A method and system for integrated monitoring and early warning of underground gas storage facilities using air, ground, and well systems.

Abstract

This invention provides an integrated monitoring and early warning method and system for underground gas storage facilities, belonging to the field of energy storage safety monitoring technology. The method includes: collecting multi-dimensional monitoring data from the ground, air, and wellbore, comprising eight modules: surface wellhead pressure and temperature, surface subsidence, well site leakage, oil-gas-water interface, wellbore leakage, tracer data, microseismic data, and UAV inspection data; standardizing and preprocessing the multi-source heterogeneous data; constructing a dynamic spatiotemporal graph with monitoring points as nodes and spatiotemporal correlations as edges; extracting multi-scale features using convolutional neural networks and graph convolutional networks; achieving deep cross-fusion of features using a cross-modal attention mechanism; simultaneously assessing wellbore integrity, reservoir stability, and leakage risk through multi-task collaborative learning to generate a comprehensive risk value; and classifying early warning into low, medium, and high levels based on risk thresholds. This significantly improves the accuracy and real-time performance of early warnings, effectively ensuring the safe operation of underground gas storage facilities.

Description

Technical Field

[0001] This invention relates to the field of energy storage safety monitoring technology, and in particular to an integrated monitoring and early warning method and system for underground gas storage facilities using air, ground, and well systems. Background Technology

[0002] The integrated air-ground-ground-well monitoring and early warning system, through the fusion of multi-source sensing technologies, constructs a three-dimensional monitoring network covering "air-space-ground-well," becoming a core support for geological disaster early warning and resource development safety. Ground subsidence monitoring technology, based on InSAR (Inductive Aperture Radar Interferometry), achieves millimeter-level deformation detection. Combined with GNSS (Global Navigation Satellite System) and a level, it forms a multi-scale verification system, controlling the error to ±2mm / year in mine subsidence monitoring. Wellbore leakage fiber optic monitoring technology employs distributed acoustic sensing (DAS) and temperature sensing (DTS) fibers to capture micron-level strain and temperature anomalies in the wellbore in real time, achieving a sensitivity of 0.1με. In shale gas field applications, it achieves a leak point location accuracy of ±5 meters. Ground well site gas leakage monitoring relies on laser methane telemetry and UAV-borne spectrometers to achieve ppb-level methane concentration detection. Combined with AI image recognition technology, the leak identification rate exceeds 95%. Microseismic monitoring technology, through the linkage of deep well geophone arrays and surface seismic networks, achieves a positioning accuracy of 50 meters and can capture M-1.0 level micro-fracture events, supporting the evaluation of hydraulic fracturing effects and early warning of fault activation.

[0003] The aerospace monitoring layer uses satellite remote sensing as a macro framework. Optical satellites conduct general surveys of surface deformation, SAR satellites penetrate clouds and rain to monitor millimeter-level displacements, and UAVs equipped with LiDAR and multispectral sensors fill ground-based blind spots, constructing a three-tiered system of "space-based general survey - air-based detailed survey - ground verification." The downhole monitoring layer integrates fiber optic sensing, downhole robots, and wireless sensor networks to achieve full lifecycle monitoring of underground facilities such as salt wells and coalbed methane wells. For example, the automatic control system for coalbed methane well drainage achieves remote control accuracy of ±0.1 MPa through embedded development. At the data fusion level, a WebGIS-based early warning platform integrates multi-source data and constructs an evaluation system of "target layer - criterion layer - sub-criterion layer" using the analytic hierarchy process, enabling automatic early warning of leakage risks in scenarios such as CO2 sequestration.

[0004] The technology has been applied in fields such as geological disasters, energy development, and carbon sequestration. For example, in underground gas storage facilities, laser scanning and sonar fusion modeling achieves centimeter-level accuracy, supporting dynamic assessment of surrounding rock stability; in coal mine goaf areas, surface displacement sensors and underground borehole inclinometers work together to provide over 90% accuracy in early warning of surface collapse. AI-driven multimodal data cleaning technology can automatically remove abnormal data, improve the reliability of monitoring data, and provide decision support for the "prediction-early warning-rehearsal-contingency plan" closed loop.

[0005] The integrated monitoring and early warning system for air, ground, and wellbore systems faces multiple technical bottlenecks. Regarding environmental adaptability, highly reflective surfaces (such as salt cavities) cause laser signal overexposure, resulting in a point cloud loss rate of up to 30%, while dust and humidity interference further reduce scanning quality. The high temperature and humidity environment underground (over 60°C) shortens the lifespan of fiber optic sensors by 50%, and while explosion-proof designs improve safety, equipment size and power consumption limit portability. In terms of data fusion, the spatiotemporal references of multi-source data (such as InSAR and fiber optic sensing) are inconsistent, resulting in fusion errors of ±15%. Although WebGIS platforms support visualization, real-time processing delays exceed 10 minutes, making it difficult to meet the needs of emergency disaster response.

[0006] The technology has significant limitations. Microseismic monitoring has an accuracy rate of less than 40% for events below magnitude M-1.0; the deployment cost of deep-well geophone arrays is prohibitively high, exceeding one million yuan per well; gas leak monitoring in complex terrain (such as mountainous areas) has a drone aerial survey coverage rate of only 60%, and laser methane detectors are affected by smog, increasing the false alarm rate by 25%. In downhole monitoring, fiber optic sensors experience signal attenuation of up to 50% in curved wellbores, and wireless sensor networks suffer from communication interruptions exceeding 30% in deep formations, leading to data loss. At the algorithm level, AI models achieve an accuracy rate of less than 70% in feature-sparse scenarios (such as salt cavity fractures), and semi-automatic methods rely on manual annotation, resulting in low efficiency.

[0007] System integration presents significant challenges. Hardware synchronization calibration is difficult; for example, the time synchronization error between the UAV and ground sensors exceeds 0.1 seconds, affecting the accuracy of 3D modeling. Explosion-proof design results in low modularity of equipment, increasing maintenance costs by 30%. Furthermore, the lack of standardized procedures for sonar and laser data fusion leads to uneven accuracy in the 3D model, affecting the reliability of surrounding rock stability analysis. Summary of the Invention

[0008] This invention aims to solve at least one of the technical problems existing in the prior art, and proposes an integrated monitoring and early warning method and system for underground gas storage facilities, which solves the problems of data silos, inconsistent spatiotemporal references, strong environmental interference, and low positioning accuracy in the prior art.

[0009] In a first aspect, embodiments of the present invention provide an integrated monitoring and early warning method for underground gas storage facilities, including:

[0010] Using each monitoring point in the underground gas storage facility as a node and the physical and spatiotemporal relationships between monitoring points as edges, a dynamic time-weighted matrix is ​​introduced to characterize the dynamic evolution of the relationship strength, thus constructing a dynamic spatiotemporal graph structure.

[0011] Based on the dynamic spatiotemporal graph structure, multi-scale feature extraction is performed on the monitoring data of each monitoring point, including extracting local anomaly patterns using a one-dimensional convolutional neural network and integrating information of spatially adjacent nodes using graph convolution operations.

[0012] A cross-modal attention mechanism is adopted to cross-fuse features extracted from different monitoring modules in order to uncover deep semantic relationships between different monitoring data.

[0013] Based on the fused features, wellbore integrity assessment, reservoir stability analysis and leakage risk warning are performed simultaneously through a multi-task collaborative learning module to generate comprehensive risk assessment results.

[0014] Based on the comprehensive risk assessment results, the risk levels are divided into low-risk, medium-risk, and high-risk categories, and early warning signals are issued.

[0015] The technical advantages of the integrated monitoring and early warning method for underground gas storage facilities disclosed in this invention are: it achieves the integration of multi-source heterogeneous data from acquisition, alignment, correlation modeling to intelligent diagnosis throughout the entire process. Compared with traditional methods that analyze data from each module independently, this method can reveal the complex correlations hidden between data (such as the relationship between microseismic events and injection / production pressure), shorten the early warning response time from more than 10 minutes in traditional methods to less than 100 ms, reduce the false alarm rate to below 3%, and improve the detection accuracy to over 92%, significantly improving the intelligence level and reliability of underground gas storage facility safety monitoring.

[0016] Furthermore, the monitoring module includes:

[0017] The surface wellhead pressure and temperature monitoring module is used to monitor the temperature and pressure at the wellhead and surrounding key areas.

[0018] A real-time ground settlement monitoring module is used to monitor ground deformation by integrating leveling measurements with InSAR technology;

[0019] A real-time well site leakage monitoring module is used to monitor the methane concentration at the well site using an all-fiber photothermal spectroscopy method.

[0020] The oil-gas-water interface monitoring module is used to monitor the location of the oil-water or gas-water interface based on the difference in specific heat capacity and distributed fiber optic temperature measurement technology.

[0021] The wellbore leakage monitoring module is used to monitor wellbore leakage signals through distributed fiber optic acoustic sensing (DAS) and distributed fiber optic temperature sensing (DTS).

[0022] The tracer concentration change and transport dynamic monitoring module is used to evaluate fluid transport patterns and caprock sealing by injecting and detecting tracers.

[0023] The underground storage microseismic monitoring module is used to acquire microseismic events through surface monitoring, multi-well + single-level monitoring, or single-well + multi-level monitoring.

[0024] The drone intelligent inspection module is used to monitor gas leaks and collect evidence of suspicious persons by using gas sensors and high-definition cameras.

[0025] Furthermore, the leakage judgment criteria of the wellbore leakage monitoring module are: the frequency amplitude of the DAS signal undergoes continuous dynamic changes, and the temperature difference of the DTS signal exceeds 0.2℃.

[0026] Furthermore, the real-time well site leakage monitoring module adopts a methane monitoring method based on all-fiber photothermal spectroscopy, and its monitoring accuracy reaches within 1 ppm within a 1 km range.

[0027] Furthermore, in the step of multi-scale feature extraction of time-series monitoring data from each monitoring point, the formula for the graph convolution operation is:

[0028] ;

[0029] in, Let be the dynamic adjacency matrix at time t. Its degree matrix, Let l be the feature matrix of the l-th layer. Let be the parameter matrix, and σ be the activation function.

[0030] Furthermore, the cross-fusion of features extracted from different monitoring modules includes a multi-source data standardization preprocessing step beforehand.

[0031] Heterogeneous data from different monitoring modules are cleaned, aligned, and normalized.

[0032] For each type of monitoring data, a timeliness factor M and a weighting factor W are set, where the values ​​of the timeliness factor M and the weighting factor W are both in the range of 0 to 1;

[0033] The monitoring data includes at least: ground subsidence data S, wellhead temperature T and pressure P, gas diffusion concentration C, interface depth location Inf, distributed fiber optic acoustic data Das, distributed fiber optic temperature data Dts, tracer data Tra, microseismic data Ms, and UAV emergency inspection data Uav.

[0034] Furthermore, the comprehensive risk assessment result is used to calculate the uniform variable F using the following formula. all get:

[0035] ;

[0036] in, This is the normalized monitoring dataset for the i-th monitoring module. Its weighting factor, Its timeliness factor.

[0037] Furthermore, the criteria for classifying low-risk, medium-risk, and high-risk levels are as follows:

[0038] When F all The value is between 0 and 30% of Max{F all When the range is within a certain range, it indicates a low-risk warning.

[0039] When F all The value of Max{F} is between 30% and 60%. all When the range is within a certain range, it is considered a medium-risk warning.

[0040] When F all The value of Max{F} is between 60% and 100%. all When the value is within the specified range, a high-risk warning is issued.

[0041] Furthermore, the adoption of the cross-modal attention mechanism includes the feature F extracted from the wellbore leakage module. DAS Features of ground settlement module F S Calculate cross-attention:

[0042] ;

[0043] in, These are the query projection matrix, key projection matrix, and value projection matrix, respectively. k To query the feature dimensions of Q and key K.

[0044] Secondly, embodiments of the present invention also provide an integrated monitoring and early warning system for underground gas storage facilities, comprising:

[0045] The dynamic spatiotemporal graph construction module is configured to use each monitoring point of the underground gas storage as a node, the physical and spatiotemporal relationships between the monitoring points as edges, and introduce a time dynamic weight matrix to represent the dynamic evolution of the relationship strength, thereby constructing a dynamic spatiotemporal graph structure.

[0046] The feature extraction module is configured to perform multi-scale feature extraction on the monitoring data of each monitoring point based on the dynamic spatiotemporal graph structure, including extracting local anomaly patterns using a one-dimensional convolutional neural network and integrating information of spatially adjacent nodes using graph convolution operations.

[0047] The feature fusion module is configured to use a cross-modal attention mechanism to cross-fuse features extracted by different monitoring modules in order to uncover deep semantic relationships between different monitoring data.

[0048] The risk assessment module is configured to perform wellbore integrity assessment, reservoir stability analysis, and leakage risk warning simultaneously through a multi-task collaborative learning module based on the fused features, generating a comprehensive risk assessment result.

[0049] The early warning output module is configured to classify risks into low, medium, and high risk levels based on the comprehensive risk assessment results and output early warning signals.

[0050] The technical advantages of the integrated monitoring and early warning system for underground gas storage facilities, disclosed in this invention, are as follows: It integrates the dispersed monitoring networks of ground, air, and wells into a cohesive whole, achieving a closed loop from data acquisition to intelligent decision-making through hardware and software collaboration. The platform-based design improves the system's modularity and maintainability, while the standardized data processing flow solves the problem of multi-source data fusion, providing underground gas storage facilities with a highly available (99.99%) integrated safety assurance solution. Attached Figure Description

[0051] Figure 1 This is a flowchart illustrating an integrated monitoring and early warning method for underground gas storage facilities using air, ground, and well systems, as provided in an embodiment of the present invention. Detailed Implementation

[0052] To enable those skilled in the art to better understand the technical solutions of the present invention, exemplary embodiments of the present invention are described below in conjunction with the accompanying drawings, including various details of the embodiments of the present invention to aid understanding. These should be considered merely exemplary. Therefore, those skilled in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0053] An integrated monitoring and early warning method for underground gas storage facilities, encompassing air, ground, and well sources, mainly comprises the following three parts: 1. Multi-dimensional integrated monitoring of underground gas storage facilities; 2. Fusion and analysis of multi-source, multi-dimensional, heterogeneous data; 3. Technical route and implementation plan for multi-source monitoring data fusion and analysis of underground gas storage facilities. (Reference) Figure 1 As shown, the specific steps include:

[0054] S1, Multi-source data acquisition and standardized preprocessing. Using each monitoring point in the underground gas storage facility as a node and the physical and spatiotemporal correlations between monitoring points as edges, a dynamic temporal weight matrix is ​​introduced to represent the dynamic evolution of the correlation strength, thus constructing a dynamic spatiotemporal graph structure.

[0055] The data comes from the underground gas storage facility's integrated monitoring system for air, ground, and well depths, which comprises eight modules:

[0056] (1) Surface wellhead pressure and temperature monitoring module: Sensors are arranged at and around the wellhead to monitor temperature T and pressure P in real time.

[0057] (2) Real-time ground settlement monitoring module: Leveling is carried out using deeply buried leveling stones, and the InSAR technology processing flow is optimized. By using the INSAR data interpolation method constrained by the leveling results, the high precision of the leveling points is combined with the wide coverage advantage of InSAR to achieve sub-millimeter level monitoring accuracy and obtain ground settlement data S.

[0058] (3) Real-time well site leakage monitoring module: A multi-channel methane detection system based on all-fiber photothermal spectroscopy is adopted. A DFB laser with a wavelength of 1650.9nm is used as the pump light. Combined with a narrow linewidth laser, the refractive index change caused by the photothermal effect is demodulated by a lock-in amplifier to calculate the methane concentration C in real time with an accuracy of less than 1ppm within a 1km range. The sensor housing is made of stainless steel and has a waterproof and breathable membrane.

[0059] (4) Oil-gas-water interface monitoring module: Based on the differences in specific heat capacity of oil, gas, and water (e.g., diesel 2.0×10^3, air 1.005×10^3, water 4.2×10^3 J / (kg·℃)). The medium in the well is heated by an energized resistance wire, and the temperature curve along the wellbore is measured using a distributed optical fiber temperature measurement system (DTS). At the oil-water or gas-water interface, the difference in specific heat capacity leads to a difference in the temperature rise after heat absorption, forming a significant temperature gradient, thereby accurately determining the interface depth position Inf.

[0060] (5) Wellbore Leakage Monitoring Module: Utilizing an SCTGPGP-encapsulated armored optical cable adapter with a distributed fiber optic demodulator. The DAS captures leakage acoustic vibration signals, while the DTS monitors for temperature anomalies. The established leakage judgment criterion is: a continuous dynamic change in DAS frequency amplitude and a DTS temperature difference exceeding 0.2℃. Combined with wellhead pressure and temperature data, the positioning accuracy is improved, with a test error of less than 1 meter.

[0061] (6) Tracer concentration change and transport dynamic monitoring module: Selected tracer is injected into the gas injection well, and samples are taken from the monitoring well every 2 to 3 months to measure the tracer concentration Tra in order to evaluate fluid transport and caprock sealing.

[0062] (7) Microseismic monitoring module for underground storage: Considering the dispersed nature of salt cavern cavities, a "multi-well + single-stage" monitoring method is preferred, balancing coverage and weak event detection capability. Noise suppression, event identification and location are performed on continuous waveform data to obtain microseismic data Ms.

[0063] (8) Intelligent inspection module for drones: DJI M350 / M30 drones are used, equipped with laser methane sensor (monitoring accuracy <100ppm), high-definition camera (≥20 million pixels) and infrared thermal imager (temperature measurement accuracy ±1℃) to carry out regular inspections and emergency inspections, and obtain Uav emergency inspection data of drones for gas leak monitoring and evidence collection of suspicious persons.

[0064] The heterogeneous data collected by the above modules are cleaned (e.g., missing values ​​and outliers are removed), aligned (using the PTP protocol to achieve nanosecond-level time synchronization and unify spatial coordinates), and normalized. Attributes are assigned to each type of data: Type 1 (Day): S; Type 2 (Ground): T, P, C, Inf, Tra; Type 3 (Well): Das, Dts, Ms; Type 4 (Null): Uav. And set the timeliness factor M and the weight factor W, for example: S(M=0.2, W=0.9), T / P(M=0.8, W=0.8), C(M=0.9, W=0.5), Inf(M=0.1, W=0.7), Das(M=0.6, W=0.3), Dts(M=0.3, W=0.6), Tra(M=0.1, W=0.4), Ms(M=0.8, W=0.6), Uav(M=0.9, W=0.9).

[0065] S2, Dynamic Spatiotemporal Graph Construction. Based on the dynamic spatiotemporal graph structure, multi-scale feature extraction is performed on the monitoring data of each monitoring point, including extracting local anomaly patterns using a one-dimensional convolutional neural network and integrating information of spatially neighboring nodes using graph convolution operations.

[0066] The preprocessed multi-source monitoring data is modeled as a dynamic heterogeneous graph structure. Among them, the node set Corresponding to each monitoring point (such as wellhead pressure sensor, leveling point, fiber optic sensing point, microseismic detector, etc.). Edge aggregation. The adjacency matrix is ​​defined by the physical associations (such as sensors in the same wellbore) and spatial proximity between nodes. The initial association strength is described by the adjacency matrix A. To characterize the dynamic evolution of the system state, a time-dynamic weight matrix is ​​introduced to obtain the dynamic adjacency matrix at time t. =A⊙Wt, where ⊙ represents the Hadamard product. At each time step t, the observation data from all N nodes constitute the feature matrix. D is the feature dimension.

[0067] S3, Multi-scale Feature Extraction and Cross-modal Fusion. A cross-modal attention mechanism is employed to cross-fuse features extracted from different monitoring modules in order to uncover deep semantic relationships between different monitoring data.

[0068] First, the local feature extraction layer uses a one-dimensional CNN to process the temporal data X for each node. t Extract local abnormal patterns such as sudden pressure drops and abrupt temperature gradient changes.

[0069] Secondly, the region association layer integrates the information of spatially associated nodes using a graph convolutional network (GCN), and its operation formula is as follows:

[0070] ;

[0071] in, Let be the dynamic adjacency matrix at time t. Its degree matrix, Let l be the feature matrix of the l-th layer. Let be the parameter matrix, and σ be the activation function.

[0072] Next, the cross-modal feature cross-fusion layer employs an attention mechanism. For example, for the feature F extracted from the wellbore leakage module... DAS Features of ground settlement module F S Calculate cross-attention:

[0073] ;

[0074] in, These are the query projection matrix, key projection matrix, and value projection matrix, respectively. k To query the feature dimensions of Q and key K.

[0075] This mechanism can uncover the deep semantic correlation between "intensified underground microseismic activity" and "accelerated local ground subsidence rate," achieving cross-modal information complementarity.

[0076] S4, Multi-task Collaborative Diagnosis and Risk Warning. Based on the fused features, the multi-task collaborative learning module simultaneously performs wellbore integrity assessment, reservoir stability analysis, and leakage risk warning, generating a comprehensive risk assessment result.

[0077] Design a multi-task learning module to simultaneously optimize three objectives: wellbore integrity assessment ( ), reservoir stability analysis ( ) and comprehensive leakage risk warning ( The total loss function is:

[0078] ;

[0079] Dynamically adjust task weights using a gradient normalization strategy. Balance learning across all tasks.

[0080] The model outputs a comprehensive risk assessment result, which is reflected in the unified variable F. all Its calculation integrates data from each module, their weights, and timeliness.

[0081] ;

[0082] in, This is the normalized monitoring dataset for the i-th monitoring module. Its weighting factor, Its timeliness factor.

[0083] S5. Based on the comprehensive risk assessment results, classify the risks into low, medium, and high levels, and output early warning signals.

[0084] When F all The value is between 0 and 30% of Max{F all When the range is within a certain range, it indicates a low-risk warning.

[0085] When F all The value of Max{F} is between 30% and 60%. all When the range is within a certain range, it is considered a medium-risk warning.

[0086] When F all The value of Max{F} is between 60% and 100%. all When the value is within the specified range, a high-risk warning is issued.

[0087] As a crucial facility for strategic natural gas reserves, the safe operation of underground gas storage facilities is directly related to national energy security and regional economic development. The operation of underground gas storage facilities faces multiple risks and challenges, including wellbore leakage, ground subsidence, microseismic activity, well site leaks, and changes in the oil-gas-water interface. With the continuous development of monitoring technology, underground gas storage facilities have deployed various monitoring methods, including pressure sensors, temperature sensors, distributed fiber optic sensors, microseismic monitoring equipment, and InSAR remote sensing systems, generating massive amounts of multi-source heterogeneous monitoring data. However, the utilization of current monitoring data still faces many problems: the various monitoring systems are relatively independent, data formats are inconsistent, and spatiotemporal references are inconsistent, resulting in severe information silos and making it difficult to form a unified safety risk assessment system.

[0088] Multi-source, multi-dimensional heterogeneous data fusion analysis, as a core technology of modern intelligent systems, aims to achieve high-level information integration and value enhancement by integrating data from different sources, types, and scales. This technology addresses the compatibility and consistency issues between monitoring data from different sources through advanced information processing methods, encompassing multiple stages such as data acquisition, preprocessing, feature extraction, data fusion, and information reasoning. In the specific application scenario of underground gas storage facilities, data fusion technology can effectively solve the data silos between different monitoring modules, constructing a comprehensive security monitoring system. In industrial internet platform intrusion detection systems, multi-dimensional data fusion technology integrates various heterogeneous data such as network traffic data, system logs, and user behavior data to form a comprehensive information view, thereby improving the accuracy and efficiency of intrusion detection. The core of multi-dimensional data fusion technology lies in four stages: data preprocessing, feature extraction, data fusion, and decision support.

[0089] The core challenges facing underground gas storage monitoring systems lie in the multi-source, heterogeneous, and spatiotemporally multi-scale characteristics of the monitoring data. Wellbore leakage monitoring modules involve the fusion processing of data from various sensors, including distributed fiber optic sensors (DTS / DAS), pressure sensors, and temperature sensors. In land subsidence monitoring, the fusion technology of GPS and InSAR data integrates the advantages of both technologies, achieving complementary spatiotemporal resolution monitoring effects. Microseismic monitoring technology determines the location and energy magnitude of microseismic events by capturing seismic waves generated when energy is released during underground rock fracturing. Well site leakage and oil-gas-water interface testing and monitoring modules also need to process data collected from different types of sensors, such as vibration sensors, gas sensors, sound sensors, and humidity sensors. These monitoring data differ significantly in sampling frequency, spatial resolution, and data format, making it difficult for traditional data analysis methods to effectively extract their deeper value.

[0090] Based on the same inventive concept, this invention also provides an integrated monitoring and early warning system for underground gas storage facilities, including:

[0091] The dynamic spatiotemporal graph construction module is configured to use each monitoring point of the underground gas storage as a node, the physical and spatiotemporal relationships between the monitoring points as edges, and introduce a time dynamic weight matrix to represent the dynamic evolution of the relationship strength, thereby constructing a dynamic spatiotemporal graph structure.

[0092] The feature extraction module is configured to perform multi-scale feature extraction on the monitoring data of each monitoring point based on the dynamic spatiotemporal graph structure, including extracting local anomaly patterns using a one-dimensional convolutional neural network and integrating information of spatially adjacent nodes using graph convolution operations.

[0093] The feature fusion module is configured to use a cross-modal attention mechanism to cross-fuse features extracted by different monitoring modules in order to uncover deep semantic relationships between different monitoring data.

[0094] The risk assessment module is configured to perform wellbore integrity assessment, reservoir stability analysis, and leakage risk warning simultaneously through a multi-task collaborative learning module based on the fused features, generating a comprehensive risk assessment result.

[0095] The early warning output module is configured to classify risks into low, medium, and high risk levels based on the comprehensive risk assessment results and output early warning signals.

[0096] The system includes:

[0097] (i) Hardware perception layer: namely, the multi-dimensional integrated monitoring system of underground gas storage, air and ground wells as described in claim 2, which includes eight modules of sensors, equipment and transmission networks (such as optical fiber and wireless network) responsible for raw data acquisition.

[0098] (ii) Data fusion and analysis platform: deployed on a server or cloud platform, including:

[0099] Data access and preprocessing unit: Receives data from each module and performs data cleaning, alignment, normalization, and attribute (timeliness M, weight W) labeling.

[0100] Dynamic graph modeling and computation unit: Constructs dynamic spatiotemporal graphs and runs multi-scale feature extraction (CNN, GCN) and cross-modal attention fusion algorithms.

[0101] Multi-task diagnosis and early warning unit: Executes a multi-task collaborative learning model, calculates the unified risk variable Fall, and generates a three-level early warning signal based on the threshold.

[0102] Human-computer interaction and visualization unit: Based on the WebGIS platform, it displays data from each module, fusion analysis results, and early warning levels in real time, and supports historical data backtracking and decision support.

[0103] (iii) Early warning execution terminal: pushes early warning information to the monitoring center screen, management personnel's mobile terminals, etc., to trigger the corresponding emergency plan.

[0104] Beneficial effects:

[0105] (1) The underground gas storage multi-dimensional integrated monitoring system of ground, air and well is composed of the latest and most comprehensive safety monitoring and early warning technologies for underground gas storage. It consists of 8 modules and forms an integrated safety monitoring network of ground, air and well, which can realize all-weather real-time monitoring and early warning of underground gas storage groups.

[0106] (2) Multi-source, multi-dimensional, heterogeneous data fusion analysis method. By integrating data from different sources, types, and dimensions, the shortcomings of a single data source can be compensated for, and more dimensions and more complete information can be obtained, thereby improving the accuracy of decision-making; multi-source data fusion can avoid information silos, and obtain more effective and high-quality data through the combination of different data sources, thereby reducing the cost and difficulty of data analysis and processing; in the process of data fusion, various parallel computing technologies can be applied to simplify the complexity of data processing and significantly improve processing speed and efficiency; multi-source data fusion helps to better understand the overall situation of complex systems and respond to emergencies in a timely manner; by analyzing and mining multi-source data, the allocation and utilization efficiency of resources can be optimized; multi-source data fusion can reveal the implicit relationships between data, discover patterns or laws that cannot be captured by single-source data, and provide support for business innovation; by integrating information from multiple data sources, the errors and limitations of a single data source can be eliminated, and the accuracy and reliability of information can be improved; multi-source data fusion technology can adapt to the needs of real-time and streaming analysis and support low-latency data processing and decision-making.

[0107] In summary, the multi-source, multi-dimensional heterogeneous data fusion and analysis method significantly improves information quality, decision-making efficiency, and business innovation capabilities by integrating the advantages of multi-source data, becoming an important support for digital transformation in various fields.

[0108] (3) Technical route and implementation plan for multi-source monitoring data fusion analysis for underground gas storage.

[0109] The core objective of this solution is to construct an intelligent, adaptive, and scalable multi-source monitoring data fusion and analysis system for underground gas storage facilities, enabling comprehensive monitoring and precise early warning of the safety status of these facilities. Specific objectives include achieving a high degree of integration and collaborative analysis of multi-source monitoring data to improve the accuracy, timeliness, and reliability of safety risk identification and early warning. Specifically, the system needs to meet the following technical indicators: target detection accuracy of 99.3% (false alarm rate <0.5%), and response latency of <100ms. In terms of wellbore leakage monitoring, the system should achieve a false alarm rate below 3%, a detection accuracy of over 92%, a 50% reduction in early warning response time, and a system availability of 99.99%. The design principles follow four basic principles: systematicity, practicality, advancement, and scalability. The systematic principle requires treating the underground gas storage facility as a complete dynamic system, rather than a simple aggregation of isolated monitoring modules. The practicality principle emphasizes that the solution must closely integrate with actual engineering needs to ensure the feasibility and effectiveness of the technical approach. The advancement principle requires that the solution fully absorb and draw upon the latest technological achievements, especially the latest advancements in deep learning and graph neural networks in complex system modeling. The scalability principle ensures that the system can adapt to future technological developments and changes in monitoring needs.

[0110] The core concept of this scheme can be summarized as: "A framework for the fusion and analysis of multi-source monitoring data from underground gas storage facilities based on dynamic graph attention spatiotemporal networks." The core innovation of this concept lies in modeling the multi-source monitoring data of underground gas storage facilities as a dynamic heterogeneous graph. Through a graph attention mechanism, it adaptively learns the dynamic correlation strength between different monitoring modules, constructing an end-to-end risk identification and early warning system. Compared to traditional methods, the differentiated innovation of this scheme is mainly reflected in three key aspects: First, it introduces a multi-source spatiotemporal graph structure to explicitly model the physical correlation and spatiotemporal dependence between monitoring variables; second, it designs a multi-granularity feature cross-fusion module to achieve cross-scale reasoning from local anomalies to system risks; and finally, it establishes a multi-task collaborative learning mechanism to simultaneously optimize multiple objectives such as leak detection, settlement prediction, and risk assessment. This framework breaks through the limitation of treating each monitoring module as an independent system in existing research, achieving true multi-source collaborative analysis.

[0111] Deep learning and graph neural networks are another core technology selection for this solution. Deep learning models demonstrate powerful nonlinear fitting and feature extraction capabilities in data fusion. Neural networks can directly learn the nonlinear mapping of the system, replacing or assisting in linearization steps. For example, in UAV attitude estimation, LSTM networks are used to learn the nonlinear mapping of "angular velocity → attitude angle," replacing the Taylor expansion approximation of the traditional EKF, reducing attitude estimation errors by 30% to 50%. Graph neural networks excel in processing spatially correlated data.

[0112] Example embodiments have been disclosed herein, and while specific terminology has been used, it is for illustrative purposes only and should be construed as such, and is not intended to be limiting. In some instances, it will be apparent to those skilled in the art that features, characteristics, and / or elements described in conjunction with particular embodiments may be used alone, or in combination with features, characteristics, and / or elements described in conjunction with other embodiments, unless otherwise expressly indicated. Therefore, those skilled in the art will understand that various changes in form and detail may be made without departing from the scope of the invention as set forth in the appended claims.

Claims

1. A method for integrated monitoring and early warning of underground gas storage facilities using air, ground, and well systems, characterized in that: include: Using each monitoring point in the underground gas storage facility as a node and the physical and spatiotemporal relationships between monitoring points as edges, a dynamic time-weighted matrix is ​​introduced to characterize the dynamic evolution of the relationship strength, thus constructing a dynamic spatiotemporal graph structure. Based on the dynamic spatiotemporal graph structure, multi-scale feature extraction is performed on the monitoring data of each monitoring point, including extracting local anomaly patterns using a one-dimensional convolutional neural network and integrating information of spatially adjacent nodes using graph convolution operations. A cross-modal attention mechanism is adopted to cross-fuse features extracted from different monitoring modules in order to uncover deep semantic relationships between different monitoring data. Based on the fused features, wellbore integrity assessment, reservoir stability analysis and leakage risk warning are performed simultaneously through a multi-task collaborative learning module to generate comprehensive risk assessment results. Based on the comprehensive risk assessment results, the risk levels are divided into low-risk, medium-risk, and high-risk categories, and early warning signals are issued.

2. The method according to claim 1, characterized in that, The monitoring module includes: The surface wellhead pressure and temperature monitoring module is used to monitor the temperature and pressure at the wellhead and surrounding key areas. A real-time ground settlement monitoring module is used to monitor ground deformation by integrating leveling measurements with InSAR technology; A real-time well site leakage monitoring module is used to monitor the methane concentration at the well site using an all-fiber photothermal spectroscopy method. The oil-gas-water interface monitoring module is used to monitor the location of the oil-water or gas-water interface based on the difference in specific heat capacity and distributed fiber optic temperature measurement technology. The wellbore leakage monitoring module is used to monitor wellbore leakage signals through distributed fiber optic acoustic sensing (DAS) and distributed fiber optic temperature sensing (DTS). The tracer concentration change and transport dynamic monitoring module is used to evaluate fluid transport patterns and caprock sealing by injecting and detecting tracers. The underground storage microseismic monitoring module is used to acquire microseismic events through surface monitoring, multi-well + single-level monitoring, or single-well + multi-level monitoring. The drone intelligent inspection module is used to monitor gas leaks and collect evidence of suspicious persons by using gas sensors and high-definition cameras.

3. The method according to claim 2, characterized in that, The leakage judgment criteria of the wellbore leakage monitoring module are: the frequency amplitude of the DAS signal changes continuously and dynamically, and the temperature difference of the DTS signal changes by more than 0.2℃.

4. The method according to claim 2, characterized in that, The real-time well site leakage monitoring module adopts a methane monitoring method based on all-fiber photothermal spectroscopy, and its monitoring accuracy reaches within 1 ppm within a 1 km range.

5. The method according to claim 1, characterized in that, In the step of multi-scale feature extraction of time-series monitoring data from each monitoring point, the formula for graph convolution operation is: ； in, Let be the dynamic adjacency matrix at time t. Its degree matrix, Let l be the feature matrix of the l-th layer. Let be the parameter matrix, and σ be the activation function.

6. The method according to claim 1, characterized in that, The cross-fusion of features extracted from different monitoring modules also includes a multi-source data standardization preprocessing step: Heterogeneous data from different monitoring modules are cleaned, aligned, and normalized. For each type of monitoring data, a timeliness factor M and a weighting factor W are set, where the values ​​of the timeliness factor M and the weighting factor W are both in the range of 0 to 1; The monitoring data includes at least: ground subsidence data S, wellhead temperature T and pressure P, gas diffusion concentration C, interface depth location Inf, distributed fiber optic acoustic data Das, distributed fiber optic temperature data Dts, tracer data Tra, microseismic data Ms, and UAV emergency inspection data Uav.

7. The method according to claim 6, characterized in that, The comprehensive risk assessment result is calculated using the following formula to determine the uniform variable F. all get: ； in, This is the normalized monitoring dataset for the i-th monitoring module. Its weighting factor, Its timeliness factor.

8. The method according to claim 7, characterized in that, The criteria for classifying low-risk, medium-risk, and high-risk levels are as follows: When F all The value is between 0 and 30% of Max{F all When the range is within a certain range, it indicates a low-risk warning. When F all The value of Max{F} is between 30% and 60%. all When the range is within a certain range, it is considered a medium-risk warning. When F all The value of Max{F} is between 60% and 100%. all When the value is within the specified range, a high-risk warning is issued.

9. The method according to claim 1, characterized in that, The adoption of a cross-modal attention mechanism includes the feature F extracted from the wellbore leakage module. DAS Features of ground settlement module F S Calculate cross-attention: ； in, These are the query projection matrix, key projection matrix, and value projection matrix, respectively. k To query the feature dimensions of Q and key K.

10. An integrated monitoring and early warning system for underground gas storage facilities, characterized in that, include: The dynamic spatiotemporal graph construction module is configured to use each monitoring point of the underground gas storage as a node, the physical and spatiotemporal relationships between the monitoring points as edges, and introduce a time dynamic weight matrix to represent the dynamic evolution of the relationship strength, thereby constructing a dynamic spatiotemporal graph structure. The feature extraction module is configured to perform multi-scale feature extraction on the monitoring data of each monitoring point based on the dynamic spatiotemporal graph structure, including extracting local anomaly patterns using a one-dimensional convolutional neural network and integrating information of spatially adjacent nodes using graph convolution operations. The feature fusion module is configured to use a cross-modal attention mechanism to cross-fuse features extracted by different monitoring modules in order to uncover deep semantic relationships between different monitoring data. The risk assessment module is configured to perform wellbore integrity assessment, reservoir stability analysis, and leakage risk warning simultaneously through a multi-task collaborative learning module based on the fused features, generating a comprehensive risk assessment result. The early warning output module is configured to classify risks into low, medium, and high risk levels based on the comprehensive risk assessment results and output early warning signals.

Images