A method and system for evaluating a tight sandstone reservoir based on a knowledge graph and a storage medium

Abstract

This invention discloses a knowledge graph-based method, system, and storage medium for evaluating tight sandstone reservoirs, relating to the field of geological exploration and development technology. It addresses the problems of data fragmentation and static models often found in existing evaluation methods. The method includes: Step S100: Collecting tight sandstone reservoir evaluation data, constructing an ontology knowledge base, and building an ontology layer for the tight sandstone reservoir domain based on knowledge and its relationships; Step S200: Constructing a tight sandstone reservoir evaluation model, which includes a knowledge extraction module and a knowledge reasoning module. The knowledge extraction module extracts triples from entity relationships labeled with tight sandstone reservoir sequences, constructing a knowledge graph of tight sandstone reservoirs with joint entity relationships; the knowledge reasoning module uses a rule-embedded structure-aware convolutional network for multi-task collaborative reasoning; Step S300: Performing multi-task collaborative reasoning of the tight sandstone reservoir based on the evaluation model.

Description

Technical Field

[0001] This invention relates to the field of geological exploration and development technology, and more specifically, to a knowledge graph-based method, system, and storage medium for evaluating tight sandstone reservoirs. Background Technology

[0002] Tight reservoirs, as an important exploration area for tight oil and gas resources, are characterized by strong heterogeneity and complex pore structures. Conventional oil and gas reservoir production operations and exploration methods are insufficient for identifying the geological types of tight sandstone reservoirs. During the evaluation process, the significant variations in reservoir characteristics across different regions and even within the same basin necessitate different parameter configurations and model selections to improve the accuracy of reservoir evaluations. Furthermore, the differing evaluation methods and standards employed by various scholars and research institutions make it difficult to unify and compare evaluation results. Therefore, integrating multi-source data from geology, logging, and experiments to construct a comprehensive reservoir characteristic knowledge base, and combining this with technologies such as artificial intelligence to design intelligent tight sandstone reservoir evaluation methods, has become a key research focus.

[0003] As knowledge graphs continue to expand in scale and their applications deepen, graph neural networks employ a unique method for describing nodes, obtaining states that encompass both neighbor node information and graph topological features. By performing corresponding convolutional learning on the graph and aggregating relevant features and information, the prediction process becomes more interpretable. This approach effectively considers both semantic and structural information of the knowledge graph simultaneously, particularly capturing its structural information. Existing evaluation methods often suffer from data fragmentation and model staticity. Since the evaluation of tight sandstone reservoirs requires comprehensive consideration of reservoir dynamics and actual conditions, human decision-making is easily limited by subjective experience, resulting in low accuracy and stability. Summary of the Invention

[0004] The technical problem to be solved by this invention is:

[0005] Existing evaluation methods often suffer from data fragmentation and static models, while manual evaluation processes are easily influenced by subjective experience.

[0006] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0007] This invention provides a knowledge graph-based method for evaluating tight sandstone reservoirs, comprising the following steps:

[0008] Step S100: Collect tight sandstone reservoir evaluation data, preprocess the tight sandstone reservoir evaluation data, construct an ontology knowledge base, and construct an ontology layer for the tight sandstone reservoir domain based on knowledge and the relationships between them.

[0009] Step S200: Construct a tight sandstone reservoir evaluation model. The evaluation model includes a knowledge extraction module and a knowledge reasoning module. The knowledge extraction module is used to extract triples from the entity relationships of the tight sandstone reservoir sequence annotation and construct a tight sandstone reservoir knowledge graph with joint entity relationships. The knowledge reasoning module is based on a rule-embedded structure-aware convolutional network and is used for multi-task collaborative reasoning.

[0010] Step S300: Based on the tight sandstone reservoir evaluation model, perform multi-task collaborative reasoning for tight sandstone reservoirs.

[0011] Furthermore, the tight sandstone reservoir evaluation data mentioned in step S100 includes: text, numerical, and image data.

[0012] Furthermore, the construction of the bulk layer in the tight sandstone reservoir field includes:

[0013] Step S110: Define the scope of the tight reservoir field: The core information sources in the reservoir field include: the tight reservoir basic information database, the terminology set and academic literature related to the tight reservoir evaluation industry standards, and the systematic organization and summary of the practical experience of reservoir evaluation experts and their professional teams.

[0014] Step S120, Entity Concept and Attribute Definition: There are two strategies for concept extraction. One is to extract concepts from macro to micro and subdivide them step by step according to the pre-set business process architecture. The other strategy is to use collection and retrieval methods to deeply mine information from resources such as documents and basic information databases in order to extract the required concepts to be defined.

[0015] Step S130, Relationship Definition: The characteristics of the classification relationship between attributes are described using the attribute ontology architecture and clustering attribute method. Based on the extracted concepts and the results obtained from the requirements analysis, the association between each category is clarified.

[0016] Furthermore, the knowledge extraction module in step S200 includes: an embedding module, an encoding module, and a decoding module;

[0017] The embedding module is built with a BERT module, which is used to transform sentences into dynamic semantic information at the character and word levels.

[0018] The encoding module is constructed with a BiLSTM module and a CNN module. The BiLSTM module is used to extract contextual information of dense reservoir sequence data, and the CNN module is used to extract detailed features of the data using multiple convolutional kernels of different sizes, and to select the maximum value of the feature map through a pooling layer to further aggregate the features.

[0019] The decoding layer is equipped with a CRF module, which is used to obtain the globally optimal label sequence of an entity;

[0020] Ultimately, the system extracts triples from entity relationships and constructs a knowledge graph of tight sandstone reservoirs that combines entity relationships.

[0021] Furthermore, the CNN module uses 1×3 convolutions to capture entity features in individual geological terms and reservoir texts, and uses 5×5 convolutions to capture combinations of geological terms and their contextual relationships.

[0022] Furthermore, the knowledge reasoning module in step S200 includes a rule-enhanced graph convolutional layer and a rule-constrained prediction layer. The rule-enhanced graph convolutional layer is used in the encoding stage to dynamically model the matching degree between nodes and geological rules through a differentiable rule encoder, and to realize the directional filtering and aggregation of neighborhood node information by using rule weight gating in the multi-head attention mechanism. In the decoding stage, the rule-constrained prediction layer introduces a two-stage correction mechanism. First, it filters illegal predictions through hard constraints, and then constructs residual correction terms by combining soft rules. The physical equations are embedded into the fully connected layer in a learnable form, and finally realizes the full-link closed-loop constraint of geological knowledge from the feature space to the decision space.

[0023] Furthermore, the implementation process of the rule-enhanced graph convolutional layer is as follows:

[0024] In the reservoir knowledge graph G=(V,E,R), node feature updates employ a reservoir rule-aware graph convolution operation, calculated using the following formula:

[0025]

[0026] in, For the first Layer nodes Feature representation, For the activation function, R rule This represents a set of relationships related to the rules. Indicates a relationship Next, node The set of neighboring nodes, Indicates a relationship Next node right Regularized attention weights, Indicates the first Layers, Relationships The transformation matrix under, Indicates the first Layer nodes Feature representation, Indicates the first Transformation matrices related to rules in the layer, This represents the feature representation related to rule triggering. The calculation formula is as follows:

[0027]

[0028] in, Represents an exponential function. , For the rule matching degree function, This represents the weight vector in the attention mechanism. This represents the representation of node i. Representation of node j To splice, This represents the representation of node i. Representation of node k To splice, Let i represent the set of neighboring nodes. This represents an activation function used to introduce a nonlinear transformation;

[0029] By defining the rule matching degree function The importance of neighboring nodes is adjusted using the following formula:

[0030]

[0031] in, This represents the atomic formula input to the rule encoder. A rule encoder used to map atomic formulas to computable conditions in a vector space:

[0032]

[0033] in, and Represents the weight matrix. and This indicates the bias term.

[0034] Furthermore, the loss function of the rule-constrained prediction layer is a dynamically weighted loss function based on classification cross-entropy, regression mean squared error, and rule constraints:

[0035]

[0036]

[0037] in, This is a soft rule error term. This is the model's predicted output for the sample. For reference output related to soft rules, The loss function for the rule-constrained prediction layer. The basic prediction loss of the model, Represents a set of hard rules. Represents a set of soft rules. Penalties for violating hard rules This is a soft rule error term. and It is a weight parameter used to control the degree of influence of rule constraints.

[0038] This invention provides a knowledge graph-based tight sandstone reservoir evaluation system. The system has program modules corresponding to the steps of any of the above-described technical solutions, and executes the steps in the knowledge graph-based tight sandstone reservoir evaluation method during runtime.

[0039] The present invention also provides a computer-readable storage medium storing a computer program configured to, when invoked by a processor, implement the steps in the knowledge graph-based tight sandstone reservoir evaluation method described in any of the above technical solutions.

[0040] Compared with the prior art, the beneficial effects of the present invention are:

[0041] This invention addresses the challenges of dense technical terms, complex and implicit entity relationships in tight reservoir text data. It constructs a joint entity relationship extraction module based on sequence labeling patterns, overcoming the error accumulation problem caused by the separation of entity recognition and relationship classification in traditional pipeline extraction methods. Through a unified coding framework, it synchronously achieves end-to-end extraction of key reservoir entities and their associated relationships. Furthermore, this invention distills expert experience and formulates standard expressions for inference model recognition, constructing a rule-embedded structure-aware network inference module to realize multi-task inference in the field of tight sandstone reservoir evaluation. Combined with business context, it designs a knowledge graph-based tight reservoir evaluation system capable of quickly and accurately classifying tight reservoirs and predicting reservoir parameters. This not only improves work efficiency but also reduces reliance on human experience, subjective errors, and redundancy of multi-source data, demonstrating high application value. Attached Figure Description

[0042] Figure 1 This is a schematic diagram of the knowledge graph-based tight sandstone reservoir evaluation system in an embodiment of the present invention.

[0043] Figure 2 This is a schematic diagram of entity relationship extraction based on the knowledge extraction module in an embodiment of the present invention;

[0044] Figure 3 This is a network structure diagram of the knowledge reasoning module in an embodiment of the present invention;

[0045] Figure 4This is a diagram of the rule encoder network structure in an embodiment of the present invention;

[0046] Figure 5 This is a flowchart of multi-task collaborative processing in an embodiment of the present invention;

[0047] Figure 6 This is a structural diagram of a tight reservoir in an embodiment of the present invention. Detailed Implementation

[0048] To enable those skilled in the art to better understand the present invention, exemplary embodiments or examples of the present invention will be described below in conjunction with the accompanying drawings. Obviously, the described embodiments or examples are merely some, not all, of the embodiments or examples of the present invention. All other embodiments or examples obtained by those skilled in the art based on the embodiments or examples of the present invention without inventive effort should fall within the scope of protection of the present invention.

[0049] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0050] Specific Implementation Scheme 1: This invention provides a method for evaluating tight sandstone reservoirs based on knowledge graphs, comprising the following steps:

[0051] Step S100: Collect tight sandstone reservoir evaluation data, preprocess the tight sandstone reservoir evaluation data, construct an ontology knowledge base, and construct an ontology layer for the tight sandstone reservoir domain based on knowledge and the relationships between them.

[0052] Step S200: Construct a tight sandstone reservoir evaluation model. The evaluation model includes a knowledge extraction module and a knowledge reasoning module. The knowledge extraction module is used to extract triples from the entity relationships of the tight sandstone reservoir sequence annotation and construct a tight sandstone reservoir knowledge graph with joint entity relationships. The knowledge reasoning module is based on a rule-embedded structure-aware convolutional network and is used for multi-task collaborative reasoning.

[0053] Step S300: Based on the tight sandstone reservoir evaluation model, perform multi-task collaborative reasoning for tight sandstone reservoirs.

[0054] Specific Implementation Plan Two: The evaluation data for the tight sandstone reservoir includes text, numerical, and image data. Specifically, it includes:

[0055] Regional geological data includes the region’s stratigraphic sequence and structural features, rock types and distribution, mineral resources, scale, and related geological maps and historical exploration data.

[0056] Core experimental data: including physical, chemical and structural characteristics of core samples obtained through laboratory analysis.

[0057] Curve data: Formation information obtained using well logging technology is represented in the form of curves.

[0058] Pore ​​information: Pore images are obtained using the casting thin section method, scanning electron microscopy and CT scanning. Pore characteristics are obtained from the perspective of image analysis, including but not limited to pore structure, type, morphology, size, distribution, mineral composition and distribution, particle structure and quantitative parameters extracted through image analysis.

[0059] Well logging data: includes well logging documents and raw well logging data.

[0060] Oil and gas reservoir evaluation data: includes reservoir geological and engineering evaluation documents and basic data. This implementation plan is otherwise the same as Specific Implementation Plan One.

[0061] Specific implementation plan three: Image data preprocessing includes:

[0062] Standardize the image data: unify the image size to 512×512 pixels, remove noise, and enhance the visibility of cracks or pores;

[0063] Image feature extraction: A pre-trained ResNet50 feature extraction network is used, with an SE module added after the ResNet residual block to focus on key feature channels, and FPN is applied. By fusing features from different levels, image features can be extracted.

[0064] Cross-modal mapping: Reservoir information is extracted from image-attached labels, and image feature vectors and text descriptions are embedded in the same space to establish a relationship between images and text. This implementation scheme is otherwise the same as specific implementation scheme two.

[0065] Specific implementation plan four: Preprocessing of text data includes:

[0066] Standardize the text data: Convert the original reservoir data PDF file to TXT format, and remove line breaks, spaces, punctuation marks, and numerical descriptive text that does not contain entity relationships.

[0067] Text filtering: Filter out irrelevant text and filter the text.

[0068] Text annotation: For entity boundary labels, each word in the text is marked with multiple labels, including the word's position in the entity, relation type, and entity location information. This implementation scheme is otherwise the same as specific implementation scheme three.

[0069] Specific Implementation Plan Five: Construction of the main body layer in the tight sandstone reservoir field, including:

[0070] Step S110: Define the scope of tight reservoirs: The core information sources in the reservoir field include: the primary source is the tight reservoir basic information database, followed by terminology sets and academic literature related to tight reservoir evaluation industry standards, including classic works in the field of oil and gas evaluation such as "Comprehensive Evaluation Technology of Oil and Gas Reservoirs" and "Tight Oil and Gas"; and finally, a systematic compilation and summary of the practical experience of reservoir evaluation experts and their professional teams.

[0071] A top-down domain ontology knowledge representation method is adopted to construct the domain ontology layer of tight sandstone reservoirs, and OWL DL language is selected as the modeling tool. This aims to ensure powerful descriptive expressiveness while also fully considering and achieving effective compatibility with the reasoning mechanism. The "entity-attribute-attribute value" relationship is mined from the tight sandstone reservoir information to construct the domain ontology layer of tight sandstone reservoir knowledge.

[0072] Step S120: Entity Concept and Attribute Definition: The ontology layer of the tight reservoir domain knowledge graph consists of a set of concept nodes and a set of concept relationship edges, representing concepts and relationships between concepts in the geological domain. Prior knowledge is used to classify the conceptual hierarchy relationships, attribute relationships, and semantic relationships between concepts for each element.

[0073] The concept extraction method includes two strategies. The first is based on a pre-defined business process architecture, employing a macro-to-micro, step-by-step subdivision approach. During this process, it is crucial to closely adhere to the specific operational procedures within the reservoir evaluation field. For example, when classifying reservoir performance, reservoirs are categorized into Class I, Class II, and Class III based on permeability gradients and porosity. To classify reservoirs based on this, the first step is to construct a conceptual ontology of tight sandstone reservoirs, including: reservoir, pore structure, fractures, minerals, porosity, and permeability. Furthermore, the four properties—physical properties, oil content, electrical properties, and lithology—are obtained through well logging experiments. Therefore, an ontology of these four properties must be constructed. The lithological ontology includes sandstone type and cementation type; the electrical ontology includes resistivity and spontaneous potential.

[0074] Another strategy involves in-depth mining of information from documents and basic information databases through collection and retrieval methods to extract the necessary concepts to be defined. For example, searching for the keyword "lithology" will yield rock types such as "dolomite," "gypsum," and "granite," which can be directly incorporated into the ontology as clearly defined concepts. After establishing these basic concepts, they need to be further refined in the context of specific situations, analyzing the connotation of the concepts themselves and their related attributes, and optimizing the hierarchical structure between categories accordingly. The constructed model concepts include mathematical expressions such as Archie's formula and Darcy's law, as well as standard interpretation charts and regionally adapted interpretation charts. The concepts and attributes extracted through these two approaches can maximize the coverage of reservoir evaluation knowledge and provide strong foundational support for the subsequent practical application of the ontology and attributes.

[0075] This implementation plan analyzes the knowledge base of the entire tight sandstone reservoir field and related reservoir data. Through the two detailed conceptual analysis methods mentioned above, 14 entity types are obtained in the tight reservoir field knowledge, including region, oil field, well, logging experiment, four properties, pore structure, mineralization, formula, chart, stratigraphy and geological data, as shown in Table 1.

[0076] Table 1 Reservoir Entity Type Information

[0077] Entity type Chinese meaning For example Area area Sanzhao Depression Area OilField oilfield Daqing Oilfield Well well Tai 27 well area Logging Well logging projects Hole logging; Neutron logging Physical physical properties Porosity; Permeability Oiliness Oily Oil and gas saturation Electric electrical Resistivity; Spontaneous Potential Lithology Lithology mudstone pore structure Pore ​​structure Intergranular pores; dissolution pores; microcracks Mineral mineralization Quartz; Feldspar; Kaolinite Formula formula Archie formula Plate plate Well logging parameter cross plot; Stratum strata conglomerate layer, sandstone layer Sdata Geological data Reservoir literature; geological reports

[0078] Step S130, Relationship Definition: The characteristics of the classification relationships between attributes are described using attribute ontology architecture and clustering attribute methods. Based on the extracted concepts and the results of the requirements analysis, the relationships between each category are clarified. For example, oil saturation, permeability, and clay content are all calculated and analyzed using model formulas; therefore, there is a calculation relationship between the model formulas and oil properties, electrical properties, and physical properties. Simultaneously, each oilfield is located within a specific geographical region; for example, the relationship between the entity object "Fu-1 Oil Formation Structural High Point" and "Tai-27 Well Area" is "located in". It is inferred that there is an inclusion relationship between oilfields and oil wells, while there is a geographical affiliation relationship between oilfields and their respective regions.

[0079] Ultimately, the knowledge obtained in the field of well logging includes: location, calculation, classification, related formulas and usage relationships, as shown in Table 2.

[0080] Table 2

[0081] Relationship type Chinese meaning For example Include Include <Greyish-black middle layer - massive limestone, containing flint bands> Located lie in <Fuyi Oil Reservoir Structural High Point, located in Well Tai 27 Area> Calculate calculate <Archie formula, calculation, oil saturation> Divide Division <Hole diameter logging, division, lithology> Relate_formu Related formulas <SP log, related formulas, diffusion electromotive force formula> Use belong <Survey area, belonging to, alpine region of the Qinghai-Tibet Plateau>

[0082] Obtaining bulk layer information in the tight reservoir domain, such as Figure 6As shown. This implementation plan is otherwise identical to specific implementation plan four.

[0083] Specific Implementation Plan Six: (e.g.) Figure 2 As shown, the knowledge extraction module includes: an embedding module, an encoding module, and a decoding module;

[0084] The embedding module is equipped with a BERT module, which is used to transform sentences into dynamic semantic information at the character and word levels; then the semantic information is input to the encoding layer.

[0085] The encoding module is constructed with a BiLSTM module and a CNN module. The BiLSTM module is used to extract contextual information of dense reservoir sequence data, and the CNN module is used to extract detailed features of the data using multiple convolutional kernels of different sizes, and to select the maximum value of the feature map through pooling layers to further aggregate features.

[0086] The encoding module, for the input X=[Sanzhao Depression Region, Permeability, Porosity, ..., Larger], first inputs the vector representation of each word into the LSTM in the natural order of the sentence to obtain a set of sequential hidden layer representations.

[0087] The forget gate is used to determine whether to ignore certain historical information that is irrelevant to the current reservoir characteristics. The calculation is as follows:

[0088] (1)

[0089] in For the Gate of Oblivion This represents the recursive weight shared by the previous state. This represents the hidden state of the previous time step. This represents the feature vector input at time t. Indicates the bias term. It is the sigmoid activation function.

[0090] The input gate selectively introduces new geological features or rock type information based on the current reservoir property description. Its output is a vector between 0 and 1. The calculation formula for the input gate is:

[0091] (2)

[0092] in The output of the input gate, It is the sigmoid activation function. This represents the recursive weight shared by the previous state. This represents the hidden state of the previous time step. This represents the feature vector input at time t. This indicates the bias term.

[0093] It is a new reservoir feature representation generated based on current information and historical experience. The formula for updating cell state is:

[0094] (3)

[0095] in For the Gate of Oblivion This indicates the cell state at time step t-1. The output of the input gate, This represents the hyperbolic tangent activation function. This represents the recursive weight shared by the previous state. This represents the hidden state of the previous time step. This represents the feature vector input at time t. This indicates the bias term.

[0096] Among them, the output gate selectively outputs the predicted value of reservoir quality according to the current processing result, and the calculation formula is shown in (4). The calculation formula of the hidden state is shown in (5).

[0097] (4)

[0098] (5)

[0099] in This represents the hidden state at time step t. This indicates the output of the output gate at time step t. This represents the hyperbolic tangent activation function. It is a new reservoir feature representation generated based on current information and historical experience. This represents the recursive weight shared by the previous state. This represents the hidden state of the previous time step. This represents the feature vector input at time t. This indicates the bias term.

[0100] The words in the sentence are reversed in order and then fed into an LSTM to obtain a reversed hidden layer representation. Finally, the forward and reversed hidden layer representations are concatenated to obtain the final hidden layer representation for each sentence. .

[0101] The CNN module uses 1×3 convolutions to capture entity features in individual geological terms and reservoir texts, uses 5×5 convolutions to capture combinations of geological terms and their contextual relationships, and uses the ReLU function to generate higher-level feature representations that reflect the complex relationship between geological terms and attributes.

[0102] For each convolution kernel, a sliding window is used to access the input features. The feature map is obtained by sliding the text word by word or character by character. To obtain the most valuable features from the matrix obtained after convolution, max pooling is performed on the feature map, and the maximum value is selected. As a feature corresponding to a specific filter.

[0103] Finally, a fully connected layer is used for regularization adjustment, and a dropout strategy is employed to randomly discard hidden units, resulting in the final feature representation of the input sentence. Then, proceed to the decoding layer for further processing.

[0104] The decoding layer incorporates a CRF module to obtain the globally optimal label sequence. This is for the input compact reservoir feature representation. Let the corresponding actual label sequence be... The formula for calculating the score function of the label sequence is:

[0105]

[0106] in, To represent the characteristics of tight reservoirs, To represent the characteristics of tight reservoirs, The initial score matrix is ​​obtained by linear operations on the output of the BiLSTM layer. Represents the state transition matrix. For consecutive words, tags Move to label The score, Indicates the first word in the sentence The first word The score for each label. For the input sequence Corresponding output label sequence Calculate the score, and the sequence with the highest score is the final predicted label sequence.

[0107] Triples are extracted from entity relationships to construct a knowledge graph of tight sandstone reservoirs with joint entity relationships.

[0108] The obtained relationship classification results are converted into triples according to certain matching rules. First, based on the entity boundaries, entity relationships, and labeled entity role tags, relatively complete entities and relationships are obtained, including: reservoir type, rock type, porosity, and permeability. Simultaneously, the corresponding relationship type, entity role information, and location information are obtained. Only entities whose boundary tags begin with "B," whose "I" is not a header character, and whose relationship and role tags for each character within the entity are consistent can be extracted as entity relationship triples. Second, for the extracted entities with ordinary relationships, forward and backward searches are used to directly find matching entities in the text. The following rules are followed in finding entity matches: the relationship tags of the matching entities must be consistent with the current entity, while the entity role tags should be different from the current entity. Finally, after obtaining the matching entities from the forward and backward searches, the distance between the current entity and the entity to be matched needs to be determined, and the closest pair of entities is selected to form a matching pair, thus forming an entity relationship triple.

[0109] First, entities related to tight sandstone reservoirs are treated as nodes in the graph, and each node is assigned a unique ID. Next, each triple is transformed into a directed edge, pointing from the beginning entity to the end entity, and the triples are stored using the Neo4j graph database. Finally, the graph is converted into an adjacency matrix. It is used for reservoir relationship modeling and employs the TransE embedding method to map entity relationships to a vector space.

[0110] In the intelligent evaluation system for tight reservoirs, the rule base built based on expert experience is the foundation for evaluation reasoning in tight sandstone reservoirs. Targeting 263 domain experiences found in the unstructured text of geological data and core analysis reports of the Fuyu oil layer, multiple rounds of verification by geological experts resulted in an evaluation system containing 217 valid rules. To improve the machine interpretability and computability of knowledge, a production rule representation method was used to formally encode implicit experiences. At the semantic level, expert experience was deconstructed into triples (preconditions, conclusions, confidence parameters), as shown in Table 3.

[0111] Table 3

[0112]

[0113] In the knowledge graph transformation stage, the visualization and reasoning computation of rules are achieved through the mapping of predicate logic to graph structures:

[0114] For simple rules This means that all entities x that satisfy condition P(x) must be associated with Q(x). In this implementation scheme, the rule "reservoir porosity > 8% → sweet spot" is expressed as: [reservoir A porosity > 8%] → [sweet spot] (w=0.88).

[0115] Handling conjunction (∧) and disjunction (∨) in compound logic: For conjunction rules A dual-path joint reasoning mechanism is adopted: P(x)-[Rule1]→R(x) and Q(x)-[Rule1]→R(x), with weights jointly controlled by logical relations. This implementation scheme is based on the rule "(median pore throat radius of mercury intrusion) > 1". ) ∧ (CT pore coordination number > 3) → pore structure type = Class I” is represented as: [mercury porosimetry radius > 1 [AND_Gate] → [Porosity Type I] (w=0.95), [CT Coordination Number > 3] → (AND_Gate) → [Porosity Type I] (w=0.95); for the separation rule An independent inference path is constructed, with weights set separately. Activation occurs when any condition is met. In this implementation scheme, the rule "(crack density > 2 cracks / m) ∨ (crack length > 10m) → crack development level = medium-high" is represented as: [crack density > 2 cracks / m] → [crack development level = medium-high] (w=0.78), [crack length > 10m] → [crack development level = medium-high] (w=0.78). All other aspects of this implementation scheme are the same as in specific implementation scheme five.

[0116] Specific Implementation Plan Seven: (e.g.) Figure 3 As shown, the knowledge reasoning module in step S200 includes a rule-enhanced graph convolutional layer and a rule-constrained prediction layer. The rule-enhanced graph convolutional layer is used in the encoding stage. It dynamically models the matching degree between nodes and geological rules through a differentiable rule encoder and uses rule weight gating in the multi-head attention mechanism to achieve directional filtering and aggregation of neighborhood node information. In the decoding stage, the rule-constrained prediction layer introduces a two-stage correction mechanism. First, it filters illegal predictions through hard constraints, and then combines soft rules to construct residual correction terms. The physical equations are embedded into the fully connected layer in a learnable form, and finally realizes the full-link closed-loop constraint of geological knowledge from the feature space to the decision space.

[0117] The functional implementation process of the knowledge reasoning module is as follows:

[0118] Step 1: Construct rule-enhanced graph convolutional layers, design multi-layer graph convolutional structures, and define the kernel size, dilation factor, and number of layers to capture multi-scale topological relationships in the reservoir knowledge graph; embed rule attention modules into the convolutional layers and dynamically adjust feature weights based on rule matching degree.

[0119] Step 2: Train the tight sandstone reservoir knowledge reasoning module. Load entity relationships from the tight sandstone reservoir knowledge graph as rule constraints. Input the processed reservoir data into the knowledge reasoning module for training loops. Graph convolutional layers extract reservoir topological features, and the rule attention module optimizes the feature graph, enhancing high-contribution features relevant to the current rule. The optimized features are then concatenated with the original features to form a joint semantic representation.

[0120] Step 3: Optimize the model by dynamically adjusting the weight coefficients of the rule penalty terms based on the rule violation rate of the test set prediction results; optimize the network weights using the gradient descent algorithm according to the above criteria, prioritizing the retention of parameter combinations that conform to geological rules.

[0121] Step 4: Finally, output reservoir classification labels, simultaneously predict continuous parameters such as porosity and permeability, and correct the original rules to complete the reasoning for tight sandstone reservoirs. This implementation plan is otherwise the same as specific implementation plan six.

[0122] Specific implementation scheme eight: The functional implementation process of the rule-enhanced graph convolutional layer is as follows:

[0123] In the reservoir knowledge graph G=(V,E,R), node feature updates employ a reservoir rule-aware graph convolution operation, calculated as follows:

[0124]

[0125] in, For the first Layer nodes Feature representation, For the activation function, R rule This represents a set of rules-related relations, containing various relations used for reservoir rule-aware graph convolution. , Indicates a relationship Next, node The set of neighboring nodes, Indicates a relationship Next node right The regularized attention weights are determined by the reservoir rule matching degree. Determined in conjunction with the topology, Indicates the first Layers, Relationships The transformation matrix under, Indicates the first Layer nodes Feature representation, Indicates the first Transformation matrices related to rules in the layer, This represents the feature representation related to rule triggering. The calculation formula is as follows:

[0126]

[0127] in Represents an exponential function. , For the rule matching degree function, This represents the weight vector in the attention mechanism. This represents the representation of node i. Representation of node j To splice, This represents the representation of node i. Representation of node k To splice, Let i represent the set of neighboring nodes. This represents an activation function used to introduce a nonlinear transformation.

[0128] Rule matching degree measures how well a node conforms to a specific rule. This includes situations where, under the "crack validity rule," some cracks contribute significantly to seepage, while others may be ineffective. This is achieved by defining a rule matching degree function. The importance of neighboring nodes is adjusted so that nodes that conform to the rules receive higher weight during information propagation, while the influence of nodes that do not conform to the rules is weakened. This ensures that the learned representation conforms to the knowledge of the reservoir domain. The calculation formula is as follows:

[0129]

[0130] in, This represents the atomic formula input to the rule encoder. This is a rule encoder whose goal is to map atomic formulas into computable conditions in a vector space, as follows:

[0131]

[0132] in, and Represents the weight matrix. and This indicates the bias term.

[0133] Rule Encoder The network structure is a shared MLP, such as Figure 4 As shown, for the node pairs participating in the rule determination The hidden state and relation embedding are concatenated. Assuming the feature dimension of each node is d and the relation embedding dimension is k, the total input dimension is 2d + k. The output is a scalar value, and the rule matching degree is obtained by passing a sigmoid function. This allows its output range [0,1] to reflect the degree of compliance with the rules.

[0134] In the reasoning process for tight reservoir evaluation, a collaborative reasoning framework is constructed by integrating well logging data, image features, geological text descriptions, and domain knowledge rules required for intelligent tight reservoir evaluation. Based on the reservoir evaluation knowledge system, semantic association rules between entity nodes of "reservoir feature - reservoir attribute" are defined, a weighted adjacency matrix is ​​constructed, and a weighted graph convolutional network (WGCN) is used to process the graph structure data of reservoir samples, generating entity embedding vectors. During GCN iteration, the semantic propagation weights between entity nodes of "reservoir feature - reservoir attribute" type are dynamically adjusted to solve the oversmoothing problem caused by the heterogeneity of geological relationships in traditional GCN. Geologically meaningful entity relationship embedding vectors are generated, enabling the deep network to be better trained and optimized in tight reservoir classification tasks. Other aspects of this implementation scheme are the same as in specific implementation scheme seven.

[0135] Specific implementation scheme nine: The functional implementation process of the prediction layer of the rule constraint is as follows:

[0136] During the decoding process, for the preliminary tight sandstone reservoir evaluation results and parameter predictions generated from the Conv-TransE layer, the results are first filtered using constraints constructed based on logical rules and expert knowledge to remove predictions that violate hard constraints. Then, results that conform to hard constraints but have biases are weighted and adjusted to better align with reservoir rules. For known empirical relationships, weighting factors are used to adjust the prediction results to better reflect domain knowledge. Finally, during the training phase, rule constraints are explicitly introduced into the loss function, enabling the model to learn prediction patterns that conform to domain knowledge. Prediction results that violate hard rule constraints are penalized using a large penalty term to significantly increase the loss and prevent the generation of illegal predictions. The model's adaptability to soft rules is improved by minimizing the error between the calculated prediction value and the expected value of the rule; the calculation formula is as follows:

[0137]

[0138]

[0139] in, This is a soft rule error term. This is the model's predicted output for the sample. For reference output related to soft rules, For the overall loss function, The basic prediction loss of the model, Represents a set of hard rules. Represents a set of soft rules. Penalties for violating hard rules This is a soft rule error term. and This is a weighting parameter used to control the degree of influence of rule constraints. This implementation scheme is otherwise the same as specific implementation scheme eight.

[0140] Specific Implementation Plan Ten: (e.g.) Figure 5 As shown, the multi-task collaborative reasoning in step S300 includes the following steps:

[0141] Reservoir classification task:

[0142] Based on traditional reservoir classification, a static parameter classification system is constructed using geological expert experience and parameter association criteria from a knowledge graph: Class I reservoirs: porosity > 8%, permeability > 0.1 mD; Class II reservoirs: porosity 5%–8%, permeability 0.05–0.1 mD; Class III reservoirs: porosity < 5%, permeability < 0.05 mD. Reservoir quality grades are output based on the input reservoir characteristics. A rule-constrained cross-entropy loss is used to ensure consistency between the classification results and expert experience. The calculation formula is as follows:

[0143]

[0144] in, This is the embedding vector for reservoir evaluation rules in the knowledge graph;

[0145] The loss function combines data-driven loss and knowledge-constrained loss, and the calculation formula is as follows:

[0146]

[0147] Wherein, when the rules are met, the indicator function It is 1; otherwise, it is 0.

[0148] Parameter prediction task:

[0149] Simultaneous prediction includes continuous physical parameters such as porosity and permeability, reducing the risk of overfitting through multi-task feature layer sharing. Parameter-specific projection is calculated using the following formula:

[0150]

[0151] in, The formula is dynamically generated using relational paths:

[0152]

[0153] Indicates parameters All related paths in the knowledge graph;

[0154] Physical equation constraints: The value of K is adjusted based on the Kozeny-Carman equation, and the calculation formula is as follows:

[0155]

[0156] In the reservoir classification task, a fully connected layer maps joint features with geological semantics to the reservoir category probability space, and a rule constraint layer performs hard validation, ultimately outputting the reservoir type label. In the parameter prediction task, after regressing the initial physical property values ​​through a residual network, the Kozeny-Carman equation physical constraints are introduced for adaptive correction, simultaneously verifying whether the predicted values ​​conform to the empirical relationship between porosity and permeability. Then, a dynamic weighted loss function is designed, including classification cross-entropy, regression mean square error, and rule penalty terms, to achieve multi-objective collaborative optimization.

[0157] In the multi-task reasoning framework, reservoir classification and parameter prediction form a bidirectional reinforcement effect through a knowledge-driven collaborative mechanism: the reservoir classification task provides geological semantic context for parameter prediction through the hierarchical division of rule constraints, constraining the physical reasonable range of physical properties; while the parameter prediction task, through high-precision regression results, including continuous values ​​of porosity and permeability, feeds back into the classification decision, uses differential equations to reveal the theoretical correlation between parameters, and corrects classification ambiguities caused by data noise. Gradient collaborative optimization is achieved through dynamic weight allocation, effectively solving the semantic conflict problem existing in traditional multi-task learning.

[0158] In the evaluation of tight sandstone reservoirs, a knowledge-driven bidirectional enhancement mechanism forms a closed-loop optimization system through positive constraints of "reservoir classification → parameter prediction" and negative corrections of "parameter prediction → reservoir classification." The tight reservoir evaluation criteria derived from geological laws and expert experience are transformed into logical rules or physical equations to constrain the output range of the parameter prediction model. Data-driven models are used to uncover implicit correlations, and the strength of knowledge constraints is dynamically adjusted based on data quality and knowledge confidence to correct local biases in expert knowledge. Other aspects of this implementation plan are the same as in specific implementation plan nine.

[0159] The knowledge graph-based evaluation method (algorithm) for tight sandstone reservoirs proposed in this invention is the underlying technical core of this invention, and various products can be derived based on the algorithm.

[0160] Based on the method proposed in this invention, a knowledge graph-based tight sandstone reservoir evaluation system is developed using a programming language. This system has program modules corresponding to the steps of the above-mentioned technical solution, and executes the steps in the knowledge graph-based tight sandstone reservoir evaluation method described above when running.

[0161] The overall framework of the knowledge graph-based evaluation model for tight sandstone reservoirs is as follows: Figure 1 As shown, it includes:

[0162] Data layer: Used for various types of data acquisition and storage, providing the necessary structured and unstructured data for constructing a knowledge graph in the field of tight sandstone reservoirs, and performing data preprocessing operations on the collected multi-source heterogeneous data.

[0163] Service Layer: This layer is the middle layer of the model and serves as the core processing module of the system. It is responsible for extracting entity relationships in the tight sandstone reservoir domain and for using a rule-enhanced structure-aware network (SCAN) reservoir quality inference method. Specifically, in the tight reservoir entity relationship joint extraction model, entity relationship extraction is transformed into a sequence labeling task, focusing on identifying and extracting nested entities and complex overlapping relationships in tight reservoir text, ultimately obtaining the globally optimal label sequence. The extracted triples are stored in the Neo4j graph database system, and a rule base is constructed based on domain expert experience. Combined with the rule-enhanced SCAN reservoir quality inference method, the evaluation results of the tight sandstone reservoir are obtained.

[0164] Application Layer: The application layer is the top layer of the model, including the knowledge graph management module, the reservoir evaluation module, and the system function configuration.

[0165] The developed system (software) computer program is stored on a computer-readable storage medium, and the computer program is configured to implement the steps of the knowledge graph-based tight sandstone reservoir evaluation method described above when called by a processor. In other words, the invention is materialized on a carrier, becoming a computer program product.

[0166] Various implementations of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, application-specific integrated circuits (ASICs), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include: implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0167] The computational programs (also referred to as programs, software, software applications, or code) of this invention include machine instructions of a programmable processor and can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, device, and / or apparatus (e.g., disk, optical disk, memory, programmable logic device PLD) for providing machine instructions and / or data to a programmable processor, including machine-readable media that receive machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal for providing machine instructions and / or data to a programmable processor.

[0168] While the present invention has been disclosed above, its scope of protection is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention, and all such changes and modifications will fall within the scope of protection of the present invention.

Claims

1. A knowledge graph-based evaluation method for tight sandstone reservoirs, characterized in that: Includes the following steps: Step S100: Collect tight sandstone reservoir evaluation data, preprocess the tight sandstone reservoir evaluation data, construct an ontology knowledge base, and construct an ontology layer for the tight sandstone reservoir domain based on knowledge and the relationships between them. Step S200: Construct a tight sandstone reservoir evaluation model. The evaluation model includes a knowledge extraction module and a knowledge reasoning module. The knowledge extraction module is used to extract triples from the entity relationships of the tight sandstone reservoir sequence annotation and construct a tight sandstone reservoir knowledge graph with joint entity relationships. The knowledge reasoning module is based on a rule-embedded structure-aware convolutional network and is used for multi-task collaborative reasoning. Step S300: Based on the tight sandstone reservoir evaluation model, perform multi-task collaborative reasoning for tight sandstone reservoirs; The knowledge reasoning module in step S200 includes a rule-enhanced graph convolutional layer and a rule-constrained prediction layer. The rule-enhanced graph convolutional layer is used in the encoding stage to dynamically model the matching degree between nodes and geological rules through a differentiable rule encoder, and to realize the directional filtering and aggregation of neighborhood node information by using rule weight gating in the multi-head attention mechanism. In the decoding stage, the prediction layer constrained by the rules introduces a two-stage correction mechanism. First, illegal predictions are filtered out through hard constraints, and then residual correction terms are constructed in combination with soft rules. The physical equations are embedded into the fully connected layer in a learnable form, and finally the geological knowledge is closed-loop constrained from the feature space to the decision space. The implementation process of the rule-enhanced graph convolutional layer is as follows: In the reservoir knowledge graph G=(V,E,R), node feature updates employ a reservoir rule-aware graph convolution operation, calculated using the following formula: ； in, For the first Layer nodes Feature representation, For the activation function, R rule This represents a set of relationships related to the rules. Indicates a relationship Next, node The set of neighboring nodes, Indicates a relationship Next node right Regularized attention weights, Indicates the first Layers, Relationships The transformation matrix under, Indicates the first Layer nodes Feature representation, Indicates the first Transformation matrices related to rules in the layer, This represents the feature representation related to rule triggering. The calculation formula is as follows: ； in, Represents an exponential function. For the rule matching degree function, This represents the weight vector in the attention mechanism. This represents the representation of node i. Representation of node j To splice, This represents the representation of node i. Representation of node k To splice, Let i represent the set of neighboring nodes. This represents an activation function used to introduce a nonlinear transformation; By defining the rule matching degree function The importance of neighboring nodes is adjusted using the following formula: ； in, This represents the atomic formula input to the rule encoder. A rule encoder used to map atomic formulas to computable conditions in a vector space: ； in, and Represents the weight matrix. and This indicates the bias term.

2. The knowledge graph-based evaluation method for tight sandstone reservoirs according to claim 1, characterized in that: The tight sandstone reservoir evaluation data mentioned in step S100 includes: text, numerical, and image data.

3. The knowledge graph-based evaluation method for tight sandstone reservoirs according to claim 2, characterized in that: The construction of the bulk layer in tight sandstone reservoirs includes: Step S110: Define the scope of the tight reservoir field: The core information sources in the reservoir field include: the tight reservoir basic information database, the terminology set and academic literature related to the tight reservoir evaluation industry standards, and the systematic organization and summary of the practical experience of reservoir evaluation experts and their professional teams. Step S120, Entity Concept and Attribute Definition: There are two strategies for concept extraction. One is to extract concepts from macro to micro and subdivide them step by step according to the pre-set business process architecture. The other strategy is to use collection and retrieval methods to deeply mine information from resources such as documents and basic information databases in order to extract the required concepts to be defined. Step S130, Relationship Definition: The characteristics of the classification relationship between attributes are described using the attribute ontology architecture and clustering attribute method. Based on the extracted concepts and the results obtained from the requirements analysis, the association between each category is clarified.

4. The knowledge graph-based evaluation method for tight sandstone reservoirs according to claim 3, characterized in that: The knowledge extraction module in step S200 includes: an embedding module, an encoding module, and a decoding module; The embedding module is built with a BERT module, which is used to transform sentences into dynamic semantic information at the character and word levels. The encoding module is constructed with a BiLSTM module and a CNN module. The BiLSTM module is used to extract contextual information of dense reservoir sequence data, and the CNN module is used to extract detailed features of the data using multiple convolutional kernels of different sizes, and to select the maximum value of the feature map through a pooling layer to further aggregate the features. The decoding layer is equipped with a CRF module, which is used to obtain the globally optimal label sequence of an entity; Ultimately, the system extracts triples from entity relationships and constructs a knowledge graph of tight sandstone reservoirs that combines entity relationships.

5. The knowledge graph-based evaluation method for tight sandstone reservoirs according to claim 4, characterized in that: The CNN module uses 1×3 convolutions to capture entity features in individual geological terms and reservoir texts, and 5×5 convolutions to capture combinations of geological terms and their contextual relationships.

6. The knowledge graph-based evaluation method for tight sandstone reservoirs according to claim 5, characterized in that: The loss function of the rule-constrained prediction layer is a dynamic weighted loss function based on classification cross-entropy, regression mean squared error, and rule constraints: ； in, This is a soft rule error term. This is the model's predicted output for the sample. For reference output related to soft rules, The loss function for the rule-constrained prediction layer. The basic prediction loss of the model, Represents a set of hard rules. Represents a set of soft rules. Penalties for violating hard rules This is a soft rule error term. and It is a weight parameter used to control the degree of influence of rule constraints.

7. A knowledge graph-based evaluation system for tight sandstone reservoirs, characterized in that: The system has a program module corresponding to the steps of the method described in any one of claims 1 to 6, and executes the steps in the knowledge graph-based tight sandstone reservoir evaluation method described above when running.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program configured to, when invoked by a processor, implement the steps of the knowledge graph-based tight sandstone reservoir evaluation method according to any one of claims 1 to 6.

Images