it specifically includes the following content:

[0130] Step 2.3.1: Marking Matrix Construction

[0131] The present invention models relation extraction as three marking tasks, corresponding to three types of marking matrices respectively:

[0132] ① Entity Head-to-Tail Marking Matrix (EH-to-ET): Mark the start and end word pairs of each entity;

[0133] ② Subject Head-to-Object Head Marking Matrix (SH-to-OH): Mark the start word positions of each pair of subject and object entities;

[0134] ③ Subject Tail-to-Object Tail Marking Matrix (ST-to-OT): Mark the end word positions of each pair of subject and object entities.

[0135] For each pair of word pairs (w i , w j ) in the sentence, construct a marking M ij = l, where l represents the marking category of this word pair. For example, in the EH-to-ET matrix, M ij = 1 indicates that words w i and w j are the head and tail boundaries of the same entity.

[0136] Step 2.3.2: Marking Matrix Generation

[0137] For each pair of words (w i , w j ), calculate its feature vector h ij :

[0138] h ij = tanh(W h · [h i ; h j + b h )

[0139] where h i and h j represent word embeddings, ; represents the concatenation operation, and W h and b h represent learnable parameters.

[0140] Use a unified classifier to predict the marking category of the word pair:

[0141] P(y ij ) = Softmax(W o · h ij + b o )

[0142] where W o and b o represent learnable parameters, and P(y ij) represents the probability distribution of word pairs (w i , w j ) belonging to various categories.

[0143] Subsequently, the matrix is sparsified, and the invalid regions in the lower triangular matrix are set to zero to save memory. At the same time, a marking scheme is used to flatten the valid regions into a one-dimensional sequence for efficient calculation.

[0144] Step 2.3.3: Triple decoding, that is, implementing the decoding strategy of the relation extraction model.

[0145] Decode triples from the marking matrix. The main steps include:

[0146] ① Entity extraction: Decode all entity boundaries through the EH-to-ET matrix;

[0147] ② Subject-object relationship decoding: Use the SH-to-OH matrix and the ST-to-OT matrix to locate the start and end words of the subject and object respectively;

[0148] ③ Relationship verification: Combine the three marking matrices to verify whether the subject and object conform to the relationship r.

[0149] The decoding process can be formally described by the following algorithm:

[0150] T = {(s, r, o) | s ∈ EH-to-ET, o ∈ EH-to-ET, (s, o)

[0151] ∈ SH-to-OH ∩ ST-to-OT}

[0152] Through the above process, Step 2.3 shows significant advantages in the following aspects: solving the overlap problem and avoiding cascading errors in traditional methods; greatly improving the decoding calculation efficiency through a unified matrix marking scheme; and significantly improving the extraction accuracy through the design of the combined marking matrix.

[0153] Step 2.4: Output structured data

[0154] Finally, the processing result is converted into triple data and stored in the graph database for subsequent construction and analysis of the knowledge graph.

[0155] Embodiment 3, this embodiment further limits the method for optimizing profile control and displacement wells based on a knowledge graph described in Embodiment 1. In this embodiment, the construction of the knowledge graph described in Step 3 is refined to ensure the comprehensiveness and accuracy of the graph construction. As Figure 5 shown, it specifically includes the following steps:

[0156] Step 3.1: Identify the key entities and relationship types involved in the knowledge graph, and design the overall structural framework of the graph to provide basic support for the standardization and systematicness of knowledge representation.

[0157] Step 3.1.1: Definition of Core Elements

[0158] Identify the types of entities involved, their attributes, and their mutual relationships. Standardize the expression methods of entity attributes and relationships to provide data specifications for subsequent graph construction.

[0159] Step 3.1.2: Structure Design

[0160] Determine the hierarchical structure of the graph, including the data layer, logic layer, and application layer, to form an efficient storage and analysis logic.

[0161] Step 3.2: Through the steps in Method 2 of the implementation, prepare the data for the knowledge graph, convert structured and unstructured data into the input data required by the knowledge graph, and provide basic support for graph construction.

[0162] Step 3.3: Based on the data preparation in Step 3.2, use graph database technology to implement the construction and storage of the knowledge graph, and improve the efficiency of querying and accessing through index optimization.

[0163] Step 3.3.1: Graph Construction

[0164] Use tools to generate a structured representation of nodes and edges, and import the cleaned entities and relationships into the graph database.

[0165] Step 3.3.2: Storage and Optimization

[0166] Adopt distributed storage and index optimization technologies to support the efficient access and intelligent analysis of large-scale data.

[0167] Step 3.4: After the above steps are completed, verify the accuracy and practicality of the knowledge graph through visualization tools and actual data, and optimize its reasoning logic and query capabilities.

[0168] Step 3.4.1: Graph Visualization

[0169] Provide an intuitive interactive interface to display the structure and entity relationships of the knowledge graph.

[0170] Step 3.4.2: Graph Verification

[0171] Compare the reasoning results of the knowledge graph with historical data to ensure its reliability, and adjust and optimize according to expert suggestions.

[0172] Through the above process, the present invention constructs a domain knowledge graph with comprehensive functions and clear structure, realizes the efficient storage, analysis and intelligent reasoning of multi-dimensional data for profile control and displacement driving, significantly improves the scientificity and efficiency of the optimization process, and provides reliable technical support for oilfield development.

[0173] Embodiment 4. This embodiment further limits a method for optimizing profile control and displacement driving wells based on a knowledge graph described in Embodiment 1. In this embodiment, the well location optimization and plan generation described in Step 4 are refined to more accurately identify and recommend the optimal well locations for profile control and displacement driving. As Figure 6 shown, it specifically includes the following steps:

[0174] Step 4.1: Use the knowledge graph for intelligent analysis. Through the knowledge graph, conduct correlation modeling on multi-source data of the oilfield, and establish the correlation between entities such as wells, reservoirs, faults, and multi-dimensional information such as injection-production relationships, connectivity, and profile control effects.

[0175] Step 4.1.1: Application of graph algorithms

[0176] Path search analysis: By searching for the shortest path between wells, evaluate the connectivity of injection-production wells, and identify key injection-production nodes with potential impacts in the reservoir.

[0177] Influence analysis: Rank the influence of injection-production wells in the knowledge graph. By comparing the injection volume, recovery volume, and reservoir connectivity characteristics of each well, screen out the candidate well locations with the greatest influence in the overall development.

[0178] Step 4.1.2: Connectivity analysis

[0179] Analyze the physical connectivity between wells through reservoir attribute data and profile control and displacement driving effect data, and identify the preferred fluid flow paths. Combine the dynamic data of injection-production wells, evaluate the potential oil displacement areas after profile control, and optimize the well location combination with the best linkage effect between injection and recovery.

[0180] Step 4.1.3: Historical effect analysis

[0181] Call the historical profile control and displacement driving operation records in the knowledge graph, and compare and analyze the profile control effect indicators of different wells (such as recovery rate improvement, injection-production ratio change). Use the time series model to predict the potential benefits of future profile control and displacement driving, and provide a quantitative reference for well location optimization.

[0182] Step 4.2: Based on the above analysis results, classify and rank the potential preferred well positions, and generate a well position recommendation list. Considering the geological and production actual situations, set the core parameters for well position optimization, including connectivity score, injection-production relationship strength, and historical profile control effect. Rate and grade the well positions to ensure that the recommended well positions have good adaptability under different optimization conditions. Finally, screen out the set of preferred well positions suitable for the current reservoir development strategy.

[0183] Step 4.3: According to the generated list of preferred well positions, propose an optimized implementation plan in combination with the relevant parameters of profile control and flooding.

[0184] Step 4.3.1: Optimization of injection-production parameters

[0185] Based on the reservoir connectivity and historical injection-production data of the preferred wells, calculate the reasonable injection dosage, injection rate, and injection pressure to ensure the maximization of the profile control effect.

[0186] Step 4.3.2: Formulation of construction plan

[0187] Formulate the profile control and flooding construction plan for the preferred wells, including the type of injected substances, injection time, and construction sequence. Provide a detailed construction feasibility analysis to ensure the implementation of the optimization plan has sufficient safety and economy.

[0188] Step 4.4: Generation of visual report

[0189] Convert the preferred results and optimization plan into visual charts and analysis reports to provide an intuitive decision-making support tool for oilfield development management. Through the knowledge graph visualization function, generate dynamic graphs including well position association relationships, profile control effect distribution, and injection-production network layout. Use map projection technology to display the geographical distribution of the preferred well positions and reservoir connectivity. Automatically generate a comprehensive report including the list of preferred well positions, profile control parameter suggestions, and expected effect evaluation. At the same time, provide a flexible data interaction interface to support development management personnel to customize the report content according to their needs.

[0190] Through the above detailed steps, the present invention can quickly and accurately select the well positions suitable for profile control and flooding from complex reservoir multi-source data, and provide a scientific and reasonable optimization plan.

[0191] In summary, by utilizing the intelligent analysis ability of the knowledge graph and the optimization ability of graph algorithms, the present invention constructs an intelligent method for optimizing profile control and flooding well positions and generating a plan, providing a new technical means for oilfield development.

[0192] In several embodiments, the system architecture, functional modules, and structured components mentioned in the present invention are all designed to implement the specific steps in the method embodiments described above. Since these implementation details have been elaborated in detail in the above method embodiments, to avoid repetition, these specific implementation manners will not be further expanded here.

[0193] The method for optimizing the selection of profile control and displacement wells based on a knowledge graph proposed by the present invention combines multi-source data correlation analysis and intelligent optimization strategies. Through the deep integration of knowledge graph and natural language processing technologies, it solves the problems in traditional reservoir development such as well location selection relying on manual experience, low data correlation, and insufficient analysis efficiency. To improve the adaptability and accuracy of the system, the present invention introduces multi-dimensional graph algorithms, natural language processing modules, and automated analysis tools for the problems of heterogeneity, dynamics, and uneven data distribution in reservoir data, ensuring that the well location selection process is more scientific and comprehensive.

[0194] The present invention particularly focuses on solving the complex problems existing in actual reservoir development. For example, to address the problem that the correlation relationship between multi-source data is difficult to reveal, the present invention constructs a dynamic knowledge graph and ensures the accuracy of well location selection through knowledge graph-related technologies; for the possible noise in the data and the ambiguity of historical records, entity recognition and relationship extraction technologies based on various feature information are incorporated, thereby improving the reliability and practicality of the knowledge graph. These technological innovations significantly improve the efficiency and accuracy of profile control and displacement well selection, providing strong technical support for oilfield development decision-making.

[0195] Compared with traditional manual screening and empirical judgment methods, the present invention effectively overcomes the problems of high labor cost, long analysis cycle, and poor result stability. By using natural language processing technology to extract and structure the key information in unstructured documents, it solves the pain points of high information redundancy and low processing efficiency in traditional text analysis. This comprehensive optimization method integrating natural language processing and knowledge graph technologies demonstrates excellent reliability and application value in the recommendation of oilfield development well locations, bringing significant economic and management benefits to actual production.

[0196] The above content details various possible implementation manners of the present invention. These embodiments are only examples of technical applications and do not exhaust all applicable scenarios. Those skilled in the art can adjust and optimize the specific implementation manners based on the basic principles and innovative concepts of the present invention without violating the core technical ideas. For example, the construction of knowledge graphs under different reservoir conditions, the corpus optimization of natural language processing modules, or the customized adjustment of graph algorithms all belong to the reasonable extension of the present invention.

[0197] In addition, the terms and expressions used in this article are intended to convey the core ideas, working principles, and technological advancements of the present invention in a clear manner, so as to ensure that those skilled in the art can fully understand and implement these technical solutions. Any variant or extended solution based on the core content of the present invention shall be regarded as falling within the scope of protection of the present invention.

Claims

1. A method for optimizing profile control and flooding wells based on knowledge graph, characterized in that: The following steps are involved: Step 1: Collect multi-source heterogeneous data of oilfield development, including structured data and unstructured data; clean and standardize the collected structured data; The processed data is uniformly stored in the database; Step 2: Use natural language processing technology to process unstructured documents and convert the processing results into structured triple data to provide input for knowledge graph construction; Step 3: Based on the structured triple data extracted above, a knowledge graph related to profile control and flooding control is constructed; a graph database is used to store the knowledge graph to achieve efficient query and analysis of data; Step 4: Use the knowledge graph for intelligent analysis, and use graph algorithms to comprehensively consider reservoir connectivity, injection-production relationships, and historical effects to identify optimal well locations suitable for profile adjustment and flooding. Generate a recommended list of well locations based on the analysis results, and propose an optimization plan based on the technical parameters related to profile adjustment and flooding. Finally, generate a visual report to provide decision support for oilfield development.

2. A profile control and flooding well optimization method based on knowledge graph according to claim 1, characterized in that: Step 2 includes: Step 2.1: Before processing the unstructured document, the unstructured document data needs to be preprocessed; Step 2.2: Use entity recognition technology to identify and extract key information from unstructured documents; Step 2.3: Apply relation extraction technology to extract implicit relations between entities in unstructured documents; Step 2.4: Convert the extracted key information and implicit relationships between entities into structured triple data to provide input for subsequent knowledge graph construction.

3. A profile control and flooding well optimization method based on knowledge graph according to claim 2, characterized in that: In step 2.1, the unstructured document is preprocessed, including: Step 2.1.1, collect unstructured data related to profile control and flooding control and unify the format; Step 2.1.2, clean up the redundant information in the document, remove duplicate paragraphs and blank documents; Step 2.1.3: Use domain dictionaries and Chinese word segmentation tools to segment the document.

4. A profile control and flooding well optimization method based on knowledge graph according to claim 2, characterized in that: Step 2.2 includes: Step 2.2.1, domain entity definition and data annotation; Step 2.2.2, use the BERT module and BiLSTM module to capture the context and form the input representation; Step 2.2.3, use the pytorch framework to implement the entity recognition model based on word pair relationships; Among them, step 2.2.3 includes: Word pair relationship representation construction; Relationship matrix generation and optimization; For the optimized relationship matrix, the double affine transformation and multi-layer perceptron are used to jointly perform relationship classification prediction; Entities are extracted from the relation matrix using decoding using predefined relation categories.

5. The method for optimizing profile control and flooding wells based on knowledge graph according to claim 2, characterized in that: Step 2.3 includes: Step 2.3.1, marker matrix construction; Step 2.3.2, marker matrix generation; Step 2.3.3, triple decoding; Among them, step 2.3.3 includes: Entity extraction; Decoding of subject-object relations; Relationship verification.

6. The method for optimizing profile control and flooding wells based on knowledge graph according to claim 1, characterized in that: Step 3 includes: Step 3.1: Identify the key entities and relationship types involved in the knowledge graph, and design the overall structural framework of the knowledge graph to provide basic support for the standardization and systematicness of knowledge representation; Step 3.2: Prepare the data for the knowledge graph through step 2, convert the structured and unstructured data into the input data required for the knowledge graph, and provide basic support for graph construction; Step 3.3: Based on the data preparation in step 3.2, use graph database technology to build and store knowledge graphs, and improve query and access efficiency through index optimization; Step 3.4: Verify the accuracy and practicality of the knowledge graph through visualization tools and actual data, and optimize its reasoning logic and query capabilities.

7. A profile control and flooding well optimization method based on knowledge graph according to claim 6, characterized in that: In step 3.1, design the overall structural framework of the knowledge graph. include, Definition of core elements; Structural design; In step 3.3, the storage of knowledge graph is realized. include, Use tools to generate structured representations of nodes and edges, and import the cleaned entities and relationships into the graph database; Adopt distributed storage and index optimization technology to support efficient access and intelligent analysis of large-scale data; In step 3.4, the accuracy and practicality of the knowledge graph are verified through visualization tools and actual data, including: Provide an intuitive interactive interface to display the structure and entity relationships of the knowledge graph; Compare the reasoning results of the knowledge graph with historical data to ensure its reliability, and adjust and optimize it according to expert advice.

8. The method for optimizing profile control and flooding wells based on knowledge graph according to claim 1, characterized in that: Step 4 includes: Step 4.1: Use knowledge graph for intelligent analysis; Step 4.2: Based on the above analysis results, classify and sort the potential optimal well locations, and generate a well location recommendation list; Step 4.3: Based on the generated list of preferred well locations, an optimization implementation plan is proposed in combination with relevant parameters of profile control and flooding; Step 4.4: Generate a visualization report.

9. A profile control and flooding well optimization method based on knowledge graph according to claim 8, characterized in that: Step 4.1 includes: Using path search and influence analysis algorithms; Connectivity analysis; Historical effect analysis; In step 4.2, a list of recommended well locations is generated, including: Based on the actual geological and production conditions, the core parameters for well location optimization are set, including connectivity score, injection-production relationship strength, and historical profile control effects; Score and rate the well locations to ensure that the recommended well locations have good adaptability under different optimal conditions; Screen out the optimal well location set suitable for the current reservoir development strategy.

10. The method for optimizing profile control and flooding wells based on knowledge graph according to claim 8, characterized in that: Step 4.3 includes: Optimize injection and production parameters; Develop construction plans; Step 4.4 includes: Through the knowledge graph visualization function, a dynamic graph including well location association, profile control effect distribution and injection-production network layout is generated; Generate a comprehensive report including a list of well site optimization, profile control parameter recommendations, and expected effect evaluation; Provides a flexible data interaction interface to support development managers to customize report content according to needs.

Images