An oil and gas reservoir parameter intelligent inversion method and system based on cloud edge fusion

Abstract

The application discloses an oil and gas reservoir parameter intelligent inversion method and system based on cloud edge fusion, and belongs to the field of cloud edge collaborative calculation. The method is as follows: receiving a real-time causal diagram constructed by an edge device according to a real-time inversion network model, wherein the real-time causal diagram is constructed and acquired from real-time logging data collected by the edge and first reservoir parameters obtained through inversion; extracting a causal conflict feature between the real-time causal diagram and each preset benchmark causal diagram; if the causal conflict feature meets an inversion anomaly judgment condition, determining an abnormal cause of the real-time inversion network model accordingly; generating an adjustment configuration parameter according to the causal conflict feature and the abnormal cause, and delivering the adjustment configuration parameter to the edge device to adjust the real-time inversion network model, and then performing reservoir parameter inversion. The application improves the matching degree of the edge model adjustment parameter. The application can improve the matching degree of the edge model adjustment parameter when the edge model inversion fails.

Description

Technical Field

[0001] This application relates to the field of cloud-edge collaborative computing, and in particular to an intelligent inversion method and system for oil and gas reservoir parameters based on cloud-edge fusion. Background Technology

[0002] In modern oil and gas exploration and development, the inversion of reservoir parameters (such as porosity, permeability, and oil saturation) in drilling sites is crucial for real-time evaluation of formation potential. Due to the high concurrency and real-time nature of logging data, and the fact that the deep learning network models relied upon for inversion often involve massive computational demands and require support from vast amounts of historical data from multiple wells, a cloud-edge fusion architecture—combining cloud-based data management and model control with lightweight, real-time inversion at the edge—has become an inevitable choice for achieving intelligent oil and gas reservoir inversion.

[0003] However, in practical applications, existing cloud-edge fusion inversion methods typically involve edge devices only receiving the basic inversion model from the cloud and directly applying it to predict real-time logging data in the current drilling area. Due to the extremely complex and variable underground geological conditions, and the unpredictable systematic drift of sensors at the drilling site, existing technologies cannot automatically detect the failure of the inversion mechanism at the edge, nor can they accurately trace and diagnose whether the anomaly is caused by data distortion due to front-end sensor failure or model mismatch due to drastic changes in natural geological mechanisms. This further prevents the cloud server from providing matching adjustment strategies to the edge devices, often resulting in blind retraining or frequent full-scale switching of the network model. This not only wastes significant cloud-edge communication bandwidth and computing resources but also leads to frequent collapses or severe accuracy distortions in the edge-end intelligent inversion system. Therefore, improving the matching degree of edge model adjustment parameters when edge model inversion fails has become a pressing technical problem in the field. Summary of the Invention

[0004] This application provides a cloud-edge fusion-based intelligent inversion method and system for oil and gas reservoir parameters, which can solve the technical problem of how to improve the matching degree of edge model adjustment parameters when the edge model inversion fails.

[0005] This application provides a cloud-edge fusion-based intelligent inversion method for oil and gas reservoir parameters, applied to a cloud server. The cloud server communicates with edge devices, which are located in corresponding drilling areas. The edge devices are configured with a real-time inversion network model for oil and gas reservoir parameter inversion. The cloud-edge fusion-based intelligent inversion method for oil and gas reservoir parameters includes:

[0006] The edge device receives a real-time causal graph constructed based on the real-time inversion network model; wherein the real-time causal graph is constructed based on the real-time logging data of the drilling area collected by the edge device and the first oil and gas reservoir parameters obtained by the real-time inversion network model based on the real-time logging data. Extract causal conflict features between the real-time causal graph and each preset benchmark causal graph; the causal conflict features include: a first causal edge difference degree characterizing the causal difference between the real-time causal graph and the benchmark causal graph, and a statistical distribution divergence characterizing the data distribution difference between the real-time causal graph and the benchmark causal graph; If the causal conflict features meet the preset inversion anomaly determination conditions, then the anomaly cause of the real-time inversion network model is determined according to the causal conflict features; wherein, each of the reference causal graphs corresponds to a detected drilling area and a reference inversion network model. Based on the causal conflict characteristics and the causes of anomalies, adjustment configuration parameters for the real-time inversion network model are generated. The adjustment configuration parameters are then sent to the edge device to adjust the real-time inversion network model. Subsequently, the oil and gas reservoir parameters of the drilling area are inverted using the adjusted real-time inversion network model.

[0007] Compared with existing technologies, the above embodiments have the following beneficial effects: The impact mechanisms of different physical failures on graph topology and data status in oil and gas inversion systems are fundamentally different. For example, front-end sensor failures typically only cause systematic deviations in logging values, without altering the inherent causal response chain of the formation's rock physics; while natural geological mechanism anomalies fundamentally disrupt the originally stable causal logic of geophysical responses. This application constructs a real-time causal graph by integrating logging data and inverted reservoir parameters at the edge device. Furthermore, by comparing the causal conflict characteristics of the real-time causal graph constructed at the edge with a preset baseline causal graph in the cloud, it obtains the statistical distribution divergence of the first causal edge difference and data distribution differences, thereby distinguishing different causes of inversion failure. This provides an objective basis for the cloud to generate and distribute matching adjustment configuration parameters for specific failure mechanisms, thus improving the matching degree of edge model adjustment parameters and avoiding the bandwidth and computing power waste caused by blind retraining in traditional technologies.

[0008] Further, the real-time causal graph includes several real-time causal nodes, real-time logging data or first oil and gas reservoir parameters corresponding to each real-time causal node, and several real-time causal edges; the baseline causal graph includes baseline causal nodes corresponding one-to-one with the real-time causal nodes, historical baseline data distributions corresponding to each baseline causal node, and several baseline causal edges; the extraction of causal conflict features between the real-time causal graph and the preset baseline causal graphs includes: For any of the aforementioned baseline causal graphs, the first causal edge difference degree between all the corresponding baseline causal edges and all the real-time causal edges is calculated based on the structural Hamming distance; For any of the aforementioned baseline causal graphs, calculate the statistical distribution divergence between each of the real-time logging data and the historical baseline data distribution corresponding to the baseline causal node in the baseline causal graph; For any of the aforementioned baseline causal graphs, the corresponding first causal edge difference degree and the statistical distribution divergence are matrix-correlated to construct the causal conflict features between the real-time causal graph and the baseline causal graph.

[0009] Compared to existing technologies, the above embodiments have the following advantages: In graph theory, structural Hamming distance can losslessly characterize the absolute discrete algebraic variation of topological edge additions, deletions, and reversals, while statistical distribution divergence in probability theory can sensitively measure the continuous displacement of the probability density function of a continuous variable. Therefore, this application, by calculating the first causal edge difference degree based on structural Hamming distance and calculating the statistical distribution divergence between each real-time logging data and the historical benchmark data distribution corresponding to the benchmark causal node, and performing matrix correlation, provides reliable data basis for subsequent anomaly cause determination, thereby improving the matching degree of the final adjusted configuration parameters.

[0010] Furthermore, the causes of the anomalies include: sensor malfunction and geological mechanism mismatch; determining the causes of the anomalies in the real-time inversion network model based on the causal conflict characteristics includes: If there exists a target baseline causal graph, the difference degree of the first causal edge corresponding to the target baseline causal graph is less than a preset first difference judgment threshold, and there exists at least one target baseline causal node in the target baseline causal graph whose statistical distribution divergence is greater than a preset distribution drift threshold, then the cause of the anomaly is determined to be the sensor malfunction. If the first causal edge difference degree corresponding to all the benchmark causal graphs is greater than or equal to the first difference determination threshold, then the anomaly is determined to be caused by the geological mechanism matching anomaly.

[0011] Compared with the prior art, the above embodiments have the following beneficial effects: Based on the physical nature that sensor failure does not change the causal chain but geological changes destroy the causal chain, this application sets a first difference judgment threshold to delineate the upper limit of the structural tolerance of the unchanged stratigraphic framework in the topological space, and further sets a distribution drift threshold to delineate the statistical disturbance boundary of normal data in the probability space, forming absolutely mutually exclusive screening judgment conditions on the decision chain, providing a high-confidence classification result for the inversion of the cause of anomalies for subsequent classification and policy implementation.

[0012] Furthermore, the edge device is equipped with a plurality of sensors, each of which collects a corresponding real-time logging data; if the anomaly is caused by a sensor malfunction, the step of generating adjustment configuration parameters for the real-time inversion network model based on the causal conflict characteristics and the cause of the anomaly, and sending the adjustment configuration parameters to the edge device to adjust the real-time inversion network model, includes: Based on the correspondence between the target reference causal node, the real-time logging data and the sensor, the fault sensor corresponding to each target reference causal node is determined; Based on the statistical distribution divergence corresponding to each of the target real-time causal nodes, the data calibration mapping coefficients corresponding to the fault sensors are determined by the statistical distribution alignment method. The data calibration mapping coefficients are sent to the edge device as adjustment configuration parameters to instruct the edge device to calibrate the real-time logging data collected by the corresponding fault sensor in real time through the data calibration mapping coefficients, thereby completing the adjustment of the input data of the real-time inversion network model.

[0013] Compared with the prior art, the above embodiments have the following beneficial effects: Since the fault caused by the systematic drift of the sensor is only a numerical surface source pollution of the data front end, the internal mapping weight of the original high-precision inversion model itself has not undergone physical degradation. Therefore, this application does not adopt a high-energy-consuming online model training strategy, but locks the faulty sensor through spatial correspondence and uses the statistical distribution alignment method to determine the data calibration mapping coefficient online, and calibrates the well logging data collected by the sensor in real time, thereby realizing the data reconstruction of the data input side of the inversion network model at the edge. It not only achieves data correction with extremely low edge computing power overhead and extremely high real-time response speed, but also protects the edge inversion model from being misled by erroneous input sources to the greatest extent.

[0014] Further, if the anomaly is determined to be caused by the geological mechanism mismatch, before generating the adjusted configuration parameters of the real-time inversion network model based on the causal conflict characteristics and the anomaly cause, the following steps are included: Real-time well logging self-causal graphs are extracted from the real-time causal graphs, and reference well logging self-causal graphs are extracted from each of the reference causal graphs; wherein, the real-time well logging self-causal graph includes only the first real-time causal node corresponding to the real-time well logging data and the first real-time causal edge between the first real-time causal nodes; the reference well logging self-causal graph includes only the first reference causal node corresponding one-to-one with the first real-time causal node and the first reference causal edge between the first reference causal nodes. For any of the aforementioned benchmark logging self-causal graphs, based on the structural Hamming distance, the second causal edge difference degree between all the corresponding first benchmark causal edges and all the first real-time causal edges is calculated; If at least one target benchmark logging self-causal graph has a second causal edge difference degree corresponding to it that is less than a preset second difference determination threshold, then the geological mechanism matching anomaly is determined to be a false anomaly; otherwise, the geological mechanism matching anomaly is determined to be a true anomaly.

[0015] Compared to existing technologies, the above embodiments have the following beneficial effects: Since conflicts in the real-time causal graph may originate from actual formation anomalies or simply from incorrect inversion models at the edges leading to distorted output parameters and contaminated graph topology, it is necessary to accurately distinguish between the two situations. Furthermore, since the well logging self-causal graph is entirely determined by the physical response of the natural formation, it naturally possesses physical isolation properties that shield against network model inversion errors. Therefore, by extracting the well logging self-causal graph containing only the causal relationships between natural well logging data from the full graph and calculating the second causal edge difference degree, this application can objectively identify, from the most fundamental rock physical causal response level, whether it is a genuine formation mechanism anomaly or a simple model mismatch, thereby further improving the matching degree of subsequent generated and adjusted configuration parameters.

[0016] Further, if the geological mechanism matching anomaly is a true anomaly, the step of generating adjustment configuration parameters for the real-time inversion network model based on the causal conflict characteristics and the anomaly causes, and distributing the adjustment configuration parameters to the edge device to adjust the real-time inversion network model, includes: Based on all the first real-time causal edges in the real-time logging self-causal graph, construct a target topological adjacency matrix that characterizes the current geological and physical causal response of the drilling area; The target topology adjacency matrix is ​​combined with preset weight coefficients to form the adjustment configuration parameters, and the adjustment configuration parameters are sent to the edge device. The edge device is instructed to calculate the matrix cross-entropy loss value between the intermediate feature correlation matrix generated by the real-time inversion network model during the inversion calculation process and the target topological adjacency matrix, so as to generate a causal topological regularization penalty term. The edge device is instructed to add the causal topology regularization penalty term to the original loss function of the real-time inversion network model through the weight coefficient to generate the target total loss function. Then, with the minimization of the target total loss function as the convergence objective, the real-time inversion network model is backpropagated to update the gradient, thereby completing the adjustment of the real-time inversion network model.

[0017] Compared to existing technologies, the above embodiments have the following beneficial effects: Traditional incremental fine-tuning of neural networks, lacking physical prior constraints, is prone to overfitting or ill-conditioned convergence when faced with real changes in formation mechanisms. However, the target topological adjacency matrix generated from the well logging self-causal graph, which characterizes the causal relationships between natural well logging data, provides a strong constraint on the real geological and physical causal relationships of the current work area. Therefore, this application utilizes the real-time well logging self-causal graph to construct the target topological adjacency matrix. By calculating the matrix cross-entropy loss value, a causal topological regularization penalty term is introduced into the original loss function for gradient updates. This forces the neural network to strictly adhere to the physical evolution logic of the new formation when optimizing weights, effectively ensuring the accuracy of the edge-based inversion network model adjustment.

[0018] Further, if the geological mechanism matching anomaly is a false anomaly, the step of generating adjustment configuration parameters for the real-time inversion network model based on the causal conflict characteristics and the cause of the anomaly, and distributing the adjustment configuration parameters to the edge device to adjust the real-time inversion network model, includes: From all the aforementioned benchmark logging autocausal graphs, select the target benchmark logging autocausal graph with the smallest second causal edge difference; The model parameters of the benchmark inversion network model corresponding to the benchmark causal graph corresponding to the target benchmark logging self-causal graph are used as the adjustment configuration parameters; The adjusted configuration parameters are sent to the edge device so that the edge device replaces the model parameters of the real-time inversion network model with the adjusted configuration parameters.

[0019] Compared with existing technologies, the above embodiments have the following beneficial effects: When the well logging self-causal graph is highly matched with a certain benchmark subgraph in the library, it fully demonstrates that the natural physical laws of the current new drilling area belong entirely to the known geological facies zone that has been explored. At this time, the model mismatch is a pure logical mounting error, and no incremental model training is required. Therefore, when it is determined to be a false anomaly, by comparing the well logging self-causal graph, the mature benchmark network model parameters with the smallest difference of the second causal edge are retrieved from the cloud library and the edge end is replaced as a whole. This achieves zero consumption of computing power and zero time delay self-healing at the edge device end, and restores the inversion accuracy to the design index within the shortest cloud-edge communication cycle.

[0020] Further, before receiving the real-time causal graph constructed by the edge device based on the real-time inversion network model, the following steps are included: Receive the first geographic coordinates corresponding to the drilling area sent by the edge device; Calculate the spatial distance between the first geographic coordinates and the second geographic coordinates of each of the detected drilling areas; The baseline inversion network model corresponding to the detected drilling area with the smallest spatial distance is sent to the edge device as the real-time inversion network model.

[0021] Compared with the prior art, the above embodiments have the following beneficial effects: Since the sedimentary evolution of geological facies zones and the distribution of reservoir properties have natural continuity, gradation and strong prior similarity in three-dimensional geographic space, this application calculates the spatial distance to retrieve and distribute the benchmark inversion network model with the smallest distance as the initial model before receiving the real-time causal graph. This greatly reduces the cold start prediction error in the initial operation stage of the system from the source, and at the same time greatly reduces the topological search space for subsequent anomaly cause determination.

[0022] Another embodiment of this application also provides an intelligent inversion system for oil and gas reservoir parameters based on cloud-edge fusion, which is applied to a cloud server. The cloud server communicates with edge devices, which are set in the corresponding drilling area. The edge devices are configured with a real-time inversion network model for oil and gas reservoir parameter inversion. The intelligent inversion system for oil and gas reservoir parameters based on cloud-edge fusion includes: a receiving module, a feature extraction module, an anomaly cause determination module, and an adjustment parameter distribution module. The receiving module is used to receive the real-time causal graph constructed by the edge device based on the real-time inversion network model; wherein, the real-time causal graph is constructed based on the real-time logging data of the drilling area collected by the edge device and the first oil and gas reservoir parameters obtained by the real-time inversion network model based on the real-time logging data. The feature extraction module is used to extract causal conflict features between the real-time causal graph and each preset benchmark causal graph; the causal conflict features include: a first causal edge difference degree characterizing the causal difference between the real-time causal graph and the benchmark causal graph, and a statistical distribution divergence characterizing the data distribution difference between the real-time causal graph and the benchmark causal graph; The anomaly cause determination module is used to determine the anomaly cause of the real-time inversion network model based on the causal conflict characteristics if the causal conflict characteristics meet the preset inversion anomaly determination conditions; wherein, each of the reference causal graphs corresponds to a detected drilling area and a reference inversion network model. The adjustment parameter distribution module is used to generate adjustment configuration parameters for the real-time inversion network model based on the causal conflict characteristics and the causes of anomalies, and distribute the adjustment configuration parameters to the edge device to adjust the real-time inversion network model, and then use the adjusted real-time inversion network model to perform oil and gas reservoir parameter inversion in the drilling area.

[0023] Further, the real-time causal graph includes several real-time causal nodes, real-time logging data or first oil and gas reservoir parameters corresponding to each real-time causal node, and several real-time causal edges; the baseline causal graph includes baseline causal nodes corresponding one-to-one with the real-time causal nodes, historical baseline data distributions corresponding to each baseline causal node, and several baseline causal edges; the feature extraction module includes: a first calculation unit, a second calculation unit, and an association unit; the feature extraction module is used to extract causal conflict features between the real-time causal graph and preset baseline causal graphs, including: The first calculation unit is configured to, for any of the aforementioned baseline causal graphs, calculate the first causal edge difference degree between all the corresponding baseline causal edges and all the real-time causal edges based on the structural Hamming distance; The second calculation unit is used to calculate, for any of the aforementioned reference causal graphs, the statistical distribution divergence between each of the real-time logging data and the historical reference data distribution corresponding to the reference causal node in the reference causal graph; The association unit is used to perform matrix association of the corresponding first causal edge difference degree and the statistical distribution divergence for any of the baseline causal graphs, and construct the causal conflict features between the real-time causal graph and the baseline causal graph. Attached Figure Description

[0024] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0025] Figure 1 This is a flowchart illustrating an intelligent inversion method for oil and gas reservoir parameters based on cloud-edge fusion provided in some embodiments of this application. Figure 2 This is a schematic diagram of the structure of a cloud-edge collaborative computing system provided in some embodiments of this application; Figure 3 This is a schematic diagram of the structure of an intelligent inversion system for oil and gas reservoir parameters based on cloud-edge fusion provided in some embodiments of this application. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0027] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0028] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0029] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0030] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0031] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0032] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0033] In practical applications, existing cloud-edge fusion inversion methods typically involve edge devices only receiving the basic inversion model from the cloud and directly applying it to predict real-time logging data in the current drilling area. Due to the extremely complex and variable underground geological conditions, and the unpredictable systematic drift of sensors at the drilling site, existing technologies cannot automatically detect the failure of the inversion mechanism at the edge, nor can they accurately trace and diagnose whether the anomaly is caused by data distortion due to front-end sensor failure or model mismatch due to drastic changes in natural geological mechanisms. This further prevents the cloud server from providing matching adjustment strategies to the edge devices, often resulting in blind retraining or frequent full-scale switching of the network model. This not only wastes significant cloud-edge communication bandwidth and computing resources but also leads to frequent collapses or severe accuracy distortions in the intelligent inversion system at the edge.

[0034] Please refer to Figure 1 To address the issue of improving the matching degree of edge model adjustment parameters when edge model inversion fails in existing technologies, this application provides an intelligent inversion method for oil and gas reservoir parameters based on cloud-edge fusion. The method is applied to a cloud server, which communicates with edge devices. The edge devices are located in the corresponding drilling area and are configured with a real-time inversion network model for oil and gas reservoir parameter inversion.

[0035] refer to Figure 2 This application provides a cloud-edge collaborative computing system, which, in terms of physical topology and logical control, is divided into a centrally managed cloud server and distributed edge devices deployed at various drilling sites. Figure 2As shown, at the production front line, edge devices A, B, C, and D utilize locally configured real-time inversion network models to perform low-latency oil and gas reservoir parameter inversion on natural logging data collected by field sensors. At the remote center, a cloud server cluster acts as the global coordinating brain, centrally maintaining a massive baseline causal graph library of known drilling areas. This baseline causal graph library includes several preset baseline causal graphs and a corresponding baseline inversion network model for each baseline causal graph. Each baseline causal graph is constructed based on the causal response mechanism between the historical baseline data distribution and historical reservoir parameters of the corresponding explored drilling area. The corresponding baseline inversion network model is obtained through supervised learning training based on historical multi-well logging inversion sample data of the corresponding explored drilling area.

[0036] Preferably, in some embodiments of this application, before receiving the real-time causal graph constructed by the edge device according to the real-time inversion network model, the following steps are included: Receive the first geographic coordinates corresponding to the drilling area sent by the edge device; Calculate the spatial distance between the first geographic coordinates and the second geographic coordinates of each of the detected drilling areas; The baseline inversion network model corresponding to the detected drilling area with the smallest spatial distance is sent to the edge device as the real-time inversion network model.

[0037] Preferably, in some embodiments of this application, before the cloud server receives the real-time causal graph uploaded by the edge device, the cloud server first initializes the edge device using geospatial prior information. Specifically, the cloud server receives the first geographic coordinates (such as the latitude and longitude coordinates of the wellhead) corresponding to the current drilling area from the edge device; subsequently, the calculation module within the cloud server compares the first geographic coordinates with the second geographic coordinates of each of the detected drilling areas in the benchmark causal graph library, and calculates the Euclidean distance or great circle route distance between them to obtain the corresponding spatial distance. Since sedimentary facies zones have natural continuity and similar geological properties in similar geographic spaces, the closer the spatial distance, the more consistent the stratigraphic structure and rock physical response patterns tend to be. In view of this, the cloud server retrieves the detected drilling area with the smallest spatial distance from the benchmark causal graph library and sends the high-precision benchmark inversion network model parameters corresponding to this area to the edge device, directly using it as its initial real-time inversion network model.

[0038] The above mechanism introduces the most relevant spatial continuity prior constraint in the early stage when the edge equipment has just entered the new work area and has not yet accumulated local well logging causal data. This effectively overcomes the inversion prediction error during the cold start of the system and significantly narrows the topology search space when determining the cause of anomalies.

[0039] Furthermore, in some embodiments of this application, the intelligent inversion method for oil and gas reservoir parameters based on cloud-edge fusion includes the following steps S101 to S104: S101: Receive the real-time causal graph constructed by the edge device based on the real-time inversion network model; wherein the real-time causal graph is constructed and obtained based on the real-time logging data of the drilling area collected by the edge device and the first oil and gas reservoir parameters obtained by the real-time inversion network model based on the real-time logging data.

[0040] Preferably, in some embodiments of this application, both the real-time inversion network model and the baseline inversion network model are deep learning network structures with temporal feature extraction capabilities (e.g., using one-dimensional residual convolutional neural networks (1D-ResNet) or long short-term memory networks (LSTM). The input layer of such network models is connected to the logging data acquisition channel of the corresponding drilling area, used to receive one-dimensional temporal data such as natural gamma and sonic transit time arranged in depth sequence in real time; the backbone network layer of the model uses cascaded convolutional kernels or recurrent gating units to perform multi-scale feature extraction and deep abstract mapping of the nonlinear evolution trend of logging data with formation depth. The aforementioned temporal mapping structure, because its internal weight matrix and bias parameters can perfectly fit the continuous transition law of rock physical properties in the vertical sedimentary sequence, can predict oil and gas reservoir parameters such as porosity and permeability distributed along depth through parallel inversion of the model's output layer.

[0041] Preferably, in some embodiments of this application, the first oil and gas reservoir parameters are obtained by inverting real-time logging data collected by inputting it into a real-time inversion network model. The real-time logging data includes, but is not limited to, conventional one-dimensional logging curves such as natural gamma, sonic transit time, compensated density, compensated neutron, and shallow and deep three-lateral resistivity at different detection depths, collected during drilling or well completion.

[0042] Furthermore, in some embodiments of this application, the real-time causal graph includes a plurality of real-time causal nodes, real-time logging data or first oil and gas reservoir parameters corresponding to each real-time causal node, and a plurality of real-time causal edges; the baseline causal graph includes baseline causal nodes corresponding one-to-one with the real-time causal nodes, historical baseline data distribution corresponding to each baseline causal node, and a plurality of baseline causal edges.

[0043] Preferably, in some embodiments of this application, the real-time causal graph is constructed and obtained through the following graph construction mechanism: First, using sliding window technology, time-series or depth slice data of the real-time logging data curves and corresponding inverted first oil and gas reservoir parameters (such as porosity, permeability, and oil saturation) within the current tunneling segment are extracted and encapsulated into a one-dimensional continuous variable matrix; then, each of the real-time logging data and each of the first oil and gas reservoir parameters are declared as independent top-level topological variables locally at the edge, thereby mapping and generating several real-time causal nodes, so that each real-time causal node has its corresponding real-time logging data or first oil and gas reservoir parameter as its node observation attribute; then, the bar-based... Causal discovery algorithms based on independence tests or scoring functions (such as improved PC algorithms or the linear non-Gaussian acyclic model LiNGAM) are used to calculate online the directional information flow, conditional independence constraints, and causal constraint strength between each real-time causal node, thereby eliminating spurious correlations. Directed connections are established between two real-time causal nodes with deterministic causal constraints, thereby constructing several real-time causal edges representing physical flow. Finally, through these real-time causal nodes and the interconnected real-time causal edges, a topological directed acyclic graph that represents the fusion of the real physical logic of rocks in nature and the internal computational state of the network is dynamically reconstructed, which serves as the real-time causal graph.

[0044] Preferably, in some embodiments of this application, the construction process of the baseline causal graph is the same as that of the real-time causal graph. In addition, each baseline causal node in the baseline causal graph also embeds the historical baseline data distribution of the well logging data corresponding to that node.

[0045] S102: Extract causal conflict features between the real-time causal graph and each preset benchmark causal graph; the causal conflict features include: a first causal edge difference degree characterizing the causal difference between the real-time causal graph and the benchmark causal graph, and a statistical distribution divergence characterizing the data distribution difference between the real-time causal graph and the benchmark causal graph.

[0046] Furthermore, in some embodiments of this application, the extraction of causal conflict features between the real-time causal graph and preset benchmark causal graphs includes the following steps S1021 to S1023: S1021: For any of the aforementioned baseline causal graphs, calculate the first causal edge difference degree between all the corresponding baseline causal edges and all the real-time causal edges based on the structural Hamming distance; S1022: For any of the aforementioned baseline causal graphs, calculate the statistical distribution divergence between each of the real-time logging data and the historical baseline data distribution corresponding to the baseline causal node in the baseline causal graph; S1023: For any of the aforementioned baseline causal graphs, the corresponding first causal edge difference degree and the statistical distribution divergence are matrix-correlated to construct the causal conflict features between the real-time causal graph and the baseline causal graph.

[0047] For example, in some embodiments of this application, feature extraction is performed by a cloud server on a specific baseline causal graph (e.g., a first baseline graph). Assume the first baseline graph contains three baseline causal nodes, corresponding to natural gamma ray, sonic transit time, and porosity in the oil and gas reservoir parameters, respectively. In this first baseline graph, based on historical priors, there are two standard baseline causal edges: a directed edge from natural gamma ray to porosity, and a directed edge from sonic transit time to porosity, whose corresponding adjacency matrix represents a skeleton structure. At this time, the real-time causal graph uploaded by the edge device also contains the corresponding three real-time causal nodes: natural gamma ray, sonic transit time, and porosity. It should be noted that natural gamma ray and sonic transit time are real-time logging data, while porosity is an oil and gas storage parameter obtained by inversion from real-time logging data.

[0048] During step S1021, if the current edge device's calculated real-time causal edge changes due to geological mechanism anomalies, resulting in the following: the directed edge pointing from natural gamma to porosity remains unchanged, but the causal edge pointing from acoustic transit time to porosity breaks and disappears, while a reverse causal edge is added out of thin air, i.e., a directed edge from porosity to natural gamma. The graph analysis module of the cloud server identifies one missing edge and one newly added edge by comparing these two sets of directed edges. Based on the algebraic definition of structural Hamming distance, the total number of mismatched edges is accumulated to 2, so the first causal edge difference degree corresponding to the first reference graph is directly output as a scalar value of 2.

[0049] During the parallel execution of step S1022, the system extracts the probability density distribution of real-time logging data (such as a one-dimensional time-series set of natural gamma values) captured by the sliding window within the current time period, and calls the Jensen-Shannon divergence algorithm to compare it with the inherent historical baseline data distribution of the natural gamma nodes in the first baseline map. Assume that due to aging drift of the front-end natural gamma sensor, its real-time values ​​have shifted towards higher values, resulting in a calculated statistical distribution divergence of 0.45 for the natural gamma nodes; while the acoustic time-of-flight sensor is operating normally, and its corresponding node has a statistical distribution divergence of 0.02. The system then arranges the divergence values ​​corresponding to each node into a one-dimensional vector to generate the distribution divergence vector under the current baseline map.

[0050] In the final execution step S1023, to enable the single scalar and multi-dimensional vector to be jointly represented in the same tensor space, the cloud server constructs a specialized feature association matrix. In this embodiment, by using the first causal edge difference scalar value 2 as the first element of the matrix and performing matrix association operations such as concatenation or weighted multiplication with the aforementioned distribution divergence vector, a comprehensive feature vector is finally constructed. This matrix fully integrates structural variation information and numerical drift information, serving as the causal conflict feature between the real-time causal graph and the baseline causal graph, providing a completely digital input source for subsequent input judgment conditions.

[0051] In graph theory, structural Hamming distance can losslessly characterize the absolute discrete algebraic variation of topological edge additions, deletions, and reversals, while statistical distribution divergence in probability theory can sensitively measure the continuous displacement of the probability density function of a continuous variable. Therefore, this application calculates the first causal edge difference based on structural Hamming distance and performs matrix correlation on the statistical distribution divergence between each real-time logging data and the historical benchmark data distribution corresponding to the benchmark causal node, providing reliable data basis for subsequent anomaly cause determination, thereby improving the matching degree of the final adjusted configuration parameters.

[0052] S103: If the causal conflict features meet the preset inversion anomaly determination conditions, then the anomaly cause of the real-time inversion network model is determined according to the causal conflict features; wherein, each of the reference causal graphs corresponds to a detected drilling area and a reference inversion network model.

[0053] Preferably, in some embodiments of this application, the process of determining whether the causal conflict features meet the inversion anomaly determination conditions is as follows: The cloud server first configures a topological difference determination threshold for the graph topology structure within the system, and at the same time configures a distribution drift threshold for the probability deviation of the bottom logging data. When the first causal edge difference degree in the causal conflict feature matrix is ​​greater than the topological difference determination threshold or there exists a statistical distribution divergence greater than the distribution drift threshold, it means that the causal conflict features meet the inversion anomaly determination conditions. The topological difference determination threshold and the distribution drift threshold are determined by prior statistical analysis based on the historical logging inversion data and fault samples of the known work area, and by cross-calibration combined with rock physics sensitivity experiments and expert experience. This application does not limit the above thresholds.

[0054] Furthermore, in some embodiments of this application, the causes of anomalies include: sensor failure and geological mechanism mismatch; determining the causes of anomalies in the real-time inversion network model based on the causal conflict characteristics includes the following steps S1031 to S1032: S1031: If there exists a target baseline causal graph, the difference degree of the first causal edge corresponding to the target baseline causal graph is less than a preset first difference judgment threshold, and there exists at least one target baseline causal node in the target baseline causal graph whose statistical distribution divergence is greater than a preset distribution drift threshold, then the cause of the anomaly is determined to be the sensor malfunction.

[0055] As an example of the first scenario, it is assumed that a specific target baseline causal graph exists in the baseline causal graph library. Among the causal conflict features extracted by the cloud server for this target baseline causal graph, since the directed edges of the real-time causal graph are completely consistent with those of the target baseline causal graph, the first causal edge difference calculated based on the structural Hamming distance has a scalar value of 0. Meanwhile, in the data distribution dimension, due to hardware aging or wellbore mud contamination of the sensor responsible for collecting natural gamma data at the front end, the statistical distribution divergence calculation result for the natural gamma node is 0.45, while the statistical distribution divergences for other nodes such as sonic transit time and compensation density are all between 0.01 and 0.03. At this point, the diagnostic engine compares the input features to the anomaly detection criteria: it identifies that the first causal edge difference degree corresponding to the target baseline causal graph is less than the preset first difference detection threshold (e.g., set to 2), thus strongly proving at the physical level that the current natural rock physical causal response chain has not broken or reversed, and that the natural geological mechanism of the current drilling area is completely matched with the mature historical drilling area corresponding to the first baseline causal graph; at the data level, the statistical distribution divergence of the natural gamma target baseline causal node is 0.45, which is greater than the preset distribution drift threshold (e.g., set to 0.15). The system then determines that the anomaly in the current real-time inversion network model is caused by sensor failure.

[0056] S1032: If the first causal edge difference degree corresponding to all the reference causal graphs is greater than or equal to the first difference determination threshold, then the abnormality is determined to be caused by the geological mechanism matching abnormality.

[0057] As an example of the second scenario, suppose the drill bit controlled by the edge device drills into a new stratigraphic segment. At this point, the cloud server extracts causal conflict features from the real-time causal graph constructed at the edge and each of the preset benchmark causal graphs in the benchmark causal graph library. The calculation results show that due to the paired absence, addition, or reversal of directed edges, the first causal edge difference degree of all preset benchmark causal graphs is greater than or equal to the preset first difference judgment threshold. This feature combination, where no benchmark graph fits the current real-time topological framework, indicates mechanistically that the system is currently facing an anomaly in the geological mechanism matching. It should be noted that the underlying cause of the full excess of the first causal edge difference degree of all benchmark graphs has two possibilities: on the one hand, it may be that the causal logic of the actual rock physical response in nature has been fundamentally overturned, leading to a break in the inherent physical chain of the strata; on the other hand, it is also highly likely that the real-time inversion network model initially issued and configured on the edge device is severely mismatched with the current actual stratum type, causing severe distortion of the first oil and gas reservoir parameters output by the model at the inference end, which in turn contaminates the topological structure of the entire real-time causal graph from the output node end. Therefore, the system does not make an absolute single attribution at this stage, but instead classifies the cause of the anomaly as a geological mechanism matching anomaly that includes the two possible failures mentioned above. This provides a precise decision-making basis for further stripping away model interference and identifying true and false geological anomalies.

[0058] Based on the physical nature of the aforementioned sensor failures not altering the causal chain while geological anomalies disrupt it, this application sets a first difference judgment threshold to delineate the upper bound of the structural tolerance of the unchanged stratigraphic framework in the topological space. It further sets a distribution drift threshold to delineate the statistical perturbation boundary of normal data in the probability space, forming absolutely mutually exclusive screening judgment conditions on the decision chain. This provides a high-confidence classification result for the inversion of anomaly causes for subsequent classification and policy implementation.

[0059] Furthermore, in some embodiments of this application, if the cause of the anomaly is determined to be the geological mechanism mismatch, before generating the adjusted configuration parameters of the real-time inversion network model based on the causal conflict characteristics and the cause of the anomaly, the following steps T1021 to T1023 are included: T1021: Extract a real-time well logging auto-causal graph from the real-time causal graph, and extract a benchmark well logging auto-causal graph from each of the benchmark causal graphs; wherein, the real-time well logging auto-causal graph includes only the first real-time causal node corresponding to the real-time well logging data and the first real-time causal edge between the first real-time causal nodes; the benchmark well logging auto-causal graph includes only the first benchmark causal node corresponding one-to-one with the first real-time causal node and the first benchmark causal edge between the first benchmark causal nodes.

[0060] To prevent the reverse contamination of the causal topology caused by the computational bias of the real-time inversion network model itself, the system enables a self-verification mechanism at the second-level subgraph level.

[0061] For example, consider a real-time causal graph currently processing three nodes: natural gamma, sonic transit time, and porosity as an output. The cloud server performs a subgraph pruning operation, directly removing the porosity node and all directed edges connected to it from the real-time causal graph. This extracts a real-time logging self-causal graph containing only the natural gamma node and the sonic transit time node, along with the first real-time causal edge between them. Simultaneously, the system uses a completely equivalent pruning rule to filter all benchmark causal graphs in the full benchmark causal graph library, similarly removing reservoir parameter output nodes such as porosity. This extracts benchmark logging self-causal graphs containing only the first benchmark causal nodes and first benchmark causal edges that correspond one-to-one with the aforementioned nodes. In this state, the directed edges within the real-time logging self-causal graph and each benchmark logging self-causal graph purely represent the geological genetic interweaving relationships between different logging physical quantities in nature, completely independent of the internal parameters of any deep learning model.

[0062] It is understood that the real-time logging self-causal graph is constructed by stripping the topological subgraph of the reservoir parameter nodes from the real-time causal full graph containing logging nodes and reservoir parameter nodes, thereby retaining only the pure logging data nodes and the causal edges between them.

[0063] T1022: For any of the aforementioned benchmark logging self-causal graphs, calculate the second causal edge difference between all the corresponding first benchmark causal edges and all the first real-time causal edges based on the structural Hamming distance.

[0064] Subsequently, for any of the aforementioned benchmark well logging autocausal graphs, the system calculates the second causal edge difference between all the first benchmark causal edges in its skeleton and all the first real-time causal edges in the real-time well logging autocausal graph, based on the structural Hamming distance. This application embodiment does not elaborate on this process.

[0065] T1023: If the difference degree of the second causal edge corresponding to at least one target benchmark logging self-causal graph in the benchmark logging self-causal graph is less than the preset second difference judgment threshold, then the geological mechanism matching anomaly is determined to be a false anomaly; otherwise, the geological mechanism matching anomaly is determined to be a true anomaly.

[0066] For example, regarding the determination of false anomalies, assuming that when comparing a specific second benchmark logging autocausal graph, it is found that the first real-time causal edge (such as the directed edge of natural gamma pointing to sonic transit time) in the real-time logging autocausal graph perfectly matches the first benchmark causal edge in the second benchmark logging autocausal graph, the calculated output difference of the second causal edge is a scalar value of 0, which is compared with a preset second difference judgment threshold (e.g., set to 1), identifying that the scalar value 0 is less than the preset second difference judgment threshold. This feature provides a decisive criterion in terms of physical mechanism: proving that the pure rock physical properties of the natural bottom layer of the current new work area are consistent with the known historical work area corresponding to the second benchmark logging autocausal graph, and that the previous full graph topology collapse was due to an incorrect version of the real-time inversion network model running on the current edge device (e.g., incorrectly configured model parameters suitable for other work areas), resulting in severe distortion of the porosity output by the model and conversely destroying the full graph. Based on this, the system determined that the geological mechanism matching anomaly was a false anomaly and locked the second benchmark logging self-causal map as the target benchmark logging self-causal map, providing a precise basis for subsequently retrieving mature high-precision model parameters of the target work area directly from the cloud for edge coverage fine-tuning.

[0067] For example, regarding the determination of genuine anomalies, if the system has traversed all the benchmark well logging self-causal graphs in the benchmark library, and the drill bit has indeed entered a new type of rock physical property formation that has never been seen before (e.g., suddenly entering an igneous intrusion from sedimentary rocks), causing the pure well logging response law in nature to be reshaped, the calculation results show that the second causal edge difference degree corresponding to all the benchmark well logging self-causal graphs is a scalar value of 2 or 3. That is, there is no benchmark well logging self-causal graph with a difference degree less than the preset second difference judgment threshold, and all data are greater than or equal to this threshold. This feature, where the topology of the entire library cannot be matched, after excluding model interference, fully demonstrates that the real rock physical causal response mechanism of nature in the current work area has undergone a fundamental change from nothing to something. Based on this, the system eliminates any doubts about model configuration errors and determines that the geological mechanism matching anomaly is a genuine anomaly, thereby guiding the cloud server to correct the model parameters of the real-time inversion model.

[0068] It should be noted that the first difference judgment threshold, the second difference judgment threshold, and the distribution drift threshold mentioned in the above embodiments are comprehensively set by performing prior statistics and distribution fitting on historical logging inversion data and past fault samples of known drilling areas, and by combining the rock physical sensitivity test results of specific strata with the experience of engineering experts. This application does not limit the specific setting method of the above thresholds.

[0069] Since conflicts in the real-time causal graph can originate from either actual formation anomalies or simply from incorrect inversion models used at the edges, leading to distorted output parameters and contaminating the graph topology, it is crucial to accurately distinguish between these two scenarios. Furthermore, because well logging self-causal graphs are entirely determined by the physical responses of the natural formation, they inherently possess physical isolation properties that shield against network model inversion errors. Therefore, this application, by extracting a well logging self-causal graph containing only the causal relationships between natural well logging data from the full graph and calculating the second causal edge difference degree, can objectively identify, from the most fundamental level of rock physical causal response, whether it is a genuine formation mechanism anomaly or simply a model mismatch, thereby further improving the matching degree of subsequent parameter adjustments.

[0070] S104: Generate adjustment configuration parameters for the real-time inversion network model based on the causal conflict characteristics and the causes of anomalies, and send the adjustment configuration parameters to the edge device to adjust the real-time inversion network model, and then use the adjusted real-time inversion network model to perform oil and gas reservoir parameter inversion in the drilling area.

[0071] Furthermore, in some embodiments of this application, the edge device is configured with a plurality of sensors, each of which collects one set of real-time logging data.

[0072] Furthermore, in some embodiments of this application, if the cause of the anomaly is a sensor malfunction, the step of generating adjustment configuration parameters for the real-time inversion network model based on the causal conflict characteristics and the cause of the anomaly, and sending the adjustment configuration parameters to the edge device to adjust the real-time inversion network model, includes the following steps T4011 to T4013: T4011: Based on the correspondence between the target reference causal node, the real-time logging data and the sensor, determine the fault sensor corresponding to each target reference causal node.

[0073] For example, assuming the target baseline causal graph currently locked by the cloud server contains a natural gamma node where the data distribution dimension exceeds the limit, the cloud server first retrieves a preset work area equipment topology mapping table. Based on the correspondence between the natural gamma node and its corresponding real-time logging data (i.e., the currently acquired one-dimensional natural gamma depth sequence curve) and the physical acquisition probes at the drilling site, the cloud server traces the source to determine that the faulty sensor corresponding to the natural gamma node is the natural gamma sensor hardware currently deployed in the downhole logging assembly at the wellhead.

[0074] T4012: Based on the statistical distribution divergence corresponding to each of the target real-time causal nodes, determine the data calibration mapping coefficients corresponding to the fault sensor using a statistical distribution alignment method.

[0075] For example, following the specific example corresponding to T4011 above, the statistical distribution divergence corresponding to the target real-time causal node currently calculated by the natural gamma node is extracted. To eliminate the systematic overall drift of measurement values ​​caused by sensor hardware aging or wellbore mud contamination, the cloud server initiates a statistical distribution alignment method. Specifically, the system retrieves the probability density distribution curve of the real-time natural gamma logging data collected within the current time window, as well as the inherent historical baseline data distribution curve of the node in the first baseline causal graph. Through quantile alignment mapping or mean-variance matching algorithms, the system online solves for the transformation characteristics that can losslessly superimpose the current distorted distribution curve to the standard historical distribution curve. In this embodiment, after probability distribution reconstruction calculation, the data calibration mapping coefficients corresponding to the faulty sensor are finally determined to include a linear scaling multiplier and a translation correction.

[0076] T4013: The data calibration mapping coefficients are sent to the edge device as the adjustment configuration parameters to instruct the edge device to calibrate the real-time logging data collected by the corresponding fault sensor in real time through the data calibration mapping coefficients, thereby completing the adjustment of the input data of the real-time inversion network model.

[0077] For example, the cloud server encapsulates the aforementioned digitized data calibration mapping coefficients as adjustment configuration parameters specific to the edge devices of this work area, and distributes them to the edge devices of the current drilling work area via the industrial wide area network. Upon receiving these adjustment configuration parameters, the edge devices directly inject them into the real-time computation stream of the edge data preprocessing control module. In this state, the edge devices do not need to stop work, unload, or replace physical probes. Instead, they instruct their internal data preprocessing control module to directly perform online real-time calibration of the one-dimensional natural gamma real-time logging data at each depth point subsequently acquired by the faulty sensor, multiplying by a linear scaling multiplier and adding a translation correction. This ensures that the probability distribution of the calibrated natural gamma data stream is strictly aligned with the historical baseline distribution. This mechanism, without changing the weights of the underlying neural network of the real-time inversion network model, directly adjusts the input data of the real-time inversion network model at the data input source, completely eliminating numerical distortion contamination caused by sensor hardware failure. This allows the accuracy of the first oil and gas reservoir parameters predicted by the edge real-time inversion network model to quickly recover to the design requirements.

[0078] Since the faults caused by the systematic drift of the sensor system are only numerical surface source contamination at the data front end, and the internal mapping weights of the originally high-precision inversion model itself have not undergone physical degradation, this application does not adopt a high-energy-consuming online model training strategy. Instead, it locks the faulty sensor through spatial correspondence and uses the statistical distribution alignment method to determine the data calibration mapping coefficients online, thereby calibrating the well logging data collected by the sensor in real time. This achieves data reconstruction on the data input side of the inversion network model at the edge, which not only achieves data correction with extremely low edge computing power overhead and extremely high real-time response speed, but also protects the edge inversion model from being misled by erroneous input sources to the greatest extent.

[0079] Furthermore, in some embodiments of this application, if the geological mechanism matching anomaly is a true anomaly, the step of generating adjustment configuration parameters for the real-time inversion network model based on the causal conflict characteristics and the cause of the anomaly, and distributing the adjustment configuration parameters to the edge device to adjust the real-time inversion network model, includes the following steps T4021 to T4024: T4021: Based on all the first real-time causal edges in the real-time logging self-causal graph, construct a target topological adjacency matrix that characterizes the current drilling area's geophysical causal response.

[0080] For example, taking a real-time well logging self-causal graph containing two first real-time causal nodes—a natural gamma node and a sonic transit time node—as an example, assume there exists a first real-time causal edge in this graph pointing from the natural gamma node to the sonic transit time node. Based on the graph topology, the cloud server digitizes the corresponding two nodes in a preset fixed order (the first dimension is natural gamma, and the second dimension is sonic transit time), thereby constructing a two-row, two-column target topological adjacency matrix. In this matrix, the first row and first column have values ​​of 0, the first row and second column have values ​​of 1 (representing an explicit physical causal flow from the natural gamma node to the sonic transit time node), and the second row and first column have values ​​of 0. This target topological adjacency matrix, in the form of a purely digital discrete graph skeleton, accurately locks down the real physical mechanism constraints exhibited by the new formation in nature.

[0081] T4022: Combine the target topology adjacency matrix with the preset weight coefficients as the adjustment configuration parameters, and send the adjustment configuration parameters to the edge device.

[0082] For example, the cloud server retrieves a preset weighting coefficient within the system to control the strength of physical constraints, packages the aforementioned two-row, two-column target topological adjacency matrix with the scalar value 0.05 together to form the adjustment configuration parameter specific to this work area, and pushes it precisely to the corresponding edge device via the communication network. The weighting coefficient is determined through hyperparameter optimization operations such as grid search or cross-validation on historical sample data, and is based on a balance between the convergence speed of the original loss term and the strength of the topological constraints, combined with on-site engineering experience. It is used to balance the convergence speed of the original inversion loss and the strong constraint to ensure the new geophysical causal relationship. This application does not limit the specific setting method of this weighting coefficient.

[0083] T4023: Instruct the edge device to calculate the matrix cross-entropy loss value between the intermediate feature association matrix generated by the real-time inversion network model during the inversion calculation process and the target topological adjacency matrix, so as to generate a causal topological regularization penalty term.

[0084] For example, when the edge device receives the adjusted configuration parameters from the cloud server, it does not interrupt the current inversion production flow. Instead, it instructs its internal real-time inversion network model to extract the dynamically generated two-row, two-column intermediate feature correlation matrix for the natural gamma feature channel and the acoustic time difference feature channel from its specific backbone feature extraction layer during the forward inference calculation. Assuming that the model has not yet adapted to the new strata, the numerical flow representing the relationship between the two feature channels in the intermediate feature correlation matrix is ​​only 0.65 at the first row, second column position. At this time, the edge computing unit uses the target topological adjacency matrix as the standard supervision label to calculate the matrix cross-entropy loss value between the intermediate feature correlation matrix and the target topological adjacency matrix online. Due to the deviation between 0.65 and the standard label 1, the calculated matrix cross-entropy loss value produces a non-zero scalar value of 0.35. The edge device directly declares this scalar value of 0.35 as the causal topological regularization penalty term.

[0085] T4024: The edge device is instructed to add the causal topology regularization penalty term to the original loss function of the real-time inversion network model through the weight coefficient to generate the target total loss function. Then, with the minimization of the target total loss function as the convergence objective, the real-time inversion network model is backpropagated to update the gradient, thereby completing the adjustment of the real-time inversion network model.

[0086] For example, when performing backpropagation gradient updates on the real-time inversion network model, the edge device synchronously and dynamically extracts the implicitly generated intermediate feature correlation matrix from the specific backbone feature extraction layer of the real-time inversion network model during the forward inference process of the inversion calculation. Next, the edge device calculates the matrix cross-entropy loss value between this intermediate feature correlation matrix and the issued target topological adjacency matrix, using this as a causal topological regularization penalty term to measure whether the internal feature flow of the neural network conforms to the physical laws of nature. Finally, the edge device multiplies the causal topological regularization penalty term by the weight coefficient and adds it linearly to the original inversion loss function of the real-time inversion network model, thereby reconstructing and generating the target total loss function. By using the minimization of this target total loss function as the convergence objective for backpropagation gradient optimization, the real-time inversion network model is forced to strictly adhere to the real geological and physical causal chain of the new work area when correcting its internal weight parameters. This achieves robust adaptive evolution of the model to a completely new mechanism work area without relying on a massive number of new well label samples.

[0087] Traditional incremental fine-tuning of neural networks is prone to overfitting or ill-conditioned convergence when faced with real-world changes in formation mechanisms due to a lack of physical prior constraints. However, the target topological adjacency matrix generated from the well logging self-causal graph, which characterizes the causal relationships between natural well logging data, provides a strong constraint on the real-world geological and physical causal relationships in the current work area. Therefore, this application utilizes the real-time well logging self-causal graph to construct the target topological adjacency matrix. By calculating the matrix cross-entropy loss value, a causal topological regularization penalty term is introduced into the original loss function for gradient updates. This forces the neural network to strictly adhere to the physical evolution logic of the new formation when optimizing weights, effectively ensuring the accuracy of the inversion network model adjustment at the edges.

[0088] Furthermore, in some embodiments of this application, if the geological mechanism matching anomaly is a false anomaly, the step of generating adjustment configuration parameters for the real-time inversion network model based on the causal conflict characteristics and the cause of the anomaly, and distributing the adjustment configuration parameters to the edge device to adjust the real-time inversion network model, includes the following steps T4031 to T4033: T4031: Select the target benchmark logging autocausality map with the smallest difference in the second causal edge from all the aforementioned benchmark logging autocausality maps.

[0089] T4032: The model parameters of the benchmark inversion network model corresponding to the benchmark causal graph corresponding to the target benchmark logging self-causal graph are used as the adjustment configuration parameters.

[0090] T4033: Send the adjusted configuration parameters to the edge device so that the edge device replaces the model parameters of the real-time inversion network model with the adjusted configuration parameters.

[0091] For example, when the cloud server determines that the geological mechanism matching anomaly is a false anomaly through a mechanism at the second-level subgraph level, it means that the causal logic of the pure rock physical response at the bottom layer of nature has not changed. Instead, the inversion distortion is caused by the mismatch between the initially issued model and the strata in the work area. Therefore, the cloud server first extracts the second causal edge difference degree corresponding to each benchmark well logging self-causal graph, and then selects the target benchmark well logging self-causal graph with the smallest second causal edge difference degree. Subsequently, the cloud server directly retrieves the full graph corresponding to the target benchmark well logging self-causal graph according to the preset graph model mapping index relationship, and uses the model parameters of the benchmark inversion network model generated by the graph in historical training as the adjustment configuration parameters. These parameters cover the set of weight matrix tensors and the set of bias vectors unique to each layer in the benchmark inversion network model. Finally, the cloud server sends the model parameter tensor of the adjustment configuration parameters to the edge device at the current drilling site, so that the edge device triggers a dynamic memory overwrite instruction locally to replace the model parameters of the currently running real-time inversion network model with the received adjustment configuration parameters.

[0092] Preferably, in some embodiments of this application, the real-time inversion network model of the edge device has the same topology as the baseline inversion network model of the cloud server. Therefore, the model parameter replacement process can be completed instantaneously without changing the underlying computational graph architecture. Through this fully automatic model parameter overlay, the real-time inversion network model at the edge is given mapping parameters that perfectly match the current actual strata, thereby completely eliminating the influence of false geological anomalies and restoring the inversion accuracy of reservoir parameters.

[0093] When the well logging self-causal graph highly matches a certain benchmark subgraph in the library, it fully demonstrates that the natural physical laws of the current new drilling area belong entirely to the known geological facies zone that has been explored. At this time, the model mismatch is a pure logical mounting error and no incremental model training is required. Therefore, when it is judged as a false anomaly, by comparing the well logging self-causal graph, the mature benchmark network model parameters with the smallest difference of the second causal edge are retrieved from the cloud library and the edge end is replaced as a whole. This achieves zero consumption of computing power and zero time delay self-healing at the edge device end, and restores the inversion accuracy to the design index within the shortest cloud-edge communication cycle.

[0094] In summary, the intelligent inversion method for oil and gas reservoir parameters based on cloud-edge fusion provided in this application has the following advantages over existing technologies: The impact mechanisms of different physical failures on graph topology and data status in oil and gas inversion systems differ fundamentally. Front-end sensor failures typically only cause systematic deviations in logging values, without altering the inherent causal response chain of the formation's rock physics; while natural geological mechanism anomalies fundamentally disrupt the originally stable causal logic of geophysical responses. This application integrates logging data and inverted reservoir parameters at the edge device to construct a real-time causal graph. Furthermore, by comparing the causal conflict characteristics of the real-time causal graph constructed at the edge with a preset baseline causal graph in the cloud, it obtains the statistical distribution divergence of the first causal edge difference degree and data distribution differences, thereby distinguishing different causes of inversion failure. This provides an objective basis for the cloud to generate and distribute matching adjustment configuration parameters for specific failure mechanisms, thus improving the matching degree of edge model adjustment parameters and avoiding the bandwidth and computing power waste caused by blind retraining in traditional technologies.

[0095] like Figure 3 As shown, based on the above-mentioned method embodiments, this application provides an intelligent inversion system for oil and gas reservoir parameters based on cloud-edge fusion, which is applied to a cloud server. The cloud server communicates with edge devices, which are set in the corresponding drilling area. The edge devices are configured with a real-time inversion network model for oil and gas reservoir parameter inversion. The intelligent inversion system for oil and gas reservoir parameters based on cloud-edge fusion includes: a receiving module 201, a feature extraction module 202, an anomaly cause determination module 203, and an adjustment parameter distribution module 204. The receiving module 201 is used to receive the real-time causal graph constructed by the edge device based on the real-time inversion network model; wherein, the real-time causal graph is constructed based on the real-time logging data of the drilling area collected by the edge device and the first oil and gas reservoir parameters obtained by the real-time inversion network model based on the real-time logging data. The feature extraction module 202 is used to extract causal conflict features between the real-time causal graph and each preset benchmark causal graph; the causal conflict features include: a first causal edge difference degree characterizing the causal difference between the real-time causal graph and the benchmark causal graph, and a statistical distribution divergence characterizing the data distribution difference between the real-time causal graph and the benchmark causal graph; The anomaly cause determination module 203 is used to determine the anomaly cause of the real-time inversion network model based on the causal conflict characteristics if the causal conflict characteristics meet the preset inversion anomaly determination conditions; wherein, each of the reference causal graphs corresponds to a detected drilling area and a reference inversion network model. The adjustment parameter distribution module 204 is used to generate adjustment configuration parameters for the real-time inversion network model based on the causal conflict characteristics and the causes of anomalies, and distribute the adjustment configuration parameters to the edge device to adjust the real-time inversion network model, and then perform oil and gas reservoir parameter inversion of the drilling area through the adjusted real-time inversion network model.

[0096] Further, in some embodiments of this application, the real-time causal graph includes a plurality of real-time causal nodes, real-time logging data or first oil and gas reservoir parameters corresponding to each real-time causal node, and a plurality of real-time causal edges; the baseline causal graph includes baseline causal nodes corresponding one-to-one with the real-time causal nodes, historical baseline data distributions corresponding to each baseline causal node, and a plurality of baseline causal edges; the feature extraction module 202 includes: a first calculation unit, a second calculation unit, and an association unit; the feature extraction module 202 is used to extract causal conflict features between the real-time causal graph and preset baseline causal graphs, including: The first calculation unit is configured to, for any of the aforementioned baseline causal graphs, calculate the first causal edge difference degree between all the corresponding baseline causal edges and all the real-time causal edges based on the structural Hamming distance; The second calculation unit is used to calculate, for any of the aforementioned reference causal graphs, the statistical distribution divergence between each of the real-time logging data and the historical reference data distribution corresponding to the reference causal node in the reference causal graph; The association unit is used to perform matrix association of the corresponding first causal edge difference degree and the statistical distribution divergence for any of the baseline causal graphs, and construct the causal conflict features between the real-time causal graph and the baseline causal graph.

[0097] In summary, the intelligent inversion system for oil and gas reservoir parameters based on cloud-edge fusion provided in this application has the following advantages over existing technologies: The impact mechanisms of different physical failures on graph topology and data status in oil and gas inversion systems differ fundamentally. Front-end sensor failures typically only cause systematic deviations in logging values, without altering the inherent causal response chain of the formation's rock physics; however, changes in natural geological mechanisms fundamentally disrupt the originally stable causal logic of geophysical responses. This application integrates logging data and inverted reservoir parameters at the edge device to construct a real-time causal graph. Furthermore, by comparing the real-time causal graph constructed at the edge with a preset baseline causal graph in the cloud using causal conflict characteristics, it obtains the statistical distribution divergence of the first causal edge difference and data distribution differences, thereby distinguishing different causes of inversion failure. This provides an objective basis for the cloud to generate and distribute matching adjustment configuration parameters for specific failure mechanisms, thus improving the matching degree of edge model adjustment parameters and avoiding the bandwidth and computing power waste caused by blind retraining in traditional technologies.

[0098] It is understood that the above system embodiments correspond to the method embodiments of this application, and can realize the intelligent inversion method for oil and gas reservoir parameters based on cloud-edge fusion provided by any of the above method embodiments of this application.

[0099] It should be noted that the system embodiments described above are merely illustrative, and some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0100] Based on the above embodiments of the intelligent inversion method for oil and gas reservoir parameters based on cloud-edge fusion, another embodiment of this application provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the intelligent inversion method for oil and gas reservoir parameters based on cloud-edge fusion of any embodiment of this application.

[0101] For example, in this embodiment, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete this application. The one or more module units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the terminal device.

[0102] The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0103] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0104] Based on the above-described method embodiments, another embodiment of this application provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is running, it controls the device where the computer-readable storage medium is located to execute the intelligent inversion method for oil and gas reservoir parameters based on cloud-edge fusion as described in any of the above-described method embodiments of this application.

[0105] The modules / units integrated in the device / terminal equipment, if implemented as software functional units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

Claims

1. A smart inversion method for oil and gas reservoir parameters based on cloud-edge fusion, characterized in that, The cloud-edge fusion-based intelligent inversion method for oil and gas reservoir parameters, applied to a cloud server and communicating with edge devices located in corresponding drilling areas, is configured with a real-time inversion network model for oil and gas reservoir parameter inversion. The method includes: The edge device receives a real-time causal graph constructed based on the real-time inversion network model; wherein the real-time causal graph is constructed based on the real-time logging data of the drilling area collected by the edge device and the first oil and gas reservoir parameters obtained by the real-time inversion network model based on the real-time logging data. Extract causal conflict features between the real-time causal graph and each preset benchmark causal graph; the causal conflict features include: a first causal edge difference degree characterizing the causal difference between the real-time causal graph and the benchmark causal graph, and a statistical distribution divergence characterizing the data distribution difference between the real-time causal graph and the benchmark causal graph; If the causal conflict features meet the preset inversion anomaly determination conditions, then the anomaly cause of the real-time inversion network model is determined according to the causal conflict features; wherein, each of the reference causal graphs corresponds to a detected drilling area and a reference inversion network model. Based on the causal conflict characteristics and the causes of anomalies, adjustment configuration parameters for the real-time inversion network model are generated. The adjustment configuration parameters are then sent to the edge device to adjust the real-time inversion network model. Subsequently, the oil and gas reservoir parameters of the drilling area are inverted using the adjusted real-time inversion network model.

2. The intelligent inversion method for oil and gas reservoir parameters based on cloud-edge fusion as described in claim 1, characterized in that, The real-time causal graph includes several real-time causal nodes, real-time logging data or first oil and gas reservoir parameters corresponding to each real-time causal node, and several real-time causal edges; the baseline causal graph includes baseline causal nodes that correspond one-to-one with the real-time causal nodes, the distribution of historical baseline data corresponding to each baseline causal node, and several baseline causal edges. The extraction of causal conflict features between the real-time causal graph and preset baseline causal graphs includes: For any of the aforementioned baseline causal graphs, the first causal edge difference degree between all the corresponding baseline causal edges and all the real-time causal edges is calculated based on the structural Hamming distance; For any of the aforementioned baseline causal graphs, calculate the statistical distribution divergence between each of the real-time logging data and the historical baseline data distribution corresponding to the baseline causal node in the baseline causal graph; For any of the aforementioned baseline causal graphs, the corresponding first causal edge difference degree and the statistical distribution divergence are matrix-correlated to construct the causal conflict features between the real-time causal graph and the baseline causal graph.

3. The intelligent inversion method for oil and gas reservoir parameters based on cloud-edge fusion as described in claim 2, characterized in that, The causes of the anomalies include: sensor malfunction and geological mechanism mismatch; determining the causes of the anomalies in the real-time inversion network model based on the causal conflict characteristics includes: If there exists a target baseline causal graph, the difference degree of the first causal edge corresponding to the target baseline causal graph is less than a preset first difference judgment threshold, and there exists at least one target baseline causal node in the target baseline causal graph whose statistical distribution divergence is greater than a preset distribution drift threshold, then the cause of the anomaly is determined to be the sensor malfunction. If the first causal edge difference degree corresponding to all the benchmark causal graphs is greater than or equal to the first difference determination threshold, then the anomaly is determined to be caused by the geological mechanism matching anomaly.

4. The intelligent inversion method for oil and gas reservoir parameters based on cloud-edge fusion as described in claim 3, characterized in that, The edge device is equipped with a plurality of sensors, each of which collects a corresponding real-time logging data. If the anomaly is caused by a sensor malfunction, the step of generating adjustment configuration parameters for the real-time inversion network model based on the causal conflict characteristics and the cause of the anomaly, and sending the adjustment configuration parameters to the edge device to adjust the real-time inversion network model, includes: Based on the correspondence between the target reference causal node, the real-time logging data and the sensor, the fault sensor corresponding to each target reference causal node is determined; Based on the statistical distribution divergence corresponding to each of the target real-time causal nodes, the data calibration mapping coefficients corresponding to the fault sensors are determined by the statistical distribution alignment method. The data calibration mapping coefficients are sent to the edge device as adjustment configuration parameters to instruct the edge device to calibrate the real-time logging data collected by the corresponding fault sensor in real time through the data calibration mapping coefficients, thereby completing the adjustment of the input data of the real-time inversion network model.

5. The intelligent inversion method for oil and gas reservoir parameters based on cloud-edge fusion as described in claim 3, characterized in that, If the anomaly is determined to be caused by an abnormal geological mechanism, before generating the adjusted configuration parameters of the real-time inversion network model based on the causal conflict characteristics and the cause of the anomaly, the process includes: Real-time well logging self-causal graphs are extracted from the real-time causal graphs, and reference well logging self-causal graphs are extracted from each of the reference causal graphs; wherein, the real-time well logging self-causal graph includes only the first real-time causal node corresponding to the real-time well logging data and the first real-time causal edge between the first real-time causal nodes; the reference well logging self-causal graph includes only the first reference causal node corresponding one-to-one with the first real-time causal node and the first reference causal edge between the first reference causal nodes. For any of the aforementioned benchmark logging self-causal graphs, based on the structural Hamming distance, the second causal edge difference degree between all the corresponding first benchmark causal edges and all the first real-time causal edges is calculated; If at least one target benchmark logging self-causal graph has a second causal edge difference degree corresponding to it that is less than a preset second difference determination threshold, then the geological mechanism matching anomaly is determined to be a false anomaly; otherwise, the geological mechanism matching anomaly is determined to be a true anomaly.

6. The intelligent inversion method for oil and gas reservoir parameters based on cloud-edge fusion as described in claim 5, characterized in that, If the geological mechanism matching anomaly is a true anomaly, the step of generating adjustment configuration parameters for the real-time inversion network model based on the causal conflict characteristics and the cause of the anomaly, and distributing the adjustment configuration parameters to the edge device to adjust the real-time inversion network model, includes: Based on all the first real-time causal edges in the real-time logging self-causal graph, construct a target topological adjacency matrix that characterizes the current geophysical causal response of the drilling area; The target topology adjacency matrix is ​​combined with preset weight coefficients to form the adjustment configuration parameters, and the adjustment configuration parameters are sent to the edge device. The edge device is instructed to calculate the matrix cross-entropy loss value between the intermediate feature correlation matrix generated by the real-time inversion network model during the inversion calculation process and the target topological adjacency matrix, so as to generate a causal topological regularization penalty term. The edge device is instructed to add the causal topology regularization penalty term to the original loss function of the real-time inversion network model through the weight coefficient to generate the target total loss function. Then, with the minimization of the target total loss function as the convergence objective, the real-time inversion network model is backpropagated to update the gradient, thereby completing the adjustment of the real-time inversion network model.

7. The intelligent inversion method for oil and gas reservoir parameters based on cloud-edge fusion as described in claim 5, characterized in that, If the geological mechanism matching anomaly is a false anomaly, the step of generating adjustment configuration parameters for the real-time inversion network model based on the causal conflict characteristics and the cause of the anomaly, and distributing the adjustment configuration parameters to the edge device to adjust the real-time inversion network model, includes: From all the aforementioned benchmark logging autocausal graphs, select the target benchmark logging autocausal graph with the smallest second causal edge difference; The model parameters of the benchmark inversion network model corresponding to the benchmark causal graph corresponding to the target benchmark logging self-causal graph are used as the adjustment configuration parameters; The adjusted configuration parameters are sent to the edge device so that the edge device replaces the model parameters of the real-time inversion network model with the adjusted configuration parameters.

8. The intelligent inversion method for oil and gas reservoir parameters based on cloud-edge fusion as described in any one of claims 1 to 7, characterized in that, Before receiving the real-time causal graph constructed by the edge device based on the real-time inversion network model, the process includes: Receive the first geographic coordinates corresponding to the drilling area sent by the edge device; Calculate the spatial distance between the first geographic coordinates and the second geographic coordinates of each of the detected drilling areas; The baseline inversion network model corresponding to the detected drilling area with the smallest spatial distance is sent to the edge device as the real-time inversion network model.

9. A smart inversion system for oil and gas reservoir parameters based on cloud-edge fusion, characterized in that, The cloud server communicates with edge devices, which are located in the corresponding drilling area. The edge devices are equipped with a real-time inversion network model for oil and gas reservoir parameter inversion. The cloud-edge fusion-based intelligent inversion system for oil and gas reservoir parameters includes: a receiving module, a feature extraction module, an anomaly cause determination module, and an adjustment parameter distribution module. The receiving module is used to receive the real-time causal graph constructed by the edge device based on the real-time inversion network model; wherein, the real-time causal graph is constructed based on the real-time logging data of the drilling area collected by the edge device and the first oil and gas reservoir parameters obtained by the real-time inversion network model based on the real-time logging data. The feature extraction module is used to extract causal conflict features between the real-time causal graph and each preset benchmark causal graph; the causal conflict features include: a first causal edge difference degree characterizing the causal difference between the real-time causal graph and the benchmark causal graph, and a statistical distribution divergence characterizing the data distribution difference between the real-time causal graph and the benchmark causal graph; The anomaly cause determination module is used to determine the anomaly cause of the real-time inversion network model based on the causal conflict characteristics if the causal conflict characteristics meet the preset inversion anomaly determination conditions; wherein, each of the reference causal graphs corresponds to a detected drilling area and a reference inversion network model. The adjustment parameter distribution module is used to generate adjustment configuration parameters for the real-time inversion network model based on the causal conflict characteristics and the causes of anomalies, and distribute the adjustment configuration parameters to the edge device to adjust the real-time inversion network model, and then use the adjusted real-time inversion network model to perform oil and gas reservoir parameter inversion in the drilling area.

10. The intelligent inversion system for oil and gas reservoir parameters based on cloud-edge fusion as described in claim 9, characterized in that, The real-time causal graph includes several real-time causal nodes, real-time logging data or first oil and gas reservoir parameters corresponding to each real-time causal node, and several real-time causal edges; the baseline causal graph includes baseline causal nodes that correspond one-to-one with the real-time causal nodes, the distribution of historical baseline data corresponding to each baseline causal node, and several baseline causal edges. The feature extraction module includes: a first calculation unit, a second calculation unit, and an association unit; the feature extraction module is used to extract causal conflict features between the real-time causal graph and preset benchmark causal graphs, including: The first calculation unit is configured to, for any of the aforementioned baseline causal graphs, calculate the first causal edge difference degree between all the corresponding baseline causal edges and all the real-time causal edges based on the structural Hamming distance; The second calculation unit is used to calculate, for any of the aforementioned reference causal graphs, the statistical distribution divergence between each of the real-time logging data and the historical reference data distribution corresponding to the reference causal node in the reference causal graph; The association unit is used to perform matrix association of the corresponding first causal edge difference degree and the statistical distribution divergence for any of the baseline causal graphs, and construct the causal conflict features between the real-time causal graph and the baseline causal graph.

Images