Physical-data coupled geologic state dynamic evaluation method and system suitable for deep water shallow formation

Abstract

The application provides a physical-data coupling geologic state dynamic evaluation method and system suitable for deep water shallow formation, and relates to the technical field of geological monitoring. Ocean dynamic data, engineering activity data and geological response data of a target geologic body are acquired in real time and standardized; a safety index field and physical field characteristics are output based on a physical mechanism analysis engine with standardized data as dynamic boundary conditions and source terms; a stability state probability distribution and a disaster risk scalar value are output based on a data-driven analysis engine according to real-time monitoring data sequences and the physical field characteristics; a stability index and a risk probability are obtained by dynamically weighting and fusing the safety index field, the stability state probability distribution and the disaster risk scalar value, and corresponding warning information is triggered according to preset multi-level threshold values and logic rules. The method realizes deep complementation of mechanism and data: the physical model provides an interpretable mechanism framework and generalization ability, the data model provides rapid identification and real-time adaptation ability for complex patterns, and both realize efficient evaluation and monitoring and early warning of formation state in the process of deep water oil and gas resource development through bidirectional coupling.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of geological monitoring technology, and in particular to a physical-data coupled dynamic assessment method and system for geological state of shallow strata in deep water. Background Technology

[0002] With the advancement of deep-sea energy development, development activities in the deep-water shallow energy-rich areas of the South China Sea are becoming increasingly frequent. For example, a certain ultra-deep-water ultra-shallow gas field and a natural gas hydrate deposit are located in the same area. The geological environment of this area is extremely complex. Under the combined effects of strong internal wave / tidal shear (natural dynamics) and resource extraction disturbance (anthropogenic dynamics), the shallow, unconsolidated strata are highly susceptible to geological disasters such as submarine landslides and shallow instability, seriously threatening operational safety.

[0003] Therefore, there is an urgent need for a dynamic assessment method for the geological state of shallow strata in deep water. Summary of the Invention

[0004] This invention provides a physical-data coupled dynamic assessment method and system for geological states in deep-water shallow strata. It can realize intelligent perception, dynamic assessment and early warning of geological stability in deep-water shallow energy-rich areas (including conventional oil and gas and natural gas hydrates) under the coupled effects of marine dynamics and engineering activities, so as to at least partially solve the above problems.

[0005] The first aspect of this invention provides a physical-data coupled dynamic assessment method for geological states of shallow deep-water strata, the method comprising: Real-time acquisition and standardized processing of marine dynamic data, engineering activity data, and geological response data of the target geological body; Based on a physical mechanism analysis engine, a multi-physics coupled numerical model is used to calculate the evolution of stress field, seepage field, and deformation field of the target geological body using standardized data as dynamic boundary conditions and source terms, and outputs safety index field and physical field characteristics. The physical information neural network model based on the data-driven analysis engine outputs the stability state probability distribution and disaster risk scalar value of geological stability according to the real-time monitoring data sequence and the physical field characteristics. The real-time monitoring data sequence is obtained based on the marine dynamic data, engineering activity data and geological response data. The safety index field, the stability state probability distribution, and the disaster risk scalar value rate are dynamically weighted and fused to obtain the stability index and the risk probability. Based on the stability index and risk probability, corresponding early warning information is triggered according to preset multi-level thresholds and logical rules. The multiphysics coupled numerical model and the physical information neural network model adopt a ensemble Kalman filter framework to achieve bidirectional dynamic coupling and correction, including: using the high confidence information output by the data-driven analysis engine and some direct observation data as observation vectors to update the model parameters and state of the physical mechanism analysis engine in reverse, and injecting the physical field features output by the physical mechanism analysis engine into the data-driven analysis engine in a positive direction.

[0006] A second aspect of this invention provides a physical-data coupled dynamic assessment system for geological states of shallow deep-water strata, the system comprising: The data processing module is used to acquire and standardize marine dynamic data, engineering activity data, and geological response data of the target geological body in real time; The physical mechanism analysis engine is used to calculate the evolution of stress field, seepage field, and deformation field of target geological bodies based on multi-physics coupled numerical models, using standardized data as dynamic boundary conditions and source terms, and output safety index field and physical field characteristics. A data-driven analysis engine is used to output the stability state probability distribution and disaster risk scalar value of geological stability based on a physical information neural network model, according to real-time monitoring data sequences and the physical field characteristics. The real-time monitoring data sequences are obtained based on the marine dynamic data, engineering activity data and geological response data. The fusion module is used to dynamically weight and fuse the safety index field, stability state probability distribution and disaster risk scalar value to obtain the stability index and risk probability. The early warning module is used to trigger corresponding early warning information based on the stability index and risk probability, according to preset multi-level thresholds and logical rules. The multiphysics coupled numerical model and the physical information neural network model adopt a ensemble Kalman filter framework to achieve bidirectional dynamic coupling and correction, including: using the high confidence information output by the data-driven analysis engine and some direct observation data as observation vectors to update the model parameters and state of the physical mechanism analysis engine in reverse, and injecting the physical field features output by the physical mechanism analysis engine into the data-driven analysis engine in a positive direction.

[0007] A third aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executed, implements the physical-data coupled dynamic assessment method for geological states applicable to deep-water shallow strata as described in the first aspect of the present invention.

[0008] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the physical-data coupled dynamic assessment method for geological states applicable to deep-water shallow strata as described in the first aspect of the present invention.

[0009] A fifth aspect of the present invention provides a computer program product, including a computer program / instructions, which are implemented by a processor as the steps in the physical-data coupled dynamic assessment method for geological states of deep-water shallow strata as described in the first aspect of the present invention.

[0010] This invention provides a system capable of real-time, dynamic, and accurate assessment of geological stability in both deep and shallow water. Specifically, through deep coupling and bidirectional feedback between a physical mechanism model and a data-driven model, an intelligent system with "perception-cognition-early warning" capabilities is formed, enabling advanced early warning of geological disasters under complex dynamic processes.

[0011] The physical-data coupled dynamic assessment method for geological state of deep-water shallow strata provided by the embodiments of the present invention achieves deep complementarity between mechanism and data: the physical model provides an interpretable mechanism framework and generalization ability, while the data model provides rapid identification and real-time adaptation to complex patterns. The two overcome their respective defects through bidirectional coupling.

[0012] It possesses dynamic self-learning and self-correction capabilities: through EnKF data assimilation, the system can continuously optimize the key parameters of the physical model using real-time observations, making it increasingly approximate the real geological environment, and significantly improving the accuracy of long-term assessments.

[0013] It achieves true early warning: the system not only monitors the current state, but also predicts the evolution trend in the short term by coupling simulation and machine learning. It can identify early warning signals of disasters and win valuable time for emergency response.

[0014] This invention forms a complete technological closed loop: from data perception and coupled analysis to intelligent decision-making, it provides an integrated solution applicable to the assessment of common geological safety risks faced by various energy development projects in deep and shallow waters. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1This is a flowchart of the steps in the physical-data coupled dynamic assessment method for the geological state of deep-water shallow strata provided in this embodiment of the invention; Figure 2 This is the overall architecture and data flow diagram of the physical-data coupled dynamic geological state assessment method for deep-water shallow strata provided in this embodiment of the invention; Figure 3 This is a schematic diagram of the multi-field coupled numerical model of the physical mechanism analysis engine in the physical-data coupled geological state dynamic assessment method for deep-water shallow strata provided in this embodiment of the invention. Figure 4 This is a diagram of the physical information neural network structure of the data-driven analysis engine in the physical-data coupled geological state dynamic assessment method for deep-water shallow strata provided in this embodiment of the invention. Figure 5 This is a flowchart of the dual-engine data assimilation process based on ensemble Kalman filtering in the physical-data coupled geological state dynamic assessment method for deep-water shallow strata provided in this embodiment of the invention. Figure 6 This is a logic diagram for dynamic fusion assessment and multi-level early warning generation in the physical-data coupled geological state dynamic assessment method for deep-water shallow strata provided in this embodiment of the invention. Detailed Implementation

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] The inventors have discovered that geological condition assessment and monitoring technologies have the following limitations: Mechanism model disconnect: Traditional numerical simulation (physical mechanism model) relies on precise initial parameters and boundary conditions, making it difficult to assimilate observation data in real time. Long-term operation will produce drift errors, and the calculation is time-consuming, making it difficult to meet real-time requirements.

[0019] Data models lack mechanisms: Although pure data-driven machine learning models can process data quickly, their predictions are like "black boxes," with poor physical interpretability, weak extrapolation ability under new working conditions they have not experienced, and questionable reliability.

[0020] Separation of monitoring and early warning: Currently, most monitoring systems mainly rely on data acquisition and simple threshold alarms. They lack the ability to deeply integrate, couple, and intelligently analyze multi-source heterogeneous data (marine dynamics, engineering activities, geological responses) under a unified mechanism framework, thus failing to achieve true process early warning and advanced early warning.

[0021] Therefore, the inventors propose an integrated technical solution that can combine physical mechanisms with real-time data to achieve dynamic self-correction and intelligent judgment.

[0022] Specifically, embodiments of the present invention provide a physical-data coupled dynamic assessment method for geological states of shallow deep-water strata, such as... Figure 1 As shown, the method includes: S101 acquires and standardizes marine dynamic data, engineering activity data, and geological response data of the target geological body in real time.

[0023] S102 is a multi-physics coupled numerical model based on a physical mechanism analysis engine. It uses standardized data as dynamic boundary conditions and source terms to calculate the evolution of the stress field, seepage field, and deformation field of the target geological body, and outputs the safety index field and physical field characteristics.

[0024] S103, a physical information neural network model based on a data-driven analysis engine, outputs the stability state probability distribution and disaster risk scalar value of geological stability according to the real-time monitoring data sequence and the physical field characteristics. The real-time monitoring data sequence is obtained based on the marine dynamic data, engineering activity data and geological response data.

[0025] S104, dynamically weight and fuse the safety index field, stability state probability distribution and disaster risk scalar value to obtain the stability index and risk probability.

[0026] S105, based on the stability index and risk probability, trigger corresponding early warning information according to preset multi-level thresholds and logical rules.

[0027] In this embodiment of the invention, a system capable of real-time, dynamic, and accurate assessment of geological stability in deep and shallow water is provided. Specifically, through deep coupling and bidirectional feedback between a physical mechanism model (multi-physics coupled numerical model) and a data-driven model (physical information neural network model), an intelligent system with "perception-cognition-early warning" capabilities is formed, enabling advanced early warning of geological disasters under complex dynamic effects.

[0028] like Figure 2 As shown, it illustrates the overall architecture and data flow diagram of an example of the present invention. Its architecture mainly includes the following three layers: the field perception layer, the edge computing layer, and the cloud data center layer.

[0029] Among them, the field perception layer is the data source, responsible for the collection of multi-source heterogeneous data, and is the input foundation of the entire system. It is divided into three types of monitoring units: marine dynamic monitoring unit, engineering activity monitoring unit, and geological response monitoring unit.

[0030] Specifically, in this embodiment of the invention, a field perception layer can be constructed based on marine dynamic monitoring units (such as ADCP, dynamic pressure sensor), engineering activity monitoring units (obtaining drilling and production parameters through OPC UA protocol), and geological response monitoring units (such as pore pressure probe, distributed fiber optic sensing system, seabed reference instrument) to obtain marine dynamic data, engineering activity data and geological response data of the target geological body in real time.

[0031] Specifically, the marine dynamic monitoring unit involves deploying a seabed-based observation array in the target area. Each base station is equipped with a high-precision acoustic Doppler current profiler (ADCP, such as the Nortek Aquadopp series) and a dynamic pressure sensor to collect three-dimensional current velocity and pressure data of the bottom seawater at a frequency of no less than 1 Hz, which is used to invert the marine shear stress acting on the seabed. .

[0032] The engineering activity monitoring unit specifically acquires engineering parameters of the drilling platform or production facilities in real time through industrial data interfaces (such as the OPC UA protocol), including but not limited to: drilling fluid equivalent density (ECD), riser pressure, gas production rate, water injection pressure, etc.

[0033] The geological response monitoring unit specifically comprises: deploying a seabed observation network, including: a) pore water pressure probes (such as Geokon vibrating wire sensors), inserted at different depths below the seabed; b) a distributed fiber optic sensing system, laying sensing optical cables on the seabed surface or in shallow boreholes, and using φ-OTDR (phase-sensitive optical time-domain reflectometry) or OFDR (optical frequency-domain reflectometry) technology to demodulate and obtain the strain along the optical cable. (a) Temperature field; (b) Seabed reference instrument, providing long-term, continuous millimeter-level displacement data.

[0034] like Figure 2 As shown, in this embodiment of the invention, real-time data preprocessing and compression can be achieved based on an edge computing layer. Low-latency data cleaning and characterization are performed at the field end, avoiding direct transmission of massive amounts of data to the cloud. This can be achieved by deploying an underwater main junction box or a shore-based front-end server. An embedded system with sufficient computing power (such as one based on an NVIDIA Jetson module) runs lightweight data preprocessing and fusion algorithms to filter, time-align, and compress the raw data. The processed structured data stream is then uploaded to the data center via submarine optical cable or satellite communication. The core process includes: Data acquisition: Directly connects to three types of monitoring units on site.

[0035] Data standardization: unifying the format, dimensions, and scale of multi-source heterogeneous data.

[0036] Outlier detection: The Isolation Forest algorithm is used to identify dirty data / outliers to ensure data quality.

[0037] Feature extraction + data compression: Extract key features from raw time series / sensor data and compress them to reduce transmission / computation overhead.

[0038] like Figure 2 As shown, in this embodiment of the invention, the cloud data center layer achieves accurate modeling and feedback through the coupling of a "physical mechanism engine + data-driven engine".

[0039] In this embodiment of the invention, a high-performance computing cluster and an artificial intelligence server can be deployed during the specific implementation process. A physical mechanism analysis engine and a data-driven analysis engine are run to perform large-scale numerical simulations and deep learning model inferences, and finally, a fusion evaluation and early warning decision are made.

[0040] Among them, the Physical Mechanism Analysis Engine (P-Engine) is based on a multiphysics coupled numerical model and governing equations, starting from physical laws to simulate the evolution mechanism of the ocean / engineering / geology.

[0041] The Data-Driven Analytics Engine (D-Engine) employs a hybrid 1D-CNN + LSTM model for its physical information neural network. It learns complex nonlinear relationships based on historical / real-time data and outputs... Risk-related indicators enable rapid, data-driven predictions.

[0042] In the bidirectional coupling and assimilation process: P-Engine provides physical constraints, D-Engine supplements data fitting capabilities, and the two work together to correct the model; at the same time, they receive feedback from the cloud evaluation results to realize model iteration.

[0043] In this embodiment of the invention, the cloud data center layer undertakes massive data storage, global computation, model training and collaboration, and outputs preprocessed data streams downstream, receives compressed feature data from the edge layer, and stores multi-source heterogeneous data; it supports global training, parameter optimization and high-precision simulation computation of P-Engine and D-Engine; and it outputs preprocessed data streams to the lower layer (dual-engine module) to ensure the continuity of engine computation.

[0044] Specifically, in this embodiment of the invention, the physical mechanism analysis engine is based on a multi-physics coupled numerical model, employing a modified Cambridge model to describe the elastoplastic behavior of soil, and combining Biot's consolidation theory to achieve full coupling of the stress field and the seepage field. The engine performs discretization solving using finite element methods (such as COMSOL or FEniCS), dynamically loading marine shear stress and engineering pressure as boundary conditions and source terms, and outputting full-field physical quantities including effective stress, excess pore water pressure, plastic strain, and element safety index.

[0045] Specifically, the multiphysics coupled numerical model can simulate physical quantities such as effective stress, excess pore water pressure, plastic strain, and safety index under the coupled action of stress field and seepage field.

[0046] Specifically, the physics mechanism analysis engine simulates the mechanical behavior of geological systems by solving a system of partial differential equations coupled with multiphysics fields. The specific steps are detailed in the appendix. Figure 3 As shown, the specific explanation is as follows: 1. Mathematical model establishment: Governing equations: 1) Momentum conservation equation (equilibrium equation): (1) in, For the total stress tensor, The bulk density of the soil. This is the acceleration due to gravity.

[0047] 2) Mass conservation equation (seepage equation): (2) in, For fluid density, Porosity The seepage velocity (following Darcy's law) , For the permeability tensor, For fluid viscosity, (Porosity pressure) Source and sink items (representing engineering extraction or injection).

[0048] Constitutive relation: Soil skeleton: The modified Cambridge model is used to describe the elastoplastic behavior of shallow cohesive soils. Its yield surface function is: (3) in, It is a deviatoric stress. For the average effective stress, The slope of the critical state line. The initial consolidation pressure is the strain of the plastic body. The function. Model parameters (such as...) , κ, M The calibration was performed using undisturbed soil samples from the target area through indoor triaxial shear tests. The compressibility index represents the slope of the logarithmic relationship between the volume change of soil and the increase in effective stress under normal consolidation (initial loading). The larger the value, the easier the soil is to compress. κThe rebound index represents the slope of the logarithmic relationship between the change in elastic volume and the change in effective stress of soil in an overconsolidated state (unloading or reloading). κ The value is much smaller than .

[0049] Coupling Relationship: Complete coupling between the stress field and the seepage field is achieved using Biot's consolidation theory. Effective Stress Principle ( (where I is the Biot exponent and I is the unit tensor) is the bridge of coupling.

[0050] 2. Numerical solution implementation: 1) Software and Discretization: The "Porous Media Elastic-Plastic" and "Darcy Flow" modules of the commercial finite element software COMSOL Multiphysics are used, or self-written code is based on the FEniCS platform. The computational domain is established as a three-dimensional geometric model based on seismic interpretation and borehole data, and tetrahedral elements are used for meshing. Local refinement is carried out around potential slip zones and wellbore (the element size can be refined to 0.5 meters).

[0051] 2) Boundary and load dynamic application: The pre-processed shear stress obtained by the field sensing layer It is dynamically applied to the grid nodes on the seabed surface in the form of surface force.

[0052] Project pressure As a source and sink item It is applied to the model element at the location of the wellbore.

[0053] 3) Solution and Output: A fully coupled implicit solver is used for transient analysis with a fixed time step of 1 hour. The full-field results are output at each time step, with the core output variables including the effective stress tensor. excess pore water pressure Plastic strain tensor Based on the Mohr-Coulomb strength criterion, the local safety index of each element is calculated. FoS_local Ultimately, the safety index field and physical field characteristics are obtained, where the physical field characteristics can specifically be the plastic shear strain field.

[0054] In this embodiment of the invention, the data-driven analysis engine is constructed based on a Physical Information Neural Network (PINN). The input consists of time series data from multiple sources and feature quantities output by the physical engine (such as average effective stress and maximum plastic shear strain). The network includes a spatiotemporal feature extraction branch (convolutional layer + bidirectional LSTM) and a physical attention mechanism, and outputs the probability distribution of stability states (stable, at-risk, unstable) and disaster risk probability values ​​for the next 3 hours.

[0055] Specifically, in this embodiment of the invention, the physical information neural network model includes: an input layer, a feature extraction branch, a physical attention mechanism unit, and a fusion output layer.

[0056] Specifically, step S103 includes: S1031, receives real-time monitoring data sequences with a fixed time window based on the input layer; S1032, based on the feature extraction branch, the convolutional layer extracts local spatiotemporal features, and based on the long short-term memory network layer, it captures long-term dependencies to obtain spatiotemporal feature vectors; S1033, the physical attention mechanism unit receives the physical field features and generates a physical attention weight vector; the physical field features include: the unit's average effective stress and maximum plastic shear strain.

[0057] S1034, based on the modulation of the spatiotemporal feature vector and the physical attention weight vector by the fusion output layer, output the stability state probability distribution and disaster risk scalar value for future time periods.

[0058] Specifically, the data-driven analysis engine, based on physical information neural networks, learns directly from data and predicts system states. The specific steps are as follows: Figure 4 As shown, In this embodiment of the invention, the network model architecture design of the physical information neural network model includes: 1) Input layer: Receives multidimensional time series data within a fixed time window. . It is engineering stress. The pore water pressure varies over time as measured by a pore water probe. It is a distributed optical fiber strain measured by optical fiber sensing technology, and it is a distributed field variable that varies with time t and spatial location.

[0059] In this embodiment of the invention, the multidimensional time series data with a fixed time window is obtained by the data-driven analysis engine (D-Engine) after processing real-time marine dynamic data, engineering activity data, and geological response data.

[0060] In this embodiment of the invention, the input layer receives a multi-source real-time monitoring data window, including real-time data input from the D-Engine and unit average effective stress from the physics mechanism analysis engine (P-Engine). Maximum plastic shear strain .

[0061] Specifically, the element-average effective stress and maximum plastic shear strain are statistical scalar characteristics obtained by post-processing the full-field tensor results output by the rational mechanism analysis engine.

[0062] The following describes the average effective stress of the element. The steps for determining this are explained below: Original Output: The physics engine outputs the effective stress tensor σ′ for each spatial cell (mesh). This is a second-order tensor containing multiple components (such as...). σxx′,σyy′,σzz′,τxyσxx′,σyy′,σzz′,τxy (etc.), which fully describes the stress state at that point in 9 components (or 6 independent components).

[0063] Post-processing calculations: Calculate the average effective stress for each element: For each element, calculate the average effective stress (i.e., the first stress invariant) based on the three normal stress components of its effective stress tensor. This step condenses the complex tensor information of each element into a single scalar value. p ′ represents the average compressive stress at that point.

[0064] Overall statistics: At the current time step, for all units... p The average effective stress level of the entire region of interest is obtained by weighting the values ​​(weighted by element volume). .

[0065] Maximum plastic shear strain The calculation method is similar and will not be repeated here.

[0066] 2) Feature extraction branch: Spatiotemporal feature branch: Two one-dimensional convolutional layers (kernel size 3, stride 1) are used to extract local spatiotemporal features, capturing short-term patterns in the input time-series data. This is followed by a bidirectional LSTM (Long Short-Term Memory) layer to capture long-term dependencies and output the time-series features.

[0067] 3) Physical Attention Mechanism Unit: At the same time step, the physical field characteristics output by the P-Engine (such as the unit's average effective stress at that moment) are... Maximum plastic shear strain As additional input, a physical attention weight vector is generated through a fully connected layer. .

[0068] 4) Fusion Output Layer: Combines spatiotemporal feature vectors with... Element-wise multiplication modulation is performed, and the output is then passed through two fully connected layers, allowing physical information to guide and constrain data-driven feature learning. The final output is the probability distribution of stable states over the next three time steps (3 hours). and a disaster risk scalar value .

[0069] In this embodiment of the invention, the comprehensive prediction results for the next three time steps are output at once. A distribution and a risk value are output within a 3-hour window.

[0070] It is a three-dimensional vector that represents the model's judgment on the overall trend over the next 3 hours. It describes the stability tendency of the system over a window of time (rather than a single moment).

[0071] It is an independent numerical value representing the probability of disaster risk, with a value range of [0,1]. It quantifies the risk of the most unfavorable scenario that may occur within the next 3-hour window.

[0072] In this embodiment of the invention, the powerful feature extraction capabilities of deep learning are utilized to process massive, high-dimensional monitoring data. Simultaneously, by introducing features calculated from physical mechanisms and fusing them at the feature level, it is ensured that the model's prediction results are not only based on data patterns but also conform to physical laws, thereby improving the reliability and interpretability of the predictions and providing strong support for safety decisions on offshore platforms.

[0073] In this embodiment of the invention, the stability state probability distribution includes: stable state probability, attention state probability, and unstable state probability, and the sum of the stable state probability, attention state probability, and unstable state probability is 1.

[0074] Among them, the stable state indicates that the target geological body is within a safe and stable range and the monitoring parameters show no abnormal trend; the state of concern indicates that the target geological body has entered a critical or warning state; and the unstable state indicates that the target geological body has reached or is close to an unstable and destructive state.

[0075] Specifically, The probability of a stable state indicates that the geological system is within a safe and stable range, and the monitored parameters show no abnormal trends. The safety index calculated by the physical model is high, indicating no significant risk.

[0076] To monitor the probability of a (critical) state, the system enters a critical or warning state. This is indicated by one or more key indicators (such as pore pressure). ,strain When an abnormal trend emerges or the threshold of concern is reached, but no immediate threat of destruction has yet been formed, this is the incubation stage of a disaster.

[0077] This represents the probability of instability (failure), indicating that the system has reached or is close to an unstable and destructive state. Monitoring data shows accelerated deformation or abrupt changes, and the physical model shows that the safety index is close to or below 1.0, the plastic zone is connected, and the risk of disasters such as landslides and wellbore instability is extremely high.

[0078] These three probability values ​​are not independent; they constitute a complete probability distribution, satisfying... The system calculates these three probability values ​​by analyzing real-time data streams, thereby quantifying the uncertainty of the geological environment.

[0079] Its working logic and importance are reflected in: A: Quantifying risk beyond simple thresholds: Traditional "yes / no" binary judgments cannot reflect the continuous changes in risk. It dynamically and continuously describes the complete risk evolution path of the system from "stable state" to "state of concern" and then to "instability state" in a probabilistic form, and can capture the risk accumulation process earlier and more flexibly.

[0080] B: Driving Intelligent Early Warning Decision-Making: In the dynamic fusion assessment and early warning generation steps, (Instability probability) is one of the direct inputs for calculating the fusion stability index and triggering graded early warnings. For example, A rapid rise in these indicators (even if other indicators have not yet fully deteriorated) can directly trigger a higher level of warning.

[0081] C: Providing decision-making basis: Different probability distributions correspond to different response strategies: [High, Low, Low]: Routine operations.

[0082] [Medium, High, Low]: Warning status, the cause needs to be investigated (e.g., is it due to the passage of a strong internal wave?).

[0083] [Low, Medium, High]: Alert status, emergency measures need to be prepared or implemented.

[0084] In this embodiment of the invention, the training steps of the physical information neural network model include: S1, collect historical data for one or more complete annual cycles for the target geological body; S2, Based on the historical data and the multi-physics coupling numerical model, obtain the security index and physical field characteristics corresponding to the historical data; S3, determine the training label corresponding to the historical data based on the security index and physical field characteristics corresponding to the historical data; S4, based on cross-entropy loss, physical constraint loss and regularization loss, train the physical information neural network model with historical data carrying the training labels.

[0085] In this embodiment of the invention, the historical data of one or more complete annual cycles of the target geological body need to cover the ocean dynamic changes and engineering activity stages in different seasons. The historical data can be divided into training set, validation set and test set in an 8:1:1 ratio.

[0086] In this embodiment of the invention, the total loss function is defined as: (4) Cross-entropy loss measures classification accuracy.

[0087] Physical constraint loss, for example, requires that the network's predicted pore pressure change have the smallest residual difference from the trend calculated by Darcy's law.

[0088] Regularization loss helps prevent overfitting.

[0089] In this embodiment of the invention, the training process can use the Adam optimizer, with an initial learning rate set to 1e-4, and the learning rate decayed when the validation set loss no longer decreases. The batch size is set to 32.

[0090] In this embodiment of the invention, the model trained using historical data of the target geological body is suitable for the dynamic evaluation of the geological body from which the data originates. It should be noted that the evaluation method provided in this embodiment of the invention can be applied to any geological body, requiring only transfer learning based on the historical data of the corresponding geological body.

[0091] In this embodiment of the invention, the labels of historical data are automatically generated by the simulation calculation of historical data through the physical mechanism analysis engine (P-Engine). This can be understood as using the judgment result of the multi-physics coupled numerical model as the learning target of the physical information neural network model.

[0092] Specifically, in this embodiment of the invention, the output of the data-driven analysis engine (D-Engine) is a probability distribution of stable states. Therefore, during training, the labels given to the model are also three-dimensional vectors in this form (e.g., [0.1, 0.2, 0.7]).

[0093] In this embodiment of the invention, considering that three-dimensional vector labels cannot be directly seen by the naked eye, three-dimensional vector labels are generated through backtracking calculations by the Physical Mechanism Analysis Engine (P-Engine).

[0094] Specifically, the actual monitoring data (marine dynamics) of a certain past year Engineering pressure (etc.) are input into the physics analysis engine (P-Engine).

[0095] The physics mechanism analysis engine simulates the physical responses of the geological body under these real loads (such as stress changes and the development of plastic zones). Furthermore, based on the local safety index and plastic zone size output by P-Engine, a rule is set to generate labels. For example: If the minimum safety index If the value is greater than 1.5 and there is no plastic region, then the label is [1, 0, 0] (stable).

[0096] If 1.0 < If the value is ≤1.5 or there are sporadic plastic areas, the label is [0, 1, 0] (attention).

[0097] if If the value is ≤1.0 or the plastic region is fully penetrated, the label is [0, 0, 1] (instability).

[0098] J further constructs training pairs based on labels and historical data: Feature (X): Multi-source monitoring data (marine dynamics, engineering disturbances) within the same time period.

[0099] Label (Y): The 3D vector label generated in the above steps .

[0100] Thus, the data-driven analysis engine (D-Engine) can learn the correspondence between monitoring data fluctuations (X) and the stability state (Y) determined by the multiphysics coupled numerical model.

[0101] In this embodiment of the invention, the multiphysics coupled numerical model and the physical information neural network model adopt an ensemble Kalman filter framework to achieve bidirectional dynamic coupling and correction, including: using the high confidence information output by the data-driven analysis engine and some direct observation data as observation vectors to update the model parameters and state of the physical mechanism analysis engine in reverse, and injecting the physical field features output by the physical mechanism analysis engine into the data-driven analysis engine in a positive direction.

[0102] In this embodiment of the invention, ensemble Kalman filtering (EnKF) is used to achieve bidirectional coupling. The high-confidence output of the physical information neural network model and some directly observed data are used as observation vectors to update key uncertain parameters (such as elastic modulus and permeability) in the multiphysics coupled numerical model. Simultaneously, the physical field features output by the multiphysics coupled numerical model are injected into the physical information neural network model to enhance its physical interpretability.

[0103] In this embodiment of the invention, the assimilation period can be set to 6 hours to achieve dynamic self-correction of the model.

[0104] Specifically, the physical mechanism analysis engine runs a 6-hour simulation based on the current target geological body's marine dynamic data, engineering activity data, and geological response data to predict the geological body's state during this period (such as effective stress, excess pore water pressure, and plastic strain).

[0105] The above prediction results are compared with the high-confidence results output by the data-driven analysis engine and the actual sensor data during the same period to calculate the deviation between the two.

[0106] Using the ensemble Kalman filter algorithm, we analyze which key parameters (such as elastic modulus and permeability) are inaccurate due to this deviation, calculate the correction amount for these parameters, and then update these parameters in the physical model in reverse.

[0107] The deep characteristic quantities calculated by the physical model (such as the average effective stress of the element and the maximum plastic shear strain) are used as inputs to provide the data-driven analysis engine for reference in its subsequent predictions.

[0108] Repeat the above steps every 6 hours to achieve continuous bidirectional correction.

[0109] In a specific embodiment of the present invention, an Ensemble Kalman Filter (EnKF) framework is used to achieve bidirectional coupling, and the specific steps are as follows: Figure 5 As shown.

[0110] 1. Assimilation period setting: For example, the assimilation period can be set to 6 hours (that is, the physical model is updated with new observation data every 6 hours).

[0111] 2. Definitions of state vector and observation vector: State vector X: Select key uncertainty parameters in the spatial distribution of the multiphysics coupled numerical model, such as the elastic modulus E and permeability k of each soil layer, and arrange their discrete grid values ​​into a one-dimensional vector.

[0112] Observation vector Y: Select high-confidence output from D-Engine and some direct observations, such as pore pressure at a specific location. Displacement obtained by fiber strain inversion .

[0113] 3. EnKF assimilation step (executed cyclically): 1) Forecast step: from the analysis set of the previous assimilation time. , Starting from there, each member operates independently. Hours, obtain the forecast set and corresponding forecast observations .

[0114] Specifically, predictive observation refers to using a model to predict conditions and then simulating what data the sensors would see in the real world. It is a virtual sensor reading, created to be compared with real sensor readings.

[0115] In simple terms, the internal state of the physical model cannot be determined (such as the elastic modulus of each grid and the pore pressure field). However, the actual installed sensors (such as pore pressure probes) can measure values ​​at certain points. Predictive observations act as a "translator," converting the unseen field states within the model into the theoretical values ​​that should be displayed at the sensor locations.

[0116] For example: The physical field characteristics of the multiphysics coupled numerical model are assumed to include pore pressure values ​​at 100,000 grid points across the entire formation. In reality, only three pore pressure probes were installed at three depths, yielding three measured values.

[0117] The forecast observation extracts the pore pressure simulation values ​​of the grid points corresponding to the three probe positions from the 100,000 grid points of the model, forming a 3D vector.

[0118] In this way, the three measured values ​​and three simulated values ​​can be compared together.

[0119] 2) Analysis step: Obtain the actual observation value at the current moment. and its error covariance matrix Calculate the mean of the forecast set. Covariance For each member, calculate the Kalman gain. and update the status: This step "distributes" the observation information to each parameter, achieving reverse correction.

[0120] In this embodiment of the invention, the actual observed value It is a hybrid vector that contains a portion of the D-Engine's output, but is not limited to the D-Engine's output.

[0121] In other words, the actual observed value It can be composed of two parts: Part 1: High-confidence output of D-Engine, which comes from the data-driven engine (verified and reliable results); Part 2: Direct observations, which come from field sensors (raw hardware data).

[0122] In this embodiment of the invention, treating the output of the D-Engine as an "observation" is a high-level technique in EnKF assimilation, namely "observation substitution" or "observation enhancement".

[0123] In this embodiment of the invention, the limitations of direct observation can be avoided. Specifically, the number of field sensors is limited (e.g., only a few pore pressure probes), which can only provide data from sparse points.

[0124] D-Engine can function as a "virtual sensor": it integrates multi-source data (such as ocean dynamics, engineering disturbances, and historical patterns) to infer the state of places without direct sensors, or to reduce noise and correct direct observations to obtain more reliable "high confidence" estimates than the original hardware readings.

[0125] In simple terms, direct observations can be understood as images captured by a regular camera (which may have noise and be blurry). The D-Engine output can be understood as an enhanced, high-definition image (clearer and more accurate). The observation vector can be understood as taking both the "regular camera image" and the "enhanced high-definition image" and using them together as a reference standard to correct the physical model.

[0126] 3) Feedback: The updated analysis set The mean as The new optimal parameter estimates are used for simulation in the next time window, thus completing the "learn-correction" cycle.

[0127] In this embodiment of the invention, P-Engine provides a prediction framework based on physical laws. D-Engine provides high-confidence observational data. EnKF acts as a bridge, integrating the two to continuously correct the system state and model parameters, thereby achieving a dynamic and accurate assessment of the stability of offshore platforms.

[0128] like Figure 2 As shown in this embodiment of the invention, the final assessment result and risk level are obtained based on the calculation results of the dual engines, completing risk quantification and graded early warning. This is the final output for platform operation and maintenance. Through weight normalization, the risk fusion of mechanism and data is achieved. Furthermore, a three-level early warning mechanism is implemented: yellow / orange / red risk warnings are generated based on the dynamic fusion assessment results and directly output to the offshore platform control room, safety management department, and emergency response center to guide decision-making and handling.

[0129] In this embodiment of the invention, the output weights of the two engines can be dynamically calculated based on information entropy and spatial variation index, and the minimum safety index of the physical engine and the risk probability of the data engine can be weighted and fused to generate a unified stability index.

[0130] Specifically, step S104 includes: S1041, Calculate the spatial variation index of the output result of the multiphysics coupling numerical model; S1042, Calculate the information entropy of the non-invariant probability distribution output by the physical information neural network model; S1043, determine the fusion weights of the multiphysics coupled numerical model and the physical information neural network model based on the spatial variation index and the information entropy.

[0131] S1044, Dynamic weighted fusion is performed based on the fusion weights of the multiphysics coupling numerical model and the physical information neural network model.

[0132] In this embodiment of the invention, a multi-indicator fusion logic is used to trigger early warning information. Three levels of early warning (yellow, orange, and red) are set based on the stability index and risk probability, and specific triggering conditions and response measures are associated to achieve a progressive early warning from "attention" to "alarm". For example: when When the rise is rapid, even if the stability index after fusion has not yet dropped significantly or the physical field characteristics are not yet obvious, the system can issue a higher level of warning in advance to gain more response time.

[0133] The final decision-making logic is: based on the stability index, and based on key characteristic signals (such as...) Rapid ascent and plastic zone penetration are used as acceleration factors, and sudden changes in the original data are used as confirmation signals for comprehensive judgment.

[0134] Specifically, step S105 includes: S1051, if the stability index after fusion is continuously lower than the first threshold or the risk probability is continuously higher than the first probability threshold, a warning of concern level is triggered. S1052, when the stability index momentarily falls below the lower second threshold, or the risk probability momentarily rises above the higher second probability threshold, and the physical mechanism analysis engine shows that the plastic zone has a tendency to penetrate; triggers a warning-level alert. S1053: An alarm-level warning is triggered when the stability index momentarily falls below a lower third threshold, the risk probability is higher than a higher third probability threshold, and the on-site monitoring data changes drastically.

[0135] In this embodiment of the invention, like Figure 6 As shown, this illustrates a flowchart of the dynamic weighted fusion and hierarchical early warning triggering steps in an embodiment of the present invention. Specifically, the stability index fusion calculation includes: calculate Output the overall minimum safety index .

[0136] calculate Output disaster risk probability .

[0137] Assess the uncertainty of the dual-engine output in this test: , ( The spatial variation index of the result.

[0138] , ( Output information entropy).

[0139] Dynamic weight allocation: The final weights are obtained after normalization. , .

[0140] Fusion stability index ： .in This is the normalized safety index.

[0141] The trigger logic for a Level 3 early warning includes: Yellow alert (attention level): When the stability index is below the threshold for two consecutive assimilation periods (12 hours). (e.g., 0.85), or If the risk level exceeds 30% for two consecutive cycles, the system will highlight the risk area on the monitoring interface and alert the operators.

[0142] Orange alert (warning level): When the stability index momentarily falls below the threshold. (e.g., 0.65), or Instantaneously above 60%, and The simulation showed that the plastic strain zone formed a potential, continuous slip surface in space. The system automatically sent alarm messages to the platform manager and safety department, suggesting that they prepare to activate the emergency response plan.

[0143] Red Alert (Alert Level): When the stability index momentarily falls below the threshold. (e.g., 0.45), and If the pressure exceeds 80% and on-site monitoring data (such as pore pressure) shows a sharp change, the system will automatically trigger the highest level alarm and can send a suggested "slow down operation" or "shut down" signal to the platform control system (which must be executed after interlocking confirmation with the platform safety system) to ensure the safety of personnel and facilities.

[0144] In this embodiment of the invention, an exemplary embodiment is also provided to illustrate the physical-data coupled dynamic assessment method for geological states applicable to deep-water shallow strata provided by the present invention.

[0145] Taking the development of a deep-water shallow gas field in the South China Sea as an example. After deploying the above system, during a period of strong inland tides, the marine dynamic monitoring unit recorded a significant increase in shear stress. The dual-engine system operation showed: Simulations indicate that plastic strain begins to accumulate in the near-wellbore region.

[0146] The risk probability is calculated based on real-time response data. It rose from 25% to 55% within 3 hours.

[0147] The data assimilation module corrected the soil softening parameters in the model accordingly.

[0148] After fusion The value dropped to 0.68, triggering an orange alert.

[0149] Based on the early warning, the platform took preventative measures to reduce the production pressure differential. Subsequently, monitoring data showed that the formation deformation rate tended to stabilize. The value rebounded, and the warning was lifted. This process successfully warned of a potential shallow instability risk, avoiding a possible major accident and economic loss.

[0150] Based on the same inventive concept, embodiments of the present invention also provide a physical-data coupled dynamic assessment system for geological states applicable to deep-water shallow strata, the system comprising: The data processing module is used to acquire and standardize marine dynamic data, engineering activity data, and geological response data of the target geological body in real time; The physical mechanism analysis engine is used to calculate the evolution of stress field, seepage field, and deformation field of target geological bodies based on multi-physics coupled numerical models, using standardized data as dynamic boundary conditions and source terms, and output safety index field and physical field characteristics. A data-driven analysis engine is used to output the stability state probability distribution and disaster risk scalar value of geological stability based on a physical information neural network model, according to real-time monitoring data sequences and the physical field characteristics. The real-time monitoring data sequences are obtained based on the marine dynamic data, engineering activity data and geological response data. The fusion module is used to dynamically weight and fuse the safety index field, stability state probability distribution and disaster risk scalar value to obtain the stability index and risk probability. The early warning module is used to trigger corresponding early warning information based on the stability index and risk probability, according to preset multi-level thresholds and logical rules. The multiphysics coupled numerical model and the physical information neural network model adopt a ensemble Kalman filter framework to achieve bidirectional dynamic coupling and correction, including: using the high confidence information output by the data-driven analysis engine and some direct observation data as observation vectors to update the model parameters and state of the physical mechanism analysis engine in reverse, and injecting the physical field features output by the physical mechanism analysis engine into the data-driven analysis engine in a positive direction.

[0151] Optionally, the physical information neural network model includes: an input layer, a feature extraction branch, a physical attention mechanism unit, and a fusion output layer; the data-driven analysis engine is used for: The input layer receives real-time monitoring data sequences within a fixed time window. Convolutional layers based on feature extraction branches extract local spatiotemporal features, while long short-term memory network layers capture long-term dependencies to obtain spatiotemporal feature vectors. The physical attention mechanism unit receives the physical field features and generates a physical attention weight vector. Based on the modulation of the spatiotemporal feature vector and the physical attention weight vector by the fusion output layer, the probability distribution of stability state and scalar value of disaster risk for future time periods are output.

[0152] Optionally, the training steps of the physical information neural network model include: Collect historical data for one or more complete annual cycles for the target geological body; Based on the historical data and the multiphysics coupled numerical model, the security index and physical field characteristics corresponding to the historical data are obtained. The training labels corresponding to the historical data are determined based on the security index and physical field characteristics corresponding to the historical data. The physical information neural network model is trained using historical data carrying the training labels, based on cross-entropy loss, physical constraint loss, and regularization loss.

[0153] Optionally, the stability state probability distribution includes: stable state probability, state of interest probability, and unstable state probability, wherein the sum of the stable state probability, state of interest probability, and unstable state probability is 1; Among them, the stable state indicates that the target geological body is within a safe and stable range and the monitoring parameters show no abnormal trend; the state of concern indicates that the target geological body has entered a critical or warning state; and the unstable state indicates that the target geological body has reached or is close to an unstable and destructive state.

[0154] Optionally, the fusion module is used for Calculate the spatial variation index of the output of the multiphysics coupled numerical model; Calculate the information entropy of the non-invariant probability distribution output by the physical information neural network model; The fusion weights of the multiphysics coupled numerical model and the physical information neural network model are determined based on the spatial variation index and the information entropy. Dynamic weighted fusion is performed based on the fusion weights of the multiphysics coupling numerical model and the physical information neural network model.

[0155] Optionally, the early warning module is used for: If the stability index after fusion is continuously lower than the first threshold or the risk probability is continuously higher than the first probability threshold, a warning of concern level will be triggered. A warning level alert is triggered when the stability index momentarily falls below the lower second threshold, or the risk probability momentarily rises above the higher second probability threshold, and the physical mechanism analysis engine shows that the plastic zone has a tendency to penetrate. An alarm-level warning is triggered when the stability index momentarily falls below a lower third threshold, the risk probability is higher than a higher third probability threshold, and the on-site monitoring data shows a sharp change.

[0156] The embodiment of the physical-data coupled dynamic geological state assessment system for deep-water shallow strata provided by this invention can be applied to any device with data processing capabilities, such as a computer. The device embodiment can be implemented through software, hardware, or a combination of both. Taking software implementation as an example, as a logical device, it is formed by the processor of any data-processing device loading the corresponding computer program instructions from non-volatile memory into memory for execution.

[0157] Based on the same inventive concept, embodiments of the present invention also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the steps of the physical-data coupled dynamic geological state assessment method applicable to deep-water shallow strata as described in any of the above embodiments.

[0158] Based on the same inventive concept, embodiments of the present invention also provide a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps in the physical-data coupled dynamic assessment method for geological states applicable to deep-water shallow strata described in any of the above embodiments.

[0159] Based on the same inventive concept, embodiments of the present invention provide a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps in the physical-data coupled dynamic geological state assessment method applicable to deep-water shallow strata described in any of the above embodiments.

[0160] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0161] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, embodiments of the present invention can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of the present invention can take the form of computer program products implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0162] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (apparatus), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable terminal device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0163] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable terminal device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0164] These computer program instructions can also be loaded onto a computer or other programmable terminal device to cause a series of operational steps to be performed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal device for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0165] Although preferred embodiments of the present invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present invention.

[0166] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0167] The present invention provides a detailed description of a physical-data coupled dynamic assessment method for geological states applicable to shallow deep-water strata. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, those skilled in the art will recognize that there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A physical-data coupled dynamic assessment method for geological state of shallow deep-water strata, characterized in that, The method includes: Real-time acquisition and standardized processing of marine dynamic data, engineering activity data, and geological response data of the target geological body; Based on a physical mechanism analysis engine, a multi-physics coupled numerical model is used to calculate the evolution of stress field, seepage field, and deformation field of the target geological body using standardized data as dynamic boundary conditions and source terms, and outputs safety index field and physical field characteristics. The physical information neural network model based on the data-driven analysis engine outputs the stability state probability distribution and disaster risk scalar value according to the real-time monitoring data sequence and the physical field characteristics. The real-time monitoring data sequence is obtained based on the marine dynamic data, engineering activity data and geological response data. The safety index field, the stability state probability distribution, and the disaster risk scalar value are dynamically weighted and fused to obtain the stability index and the risk probability. Based on the stability index and risk probability, corresponding early warning information is triggered according to preset multi-level thresholds and logical rules. The multiphysics coupled numerical model and the physical information neural network model adopt a ensemble Kalman filter framework to achieve bidirectional dynamic coupling and correction, including: using the high confidence information output by the data-driven analysis engine and some direct observation data as observation vectors to update the model parameters and state of the physical mechanism analysis engine in reverse, and injecting the physical field features output by the physical mechanism analysis engine into the data-driven analysis engine in a positive direction.

2. The method for dynamic assessment of geological state using physical-data coupling applicable to shallow deep-water strata according to claim 1, characterized in that, The physical information neural network model includes: an input layer, a feature extraction branch, a physical attention mechanism unit, and a fusion output layer; the physical information neural network model based on a data-driven analysis engine outputs the stability state probability distribution and disaster risk scalar value of geological stability according to the real-time monitoring data sequence and the physical field characteristics, including: The input layer receives real-time monitoring data sequences within a fixed time window. Convolutional layers based on feature extraction branches extract local spatiotemporal features, while long short-term memory network layers capture long-term dependencies to obtain spatiotemporal feature vectors. The physical attention mechanism unit receives the physical field features and generates a physical attention weight vector. Based on the modulation of the spatiotemporal feature vector and the physical attention weight vector by the fusion output layer, the probability distribution of stability state and scalar value of disaster risk for future time periods are output.

3. The method for dynamic assessment of geological state using physical-data coupling applicable to shallow deep-water strata according to claim 1, characterized in that, The training steps of the physical information neural network model include: Collect historical data for one or more complete annual cycles for the target geological body; Based on the historical data and the multiphysics coupled numerical model, the security index and physical field characteristics corresponding to the historical data are obtained. The training labels corresponding to the historical data are determined based on the security index and physical field characteristics corresponding to the historical data. The physical information neural network model is trained using historical data carrying the training labels, based on cross-entropy loss, physical constraint loss, and regularization loss.

4. The method for dynamic assessment of geological state using physical-data coupling applicable to shallow deep-water strata according to claim 1, characterized in that, The stability state probability distribution includes: stable state probability, state of interest probability, and unstable state probability, and the sum of the stable state probability, state of interest probability, and unstable state probability is 1. Among them, the stable state indicates that the target geological body is within a safe and stable range and the monitoring parameters show no abnormal trend; the state of concern indicates that the target geological body has entered a critical or warning state; and the unstable state indicates that the target geological body has reached or is close to an unstable and destructive state.

5. The method for dynamic assessment of geological state using physical-data coupling applicable to shallow deep-water strata according to claim 1, characterized in that, The safety index field, stability state probability distribution, and disaster risk scalar value are dynamically weighted and fused, including: Calculate the spatial variation index of the output of the multiphysics coupled numerical model; Calculate the information entropy of the non-invariant probability distribution output by the physical information neural network model; The fusion weights of the multiphysics coupled numerical model and the physical information neural network model are determined based on the spatial variation index and the information entropy. Dynamic weighted fusion is performed based on the fusion weights of the multiphysics coupling numerical model and the physical information neural network model.

6. The method for dynamic assessment of geological state using physical-data coupling applicable to shallow deep-water strata according to claim 1, characterized in that, Based on the stability index and risk probability, corresponding early warning information is triggered according to preset multi-level thresholds and logical rules, including: If the stability index after fusion is continuously lower than the first threshold or the risk probability is continuously higher than the first probability threshold, a warning of concern level will be triggered. A warning level alert is triggered when the stability index momentarily falls below the lower second threshold, or the risk probability momentarily rises above the higher second probability threshold, and the physical mechanism analysis engine shows that the plastic zone has a tendency to penetrate. An alarm-level warning is triggered when the stability index momentarily falls below a lower third threshold, the risk probability is higher than a higher third probability threshold, and the on-site monitoring data shows a sharp change.

7. A physical-data coupled dynamic assessment system for geological states of shallow and deep-water strata, characterized in that, The system includes: The data processing module is used to acquire and standardize marine dynamic data, engineering activity data, and geological response data of the target geological body in real time; The physical mechanism analysis engine is used to calculate the evolution of stress field, seepage field, and deformation field of target geological bodies based on multi-physics coupled numerical models, using standardized data as dynamic boundary conditions and source terms, and output safety index field and physical field characteristics. A data-driven analysis engine is used to output the stability state probability distribution and disaster risk scalar value of geological stability based on a physical information neural network model, according to real-time monitoring data sequences and the physical field characteristics. The real-time monitoring data sequences are obtained based on the marine dynamic data, engineering activity data and geological response data. The fusion module is used to dynamically weight and fuse the safety index field, stability state probability distribution and disaster risk scalar value to obtain the stability index and risk probability. The early warning module is used to trigger corresponding early warning information based on the stability index and risk probability, according to preset multi-level thresholds and logical rules. The multiphysics coupled numerical model and the physical information neural network model adopt a ensemble Kalman filter framework to achieve bidirectional dynamic coupling and correction, including: using the high confidence information output by the data-driven analysis engine and some direct observation data as observation vectors to update the model parameters and state of the physical mechanism analysis engine in reverse, and injecting the physical field features output by the physical mechanism analysis engine into the data-driven analysis engine in a positive direction.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the physical-data coupled dynamic assessment method for geological state of deep-water shallow strata as described in any one of claims 1-6.

9. A readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the physical-data coupled dynamic assessment method for geological state of shallow deep-water strata as described in any one of claims 1-6.

10. A computer program product comprising a computer program / instructions, characterized in that, When executed by a processor, the computer program / instruction implements the steps in the physical-data coupled dynamic assessment method for geological state of shallow deep-water strata as described in any one of claims 1-6.

Images