Distributed natural gas pipeline network safety monitoring system and method based on edge calculation

Abstract

The invention discloses a distributed natural gas pipeline network safety monitoring system and method based on edge computing, and belongs to the field of natural gas pipeline network safety monitoring, and the method comprises the steps: deploying monitoring units at a plurality of target nodes of a natural gas pipeline network, collecting the operation data of a local node by each monitoring unit, and carrying out the multi-dimensional anomaly detection through a first deep learning model; a local security risk event is identified. And the monitoring unit uploads the node operation data and the security risk event to the cloud monitoring center. And the cloud end summarizes and analyzes the data of the plurality of monitoring units to obtain global operation data of the pipe network, and identifies global safety risks through a second deep learning model. And based on the global security risk identification result, the cloud generates a corresponding control instruction and issues the control instruction to the target monitoring unit. And an execution unit connected with the target monitoring unit executes a safety action according to the control instruction. According to the method, accurate, real-time and distributed safety monitoring of the natural gas pipeline network can be realized, and the operation safety and the risk prevention and control capability of the pipeline network are improved.

Description

Technical Field

[0001] The present invention relates to the field of natural gas pipeline network security monitoring, and in particular to a distributed natural gas pipeline network security monitoring system and method based on edge computing. Background Art

[0002] With the continuous expansion of the scale of natural gas pipeline networks and the continuous increase in transmission pressure, the safe operation of pipeline networks faces many challenges. Traditional natural gas pipeline network safety monitoring methods mainly rely on manual inspections and fixed monitoring points, which have problems such as limited monitoring range, insufficient real-time performance, and delayed risk identification. It is difficult to cope with the complex and changing pipeline network operation environment and potential safety hazards.

[0003] The monitoring system in the existing technology usually adopts a centralized architecture and relies on a central control system for data processing and risk analysis. This approach has defects such as data transmission delay, network congestion and single point failure. At the same time, the traditional method for abnormal detection of pipeline nodes mainly relies on preset thresholds and simple rules, lacking the ability to deeply analyze and intelligently identify complex multi-dimensional data, and cannot effectively capture potential safety risks.

[0004] In addition, the natural gas pipeline network has a large span and complex geographical distribution, and the operating characteristics and safety risks of different nodes are different. Traditional monitoring methods are difficult to conduct accurate and personalized risk assessment and real-time monitoring based on the particularities of different nodes. Therefore, there is an urgent need for a solution to achieve distributed, intelligent, and real-time safety monitoring of natural gas pipeline networks. Summary of the invention

[0005] In view of this, an embodiment of the present invention provides a distributed natural gas pipeline network safety monitoring system and method based on edge computing to solve at least one of the above technical problems.

[0006] To achieve the above objectives, in a first aspect, a distributed natural gas pipeline network safety monitoring method based on edge computing is provided, which comprises the following steps: Monitoring units for performing edge computing are respectively set at multiple target nodes of the natural gas pipeline network, each of the monitoring units collects node operation data of a local natural gas pipeline network node, performs real-time multi-dimensional anomaly detection processing on the node operation data according to a first deep learning model, and identifies security risk events of the local natural gas pipeline network node; Each of the monitoring units uploads the safety risk events and node operation data of the local natural gas pipeline network node to the cloud monitoring center; The cloud monitoring center aggregates and analyzes the node operation data uploaded by the multiple monitoring units to obtain the global operation data of the pipeline network; obtains a global safety risk identification result based on the second deep learning model according to the global operation data of the pipeline network and the safety risk events uploaded by the multiple monitoring units; generates corresponding control instructions according to the global safety risk identification result, and sends the control instructions to one or more target monitoring units; The execution unit connected to the target monitoring unit executes the corresponding safety action according to the control instruction forwarded by the target monitoring unit.

[0007] In a second aspect, a distributed natural gas pipeline network safety monitoring system based on edge computing is provided, which includes: A plurality of monitoring units are respectively arranged at a plurality of target nodes of a natural gas pipeline network, each monitoring unit is configured with a first deep learning model, and each monitoring unit is used to collect node operation data of a local natural gas pipeline network node; perform real-time multi-dimensional anomaly detection processing on the node operation data according to the first deep learning model to identify safety risk events of the local natural gas pipeline network node; and upload the safety risk events and node operation data of the local natural gas pipeline network node to a cloud monitoring center; The cloud monitoring center is used to summarize and analyze the node operation data uploaded by the multiple monitoring units to obtain the global operation data of the pipeline network; based on the global operation data of the pipeline network and the safety risk events uploaded by the multiple monitoring units, a global safety risk identification result is obtained through a second deep learning model; a control instruction is generated according to the global safety risk identification result, and the control instruction is issued to one or more target monitoring units; An execution unit is connected to the target monitoring unit and is used to execute corresponding safety actions according to the control instructions forwarded by the target monitoring unit.

[0008] In a third aspect, an electronic device is provided, comprising: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the distributed natural gas pipeline network safety monitoring method based on edge computing as described in the first aspect.

[0009] In a fourth aspect, a computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the distributed natural gas pipeline network safety monitoring method based on edge computing as described in the first aspect is implemented.

[0010] The above technical solution has the following beneficial technical effects: The present invention realizes the distributed, intelligent and real-time safety monitoring of the natural gas pipeline network by deploying edge computing monitoring units at multiple target nodes of the natural gas pipeline network. Each monitoring unit uses the first deep learning model to perform multi-dimensional anomaly detection of local nodes, which improves the accuracy and timeliness of node-level risk identification; the cloud monitoring center uses the second deep learning model to conduct a comprehensive analysis of global operating data and risk events, and can obtain global safety risk identification results at the pipeline system level, and quickly generate targeted control instructions. Compared with traditional monitoring methods, this method has lower data transmission delay, higher risk identification intelligence level, and stronger system adaptability, which effectively improves the safety monitoring capability of the natural gas pipeline network and reduces the risk of safe operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The accompanying drawings are used to better understand the present invention and do not constitute an improper limitation of the present invention. Figure 1 It is a flow chart of a distributed natural gas pipeline network safety monitoring method based on edge computing according to an embodiment of the present invention; Figure 2 is a functional block diagram of a monitoring unit according to an embodiment of the present invention; Figure 3 is a specific flow chart of step S10 of an embodiment of the present invention; Figure 4 is a specific flow chart of step S30 of an embodiment of the present invention; Figure 5 is a flow chart of another distributed natural gas pipeline network safety monitoring method based on edge computing according to an embodiment of the present invention; Figure 6 is a functional block diagram of a multi-mode sensor module according to an embodiment of the present invention; Figure 7 It is a functional block diagram of a distributed natural gas pipeline network safety monitoring system based on edge computing according to an embodiment of the present invention; Figure 8 It is a schematic diagram of the structure of a computer system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0012] The following is a description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, including various details of the embodiments of the present invention to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0013] Embodiment 1 like Figure 1As shown, this embodiment provides a distributed natural gas pipeline network security monitoring method based on edge computing, which includes the following steps: S10: Monitoring units for performing edge computing are respectively set up at multiple target nodes of the natural gas pipeline network, each of the monitoring units collects node operation data of the local natural gas pipeline network node, performs real-time multi-dimensional anomaly detection and processing on the node operation data according to the first deep learning model, and identifies safety risk events of the local natural gas pipeline network node.

[0014] In this step, the system deploys monitoring units at key nodes of the natural gas pipeline network (such as valve stations, metering stations, distribution stations, etc.). For example, at a valve station, the monitoring unit is equipped with multi-mode sensors, including piezoelectric accelerometers, acoustic emission sensors, etc., to collect valve operation data in real time. The data collected by these sensors will be input into the first deep learning model. The first deep learning model can analyze the node operation status in real time from multiple dimensions (such as vibration characteristics, acoustic characteristics, electrical characteristics, etc.) and identify safety anomalies, such as valve mechanical wear, reduced sealing, or sudden failures.

[0015] Edge computing can process data on the device or network node closest to the user, thereby providing faster response time and better user experience. At the same time, edge computing can also reduce network load, save bandwidth resources, and improve data security, because sensitive data can be processed on edge devices without having to be transmitted to the cloud. In this embodiment, edge computing is reflected in the ability to perform local data processing and risk identification in the monitoring unit, especially through the first deep learning model to complete multi-dimensional anomaly detection at the node in real time, so as to complete the calculation on the edge side (target node) close to the data source, rather than relying on cloud processing. The advantages of edge computing are to reduce the amount of transmitted data, speed up the response speed, and realize real-time processing.

[0016] Specifically, the real-time multi-dimensional anomaly detection processing of the first deep learning model is explained in detail below.

[0017] In the data preprocessing stage, the multi-dimensional data collected by multi-mode sensors undergoes rigorous preprocessing, including data cleaning, denoising and standardization. Through sophisticated data preprocessing, noise interference is eliminated, and heterogeneous data is converted into a unified feature representation space, ensuring the quality and consistency of information generated by the subsequent fusion feature vector.

[0018] In the feature extraction stage, a hybrid architecture of Convolutional Neural Network (CNN) and Long Short-Term Memory Network (LSTM) is used to fully capture the spatial and temporal dimensional features of the data. Convolutional Neural Network (CNN) focuses on extracting spatial dimensional information of vibration, acoustic and electrical features, and captures local and global feature patterns through multi-layer convolution operations. Long Short-Term Memory Network (LSTM) analyzes the dynamic changes of time series data, learns long-term temporal dependencies through gating mechanisms, and identifies data trends and potential abnormal patterns. This multi-dimensional feature extraction method can fully capture the complex features of equipment operation data and provide a rich information basis for subsequent anomaly detection. Fusion feature vector generation is a key link connecting feature extraction and anomaly detection. Through the attention mechanism or multimodal fusion network, the spatial features extracted by CNN and the time series features captured by LSTM are deeply fused. The generated high-dimensional and semantically rich fusion feature vector comprehensively reflects the operating characteristics of the equipment in the spatial and temporal dimensions, breaking through the limitations of single-dimensional features. This composite feature representation enables the deep learning model to more accurately understand and judge the operating status of the equipment, providing a comprehensive and multi-dimensional information basis for subsequent anomaly detection.

[0019] The anomaly detection algorithm uses the fused feature vector as input to establish the normal operation standard of the multi-dimensional feature space. The anomaly score is calculated by the weighted Mahalanobis distance algorithm to quantify the degree of deviation between the real-time data and the normal operation feature space. The specific calculation formula is: , where X represents the fused feature vector, μ is the feature space mean, W is the feature weight diagonal matrix, and Σ is the covariance matrix. When the anomaly score exceeds the preset threshold, the system will determine it as a security risk event, providing data support for timely warning and risk control.

[0020] In the risk classification and assessment stage, the rich semantic information of the fused feature vector is used to achieve accurate anomaly type identification. The fused feature vector is matched with the predefined risk pattern library, and the probability distribution of each risk type is calculated using the softmax classifier. The system not only outputs the most likely risk type, but also provides a detailed risk assessment report, including risk type, impact scope, occurrence time and risk level.

[0021] To ensure the real-time and adaptability of the anomaly detection system, the first deep learning model uses a lightweight network architecture and edge computing optimization technology. The computational complexity is reduced through model compression, pruning and other methods to ensure real-time performance on edge devices. At the same time, the system has online learning capabilities, which can continuously update the judgment criteria for anomaly detection and dynamically adjust the generation of fused feature vectors and anomaly detection strategies based on new operating data. This adaptive learning mechanism ensures the long-term effectiveness of the anomaly detection system and continuously improved recognition accuracy.

[0022] S20: Each of the monitoring units uploads the safety risk events and node operation data of the local natural gas pipeline network node to the cloud monitoring center.

[0023] In this step, each monitoring unit transmits the detailed information of the safety risk events detected locally and the original operation data to the cloud monitoring center in real time through its communication module. The transmitted data can include key information such as event type, occurrence time, risk level, node location, etc. For example, if the valve station monitoring unit detects abnormal valve wear, it will immediately upload a data packet containing detailed information such as vibration data and abnormal characteristics to the cloud.

[0024] S30: The cloud monitoring center aggregates and analyzes the node operation data uploaded by the multiple monitoring units to obtain the global operation data of the pipeline network; obtains a global safety risk identification result based on the second deep learning model according to the global operation data of the pipeline network and the safety risk events uploaded by the multiple monitoring units; generates corresponding control instructions according to the global safety risk identification result, and sends the control instructions to one or more target monitoring units.

[0025] In this step, the cloud monitoring center receives data from monitoring units at different sites and processes these data comprehensively through big data analysis technology. The second deep learning model will analyze the global operation data and identify possible system-level risks. For example, if pressure anomalies or flow fluctuations occur at multiple sites at the same time, the model determines that there are potential safety hazards across the entire network. Based on the risk level and scope of impact, the system will generate specific control instructions, such as reducing pressure at local sites, launching emergency plans, or starting backup pipelines, and send them to relevant monitoring units in a targeted manner. Depending on the type and scope of impact of the risk event, the cloud monitoring center will issue control instructions to one or more relevant monitoring units. If the security risk event only involves a specific node, the control instruction is only for the monitoring unit of that node; if the security risk event is chain-like or extensible, the control instruction needs to be issued to multiple relevant monitoring units to coordinate the execution of security actions.

[0026] The first deep learning model includes an anomaly detection algorithm for processing high-dimensional spatiotemporal data, such as a detection model based on a convolutional neural network (CNN) or a recurrent neural network (RNN). The second deep learning model includes a model for complex association analysis, such as a risk identification model based on a graph neural network (GNN) or a long short-term memory network (LSTM); the first deep learning model is used in the edge computing module to process local node data, with the goal of identifying local security risks in real time, and is suitable for high-dimensional anomaly detection scenarios (such as multimodal data association analysis). The second deep learning model is used for complex risk analysis after the cloud monitoring center summarizes global data, with the goal of identifying global risks, and is suitable for spatiotemporal association modeling and fault propagation analysis of multi-node data.

[0027] S40: The execution unit connected to the target monitoring unit executes a corresponding safety action according to the control instruction forwarded by the target monitoring unit.

[0028] In this step, the execution unit is the actual operating device directly connected to the monitoring unit, such as an electric valve actuator, a pressure regulating device, etc. Upon receiving the control command issued by the cloud monitoring center, the execution unit will immediately execute the predetermined safety action. For example, if the control command requires the valve of a certain valve station to be closed urgently, the corresponding execution unit will quickly drive the valve to the specified position to achieve rapid response and risk control.

[0029] By setting up monitoring units at multiple target nodes, each monitoring unit contains independent sensors, edge computing and communication function modules. This structure reflects the characteristics of distributed deployment. Each edge computing module can further independently identify and process node security risk events locally, realizing distributed processing. Each edge computing module uploads data to the cloud monitoring center through the communication module. The cloud center combines the data of the distributed monitoring units for global analysis, thereby realizing the collaborative processing of distributed data. This distributed natural gas pipeline network safety monitoring method based on edge computing realizes real-time, accurate and comprehensive monitoring and risk management of natural gas pipeline network safety by deploying intelligent monitoring units at each key node.

[0030] Specifically, the training process of the first deep learning model includes the following steps: First, collect a large amount of historical operation data of natural gas pipeline network nodes, including normal operation data and known abnormal event data. These data need to come from multi-mode sensor modules, covering multi-dimensional indicators such as pressure, temperature, flow, vibration, etc. The data set must ensure representativeness and balance, especially for rare abnormal events, and the samples need to be expanded through data enhancement technology.

[0031] Perform comprehensive preprocessing on the collected data. This includes data cleaning, noise removal, missing value processing, and data standardization and normalization. For time series data, perform appropriate segmentation and windowing to ensure that the input data can be effectively processed by the convolutional neural network and long short-term memory network.

[0032] A hybrid deep learning model is constructed, including a convolutional neural network (CNN) and a long short-term memory network (LSTM). CNN is responsible for extracting spatial distribution features and capturing the spatial correlation and local patterns of data through convolutional layers and pooling layers. LSTM is responsible for capturing the long-term dependencies of time series and can learn the complex dynamic features of data that change over time. The two networks are integrated through a predetermined connection method to form an end-to-end anomaly detection model. Specifically, the spatial features extracted by CNN and the temporal features extracted by LSTM are connected through a cascade operation to form a unified fused feature vector, which is input into the fully connected layer for classification and anomaly detection. Alternatively, CNN and LSTM extract spatial features and temporal features respectively, and perform unified weighted fusion through the fully connected layer.

[0033] The model is trained using the prepared dataset. A supervised learning approach is used to divide the data into training, validation, and test sets. An appropriate loss function is selected, such as cross entropy loss or a custom anomaly detection loss function. The model parameters are adjusted through back propagation using optimization algorithms such as stochastic gradient descent or Adam. During the training process, the performance of the model on the validation set is continuously monitored to prevent overfitting. During the model training process, the output of the first deep learning model is a multidimensional anomaly detection result, which is specifically expressed as an anomaly probability vector for each time window or data sample. This vector contains two or more dimensions, representing abnormal states of different types or severity, such as normal, mild anomaly, moderate anomaly, severe anomaly, and dangerous anomaly. By using the Softmax activation function, the network output is converted into a probability distribution, with a probability value corresponding to each category, the sum of probabilities is 1, and the category with the highest probability is selected as the final judgment. During the training process, the loss function is responsible for calculating the difference between the model's predicted output and the true label. Through the back propagation algorithm, the network continuously adjusts the parameters to gradually reduce the deviation between the predicted output and the actual anomaly label. This process not only focuses on the classification accuracy of the model, but also focuses on the model's sensitivity and recognition ability to abnormal events.

[0034] Comprehensively evaluate the model performance on the test set, using indicators such as precision, recall, and F1 score. For high-risk scenarios such as natural gas pipeline safety monitoring, pay special attention to the model's ability to identify abnormal events. Based on the evaluation results, adjust the model architecture, add regularization terms, adjust hyperparameters, or collect more training data to improve the model's generalization ability.

[0035] Deploy the trained model to the edge computing node and integrate it into the natural gas pipeline network safety monitoring system. Establish an online learning mechanism for the model, which can continuously receive new operating data and continuously optimize the model performance through incremental learning.

[0036] like Figure 2 As shown, one monitoring unit is correspondingly arranged for each target node; the monitoring unit comprises a multi-mode sensor module, an edge computing module and a communication module; the multi-mode sensor module is used to collect node operation data of the local natural gas pipeline network node; the edge computing module is used to perform real-time multi-dimensional anomaly detection and processing on the node operation data according to the first deep learning model, and identify safety risk events of the local natural gas pipeline network node; the communication module is used for data transmission between the multi-mode sensor module and the edge computing module, and data transmission between the edge computing module and the cloud monitoring center.

[0037] Specifically, the monitoring units are installed at key nodes such as valve stations, metering stations, and distribution stations. Each monitoring unit is an independent intelligent system, including a multi-mode sensor module, an edge computing module, and a communication module.

[0038] Specifically, the multi-mode sensor module is composed of a variety of high-precision sensors, which can capture the node operation status in an all-round and multi-dimensional manner. Taking the valve station as an example, the sensors may include: piezoelectric accelerometer (accuracy ±0.1%), acoustic emission sensor (frequency response range 10kHz-1MHz), photoelectric encoder (angle resolution 0.1°) and strain gauge sensor (measurement accuracy ±0.1%). These sensors are installed on the valve actuator, valve body, rotating shaft and connection parts respectively to achieve accurate monitoring of valve mechanical wear, sealing, action response and surface deformation. The sensor can capture millisecond-level equipment status changes at a sampling frequency of 1000 times per second.

[0039] Specifically, the edge computing module is equipped with a high-performance edge computing chip with real-time deep learning reasoning capabilities. The first deep learning model built into the edge computing module is a hybrid neural network architecture that combines convolutional neural networks and long short-term memory networks. Convolutional neural networks are used to extract spatial features, such as the spatial distribution of vibration patterns; long short-term memory networks are used to capture time series features and analyze dynamic changes in equipment status. Through deep learning training of historical data, the first deep learning model can identify anomalies with millisecond response speeds, such as sudden valve jams, reduced sealing, and other potential safety risks.

[0040] Specifically, the communication module supports a variety of communication protocols and transmission methods. Internal data transmission uses high-speed industrial Ethernet or CAN bus to ensure real-time and reliable data exchange between the multi-mode sensor module and the edge computing module. External communication supports 4G / 5G cellular networks, satellite communications, and dedicated wireless private networks, with multiple backup communication paths. The communication module also has built-in data encryption and compression technology, which can achieve efficient and secure data transmission with lower data traffic under bandwidth constraints. For example, a monitoring unit can control the compressed abnormal event data packet to less than 100KB, and can quickly upload it to the cloud monitoring center even in a weak network environment.

[0041] This design ensures the real-time, intelligence and reliability of the natural gas pipeline network monitoring system, and can provide accurate and safe monitoring in complex industrial environments.

[0042] like Figure 3 As shown, step S10 may specifically include the following steps: S101: collecting node operation data of local natural gas pipeline network nodes through the multi-mode sensor module to obtain a multi-dimensional data set of the node operation data; In this step, the multi-mode sensor module comprehensively collects the operating data of the natural gas pipeline network nodes. Taking the valve station as an example, the multi-mode sensor module will synchronously collect multi-dimensional data: the piezoelectric accelerometer captures the vibration characteristics of the valve actuator (sampling frequency 1000Hz), the acoustic emission sensor records the acoustic characteristics of the valve sealing interface (frequency range 10kHz-1MHz), the photoelectric encoder monitors the valve action response angle (resolution 0.1°), and the strain gauge sensor measures the valve surface deformation (accuracy ±0.1%). These multi-dimensional data constitute a multi-dimensional data set containing four dimensions: mechanical, acoustic, angle, and stress, providing comprehensive node operation status information for subsequent analysis.

[0043] S102: Perform data optimization processing on the multidimensional data set of the node operation data to obtain optimized node operation data.

[0044] In this step, data preprocessing is first performed, including removing outliers, handling missing data, standardization, and normalization. For example, wavelet transform is used to denoise vibration data to eliminate sensor noise interference. Then, feature selection algorithms (such as principal component analysis) are applied to reduce the dimension and retain the most representative features. At the same time, adaptive filters can be used to smooth the data, remove high-frequency noise, and extract key operating features. The optimized data set is not only smaller in size, but also has a higher signal-to-noise ratio.

[0045] S103: Input the optimized node operation data into the first deep learning model, and perform real-time multi-dimensional anomaly detection processing on the multi-dimensional data set through the first deep learning model to obtain anomaly detection results for the node operation data; wherein the first deep learning model includes a convolutional neural network and a long short-term memory network; the convolutional neural network is used to extract the spatial distribution characteristics of the optimized node operation data; the long short-term memory network is used to extract the time series characteristics of the optimized node operation data.

[0046] In this step, the first deep learning model is a hybrid neural network architecture. The convolutional neural network (CNN) first processes spatial features, and extracts the spatial distribution pattern of node operation data through multiple layers of convolution and pooling layers. For example, for valve vibration data, CNN can identify the spatial distribution characteristics of different vibration intensities and directions. The long short-term memory network (LSTM) is responsible for time series feature extraction to capture the dynamic change trend of the data. LSTM can learn long-term dependencies and identify small but continuous abnormal changes. The outputs of the two networks are integrated through the fusion layer to form a multi-dimensional anomaly detection result. The model can identify potential safety risks such as valve mechanical wear and reduced sealing in real time.

[0047] S104: Based on the anomaly detection result, security risk events of local natural gas pipeline network nodes are identified in combination with an event classification rule library.

[0048] In this step, the anomaly detection results are matched with the pre-established event classification rule base. The rule base is a complex decision tree built based on a large amount of historical data and expert knowledge, containing hundreds of safety risk scenarios. For example, when CNN and LSTM detect an abnormal increase in valve vibration frequency and a sudden change in acoustic characteristics, the rule base will quickly match specific risk events such as "valve mechanical wear" or "decreased sealing". The rule base not only contains event types, but also covers multi-dimensional information such as trigger conditions and severity assessment to ensure the accuracy and comprehensiveness of risk identification.

[0049] In step S104, the safety risk event identification process is a multi-level, intelligent matching and reasoning process. First, the system takes the anomaly detection results generated by the deep learning model as input and accurately matches them through the pre-built event classification rule base. The rule base is a multi-dimensional, hierarchical knowledge graph that contains hundreds of predefined safety risk event patterns, each of which consists of multiple feature dimensions and trigger conditions. The matching process uses a weight-based fuzzy matching algorithm, which not only focuses on the precise numerical value of the anomaly detection results, but also considers the relative change trend and correlation of the features. For example, when CNN and LSTM detect an abnormal increase in the vibration frequency of the valve, a sudden change in the acoustic features, and a duration exceeding the preset threshold, the system will quickly locate possible risk event types, such as "valve mechanical wear" or "degraded sealing", through the decision tree of the rule base. The matching process introduces a probabilistic reasoning method. For abnormal features with fuzzy boundaries, the system will give multiple possible event types and their confidence levels, and sort them according to the degree of risk.

[0050] S105: Generate corresponding node security risk event description data for each security risk event, wherein the node security risk event description data includes event type, event impact scope, event occurrence time, and event risk level.

[0051] In this step, for each identified safety risk event, the system automatically generates detailed event description data. The description data not only contains basic information, but also integrates multi-dimensional analysis results. For example, for a "valve wear" event at a valve station, the description data may include: the event type is "mechanical wear", the impact range is "valve actuator", the occurrence time is accurate to milliseconds, and the risk level is divided into "medium risk" according to the degree of wear.

[0052] Specifically, node safety risk event description data generation refers to converting the raw data collected by multimodal sensors into structured and understandable safety risk event descriptions. This process is based on the anomaly detection results of the deep learning model, and through multidimensional feature analysis and rule base matching, an intelligent description model with four dimensions including event type, impact scope, occurrence time and risk level is constructed. This method can not only accurately capture the potential safety risks of natural gas pipeline network nodes, but also provide interpretable and quantifiable risk assessment information.

[0053] Specifically, event type recognition is achieved by comprehensively analyzing the anomaly detection results of the deep learning model and the dominant anomaly features of the multimodal sensor data. The recognition process first identifies possible anomaly sources based on the anomaly features of the multi-dimensional sensor data, such as vibration, acoustics, angle, and stress. Then, the abnormal event type is accurately located by matching the predefined event type rule base. The recognition process follows the priority judgment principle: first consider static feature anomalies, such as structural changes in mechanical parts; secondly, analyze dynamic change anomalies, such as parameter fluctuations; and finally evaluate cross-node propagation anomalies, such as the cascading impact of pressure or flow.

[0054] Specifically, the determination of the impact scope of an event involves an in-depth analysis of the anomaly propagation path and an accurate assessment of the coupling relationship between nodes. By constructing a topological relationship diagram of the natural gas pipeline network nodes, the system can track the path and mechanism of the spread of anomalies from the initial node to the surrounding nodes. The impact scope assessment not only considers physical connections, but also includes functional coupling and energy transfer characteristics. The system generates a dynamic and accurate impact topology map by calculating the number and range of nodes affected by the anomaly, providing comprehensive spatial dimension information for risk management.

[0055] Specifically, the precise location of the time when an event occurred is achieved through a detailed analysis of the time series of multimodal data. The system tracks the turning points of changes in each sensor data and uses millisecond-level timestamps to mark the exact moment when the anomaly occurred. This process relies on a high-precision time synchronization mechanism between sensors to ensure the accuracy of the timestamp. By constructing a time change curve of multidimensional data, the system can identify the exact time when the anomaly occurred, providing a precise time anchor point for subsequent risk analysis and tracing.

[0056] Specifically, the event risk level assessment is a comprehensive assessment process based on quantitative indicators and qualitative judgments. Quantitative indicators include the magnitude and duration of the anomaly, which are accurately quantified through mathematical models; qualitative judgments are achieved through matching and interpretation of expert knowledge bases. The risk level is divided into three levels: low risk represents minor anomalies that will not significantly affect system operation; medium risk indicates local performance degradation and requires close attention; high risk indicates serious anomalies that directly threaten system security.

[0057] This multi-dimensional, intelligent safety monitoring method makes full use of edge computing and deep learning technology to achieve the real-time, accurate and comprehensive safety monitoring of the natural gas pipeline network. The present invention improves the intelligent identification ability of the safety risks of the natural gas pipeline network nodes by introducing multi-mode sensor modules, data optimization processing, deep learning model architecture (integration of convolutional neural network and long short-term memory network) and event classification rule base in step S10. Multi-mode sensors realize multi-dimensional data acquisition, data optimization processing enhances data quality, convolutional neural network accurately extracts spatial distribution characteristics, and long short-term memory network effectively captures time series characteristics, making anomaly detection more comprehensive and accurate. Through the structured event classification rule base, not only can safety risk events be accurately identified, but also detailed event description data including event type, impact range, occurrence time and risk level can be generated, which provides a richer and more structured information basis for subsequent risk assessment and emergency disposal, and improves the intelligence level and risk prevention and control capabilities of natural gas pipeline network safety monitoring.

[0058] In some alternative embodiments, instead of using the first deep learning model, a rule-based multidimensional analysis function is used. A rule-based multidimensional analysis function is a data analysis method that combines a multidimensional data model and a rule engine to achieve complex data analysis and decision support. Multidimensional analysis is a technique for quickly processing complex queries that allows users to analyze data from multiple dimensions.

[0059] like Figure 4 As shown, in some embodiments, step S30 may specifically include the following steps: S301: The cloud monitoring center receives node operation data and corresponding security risk events uploaded by multiple monitoring units, and summarizes and analyzes the received node operation data to obtain the global operation data of the pipeline network.

[0060] Specifically, the cloud monitoring center receives real-time operation data from hundreds of monitoring units in various places through a high-performance data processing platform. These data reach hundreds of megabytes per second, covering multi-dimensional indicators such as pressure, temperature, flow, and vibration. After the data is connected, it is first structured and stored in a distributed database, and then summarized and correlated in real time through a big data analysis engine. For example, the system can quickly compare the operating parameters of valve stations in different regions, identify abnormal fluctuation trends, and generate a global operation data report of the pipeline network containing regional and global operation characteristics.

[0061] S302: The cloud-based monitoring center inputs the global operation data of the pipeline network and the safety risk events of multiple local natural gas pipeline network nodes into a second deep learning model to obtain a global safety risk identification result, wherein the global safety risk identification result includes whether there is a global safety risk, the level of the global safety risk and the scope of influence of the global safety risk.

[0062] Specifically, the second deep learning model is a multi-layer neural network for global safety risk identification. The second deep learning model combines graph neural networks (GNNs) and attention mechanisms to capture the complex correlations between pipeline network nodes. The model input includes global operating data and risk events at each node. Through multi-layer network analysis, potential systemic risks can be identified. For example, when abnormal pressures occur at valve stations in certain regions at the same time, the model can quickly determine whether this is a local problem or a global risk that may trigger a large-scale chain reaction. The risk identification results not only give whether the risk exists or not, but also evaluate the risk level (such as low, medium, high, and emergency) and the geographical scope that may be affected.

[0063] Specifically, the specific analysis process of the second deep learning model is described in detail below: (1) Data preprocessing and feature encoding stage At the beginning of the model input, the global operation data of the pipeline network and the risk events of each node are first standardized and preprocessed. For numerical indicators (such as pressure, temperature, and flow), standard scores (Z-scores) are standardized to eliminate dimension and scale differences. For discrete risk event features, one-hot encoding or embedded encoding is used to convert them into continuous feature vectors. At the same time, auxiliary information such as geographic location and pipeline network topology is spatially encoded to provide structured input for subsequent graph neural networks.

[0064] (2) Graph Neural Network Topological Correlation Analysis Graph Neural Network (GNN) is used to capture the complex associations between pipeline network nodes. First, an association graph of pipeline network nodes is constructed. Nodes represent specific valve stations or monitoring units, and edges represent physical connections, geographical proximity, or functional associations between nodes. Through the graph convolution layer, the model can aggregate feature information of neighboring nodes and learn implicit associations between nodes. For example, when a node has abnormal pressure, GNN can analyze its potential impact on surrounding nodes and identify possible cascading risk transmission paths.

[0065] (3) Dynamic feature weighting of attention mechanism The multi-head attention mechanism is introduced to dynamically learn the importance of different features and nodes. For the multi-dimensional features of the input, the attention mechanism can adaptively assign weights to highlight key information. Specifically, the model will learn the weights of different types of risk events and operating indicators in the global risk assessment. For example, pressure anomalies have a higher weight than temperature fluctuations. Multi-head attention allows the model to learn the correlation of features from different subspaces and improve the ability to model complex relationships.

[0066] (4) Multi-layer neural network risk reasoning Through stacked fully connected layers and nonlinear activation functions, the model performs deep nonlinear feature transformation and risk reasoning. These hierarchical network structures enable the model to gradually abstract high-level risk features from low-level features. Each layer of the network learns more complex and abstract feature representations, and the final output layer is activated by an activation function (softmax or sigmoid) to generate prediction results of risk probability, risk level, and impact range (please use a paragraph to explain in detail how to generate prediction results of risk probability, risk level, and impact range).

[0067] In a multi-layer neural network, through stacked fully connected layers, neurons in each layer receive feature representations from the previous layer and implement complex nonlinear feature transformations through nonlinear activation functions (Rectified Linear Unit, ReLU). This layer-by-layer feature abstraction process can extract deep information related to risks, such as specific patterns or feature associations of equipment anomalies. In the output layer, an activation function (Softmax or Sigmoid) is used to convert the final output of the network into an interpretable probability value: the probability of risk existence indicates the possibility of detecting a certain risk (the category probability in the classification problem); the risk level further determines the most likely category of different risk levels (such as mild, moderate, and severe) through the output multidimensional probability distribution; the impact range is combined with the key information in the feature vector and obtained by linearly combining the weights of the fully connected layer as the final regression prediction value of the network, indicating the equipment or area that the risk may affect.

[0068] (5) Risk assessment and uncertainty quantification In addition to outputting specific risk identification results, the model also introduces uncertainty quantification methods. Through ensemble learning or Bayesian neural network methods, the confidence of the model prediction can be estimated. For some complex or ambiguous risk scenarios, the model can give an uncertainty assessment of the risk prediction, providing decision makers with a more comprehensive basis for risk judgment.

[0069] S303: When a global security risk is identified, the cloud monitoring center determines the security execution measures according to the level of the global security risk, the impact scope of the global security risk, the global operation data of the pipeline network and the cloud event decision rule library.

[0070] Specifically, the cloud-based event decision rule library integrates the expert knowledge of multiple fields such as natural gas pipeline operation, safety, and geology. When a global safety risk is detected, the system will quickly match the most suitable safety execution measures according to the risk level and scope of impact. For example, for a cross-regional medium-risk pressure abnormality event, the system can formulate the following measures: start the staged pressure reduction process in the affected area; dispatch backup gas transmission lines; notify the gas dispatch center in cities along the line; and initiate emergency repair plans. The decision-making process takes into account multiple dimensions, including equipment status, geographical location, meteorological conditions, backup resources, etc.

[0071] Specifically, in step S303, when the cloud monitoring center identifies a global security risk, it first determines the safety execution measures according to the level and impact range of the global security risk, the global operation data of the pipeline network, and the cloud event decision rule library. The specific implementation process is: after the cloud monitoring center receives the global security risk identification result, it searches for the relevant rules in the cloud event decision rule library. The cloud event decision rule library stores different emergency response measures according to different risk levels and impact ranges. For example, for high-risk and wide-ranging events, the cloud monitoring center can choose high-priority safety measures such as emergency shutdown, closing important valves, or emergency decompression; for medium-risk events, it can choose lower-priority measures such as adjusting pressure or flow; and low-risk and local events only need local inspection or adjustment. By analyzing the global operation data of the pipeline network, the cloud monitoring center further refines the decision, such as adjusting the pressure or flow of the pipeline network node, to ensure the rationality of the decision. Subsequently, the cloud monitoring center generates a control instruction according to the determined safety execution measures, the instruction includes the safety action to be executed, the execution priority, and the identification of the target monitoring unit, and sends the control instruction to the corresponding target monitoring unit, requiring it to execute the corresponding safety measures.

[0072] S304: The cloud monitoring center generates a control instruction according to the security execution measure, wherein the control instruction includes the security action to be executed, the execution priority and the identification of the target monitoring unit, and sends the control instruction to the corresponding one or more target monitoring units.

[0073] Specifically, each control instruction is a highly structured data packet, which contains clear safety action instructions, strict execution priority and target monitoring unit identification. For example, a control instruction is as follows: the safety action is "pressure reduction 20%", the execution priority is "high", and the target monitoring unit is "A03 valve station". The instructions are transmitted through multiple encrypted channels to ensure communication security. After receiving the instruction, the target monitoring unit will immediately execute the corresponding safety action according to the instruction, and send the execution feedback back to the cloud monitoring center in real time.

[0074] This distributed natural gas pipeline network safety monitoring method based on edge computing and deep learning realizes intelligent and precise risk management from local to global, and improves the safety and emergency response capabilities of natural gas pipeline network operation. The method of the embodiment of the present invention builds a comprehensive pipeline network operation data view by aggregating node operation data of multiple monitoring units, breaking the information barriers of traditional monitoring and providing a more accurate data basis for risk analysis. The introduced second deep learning model can intelligently and finely identify global security risks, not only accurately judge the existence of risks, but also carefully evaluate the risk level and impact range. The cloud event decision rule library supports the rapid formulation of differentiated security execution measures according to risk characteristics, generates precise control instructions including specific security actions, execution priorities and target unit identifiers, and realizes intelligent regulation of risk response. This method shortens the response time from risk identification to disposal through efficient collaboration of the cloud center, and has good system scalability and continuous learning capabilities, and can dynamically adapt to new security challenges. The system transforms from passive response to active early warning, improving the safety monitoring efficiency of the natural gas pipeline network.

[0075] Specifically, the training process of the second deep learning model is described below: In the data preprocessing stage, the topological graph structure of the natural gas pipeline network is first constructed. Each pipeline node (such as valve station, metering station) is regarded as a node in the graph, and the physical connection, data flow path and geographical proximity between nodes are used as the weight of the edge. Data preprocessing not only includes data cleaning and standardization, but also enrichment of node features. In addition to basic sensor data, auxiliary information such as historical risk events, geographic location, pipeline diameter, etc. is also integrated to provide multi-dimensional and high-quality feature representation for subsequent graph neural network learning.

[0076] In the design phase of the graph neural network architecture, the model architecture adopts the Graph Attention Network (GAT), which is constructed by multiple layers of graph attention layers. The first layer of graph attention layer focuses on local feature extraction, capturing abnormal patterns of a single node and its direct neighbors; subsequent layers gradually expand the receptive field and learn a wider range of correlations. The attention mechanism allows the importance of different nodes and different types of risk events to be learned dynamically rather than fixed in advance. The last part of the network is connected to the fully connected layer and the multi-classification output layer to achieve accurate identification of global risks. This architecture enables the model to intelligently understand the complex interactions between network nodes and evaluate potential risks from a global perspective.

[0077] During the model training phase, the loss function of the model is designed using a multi-task learning paradigm to simultaneously optimize three key subtasks: risk existence classification, risk level prediction, and impact range estimation. The loss function consists of three parts: cross entropy loss for risk existence, ranking-sensitive loss for risk level, and regression loss for geographical impact range. By jointly optimizing these three subtasks, the model is able to learn a richer and more coherent risk representation.

[0078] The training process adopts a hierarchical learning strategy. First, pre-training is performed on a simulated pipeline network dataset to capture basic risk patterns; then, detailed fine-tuning is performed on real historical data. To prevent overfitting, a variety of regularization techniques are introduced, including random inactivation of graph structures, L2 regularization of attention weights, and gradient clipping of model parameters. The learning rate adopts an adaptive adjustment strategy, combining learning rate warm-up and exponential decay to balance the learning speed and training stability of the model.

[0079] like Figure 5 As shown, the method further comprises the following steps: S50: When the event risk level of a security risk event at a local natural gas pipeline node exceeds a preset level, the edge computing module generates an emergency response signal based on the event risk level and the local risk response rule base, and controls the execution unit to perform a security action corresponding to the emergency response signal.

[0080] Specifically, the local risk response rule base is an intelligent and refined decision-making system that presets multi-level and multi-dimensional response strategies for various risk scenarios that may occur at natural gas pipeline nodes. When the risk level of a node safety risk event exceeds the preset threshold, the edge computing module will immediately activate the emergency response mechanism. Taking the valve station as an example, the risk level is divided into four levels: level 1 (low risk), level 2 (medium risk), level 3 (high risk) and level 4 (emergency risk).

[0081] When the risk level reaches level three or level four, the edge computing module will quickly generate an emergency response signal based on the local risk response rule base. For example, for the scenario of "abnormal valve sealing and pressure fluctuation exceeding 15%", the system can generate the following emergency response signals: immediately start the valve automatic isolation program, trigger the local sound and light alarm system, start the backup gas pipeline, reduce the pipeline network operating pressure to below the safety threshold, and automatically record and upload a detailed risk event log.

[0082] After receiving the emergency response signal, the execution unit will respond quickly according to the preset safety action process. This intelligent local risk response mechanism can quickly identify and accurately control risks within milliseconds, minimizing the possibility and potential impact of safety accidents.

[0083] The generation process of emergency response signals involves multiple intelligent algorithms and decision-making logic, including real-time risk assessment, risk level matching, action choreography, and multiple verifications. The system continuously analyzes node operation data based on a deep learning model, accurately matches the detected risk features with the risk scenarios in the rule base, and automatically choreographs the optimal safety response action sequence based on the risk level and specific scenario.

[0084] The construction of the local risk response rule base is based on historical risk event data, expert experience knowledge, physical model simulation and machine learning algorithms. Through continuous learning and iterative updates, the local risk response rule base can continuously improve the accuracy of risk identification and response. The system also supports remote dynamic updates to ensure that it is always in the latest security protection status.

[0085] This intelligent local risk response method based on edge computing has realized the transition from passive response to active defense in natural gas pipeline network safety monitoring. Through millisecond-level risk identification and precise control, potential safety hazards can be effectively prevented, improving the safety and reliability of pipeline network operation.

[0086] like Figure 6 As shown, the multiple target nodes specifically include: valve stations, metering stations, distribution stations, compressor stations and pressure regulating stations of the natural gas pipeline network. In a specific embodiment of the present invention, multiple target nodes are key monitoring points in the natural gas pipeline network. These nodes include valve stations, metering stations, distribution stations, compressor stations and pressure regulating stations. For example, the valve station is responsible for the opening and closing control of the pipeline, and can quickly isolate the pipe section in an emergency; the metering station is responsible for accurately measuring the natural gas flow and composition; the distribution station realizes the precise distribution of natural gas; the compressor station provides gas transmission power to ensure the long-distance transportation of natural gas; the pressure regulating station ensures that the pipeline network pressure is maintained within a safe and efficient range. This multi-node distributed architecture can fully cover the key operating links of the natural gas pipeline network and provide all-round safety monitoring.

[0087] In some embodiments, the multi-mode sensor module may include: The multi-mode sensor group corresponding to the valve station includes: a piezoelectric acceleration sensor, installed at the connection between the valve actuator and the valve stem, used to obtain the mechanical wear vibration characteristic data of the valve of the valve station; an acoustic emission sensor, installed at the sealing interface between the valve body and the valve seat, used to obtain the sealing electro-acoustic coupling characteristic data of the valve of the valve station; a photoelectric encoder, installed on the rotating shaft of the valve actuator, used to obtain the action response characteristic data of the valve of the valve station; a strain gauge sensor, installed at the connection between the valve stem and the valve disc, used to obtain the surface deformation characteristic data of the valve of the valve station; The multi-mode sensor group corresponding to the metering station includes: a spectral analysis sensor, which is installed on the bypass sampling pipeline of the metering station pipeline and is located on the downstream side of the sampling valve, and is used to obtain the medium composition in the metering station pipeline; an ultrasonic flow sensor, which is installed on the straight section of the metering station pipeline, and is used to obtain the disturbance acoustic characteristic data of the fluid in the metering station pipeline; an electromagnetic interference detection sensor, which is installed in the electrical connection terminal area of ​​the metering station control cabinet housing, and is used to obtain the electromagnetic interference characteristic data of the metering station pipeline.

[0088] Specifically, in the multi-mode sensor group of the valve station, the piezoelectric accelerometer is installed at the connection between the valve actuator and the valve stem. By capturing tiny vibration signals, the degree of mechanical wear of the valve can be analyzed. For example, when the valve is in long-term operation, abnormal changes in the vibration frequency and amplitude at the connection indicate wear or impending failure of the mechanical parts. The acoustic emission sensor is installed at the sealing interface between the valve body and the valve seat, and is mainly used to monitor the sealing performance of the valve. By capturing the electroacoustic coupling characteristics of the sealing interface, tiny leakage sounds or degradation of sealing performance can be detected. For example, when the sealing performance of the valve deteriorates, the sensor can capture acoustic signals of specific frequencies and intensities, and warn in advance that the seal needs to be replaced. The photoelectric encoder is installed on the rotating shaft of the valve actuator to obtain the action response characteristics of the valve. It can accurately record the angle, speed and response time of the valve opening and closing. For example, if the opening and closing time of the valve is abnormally delayed or irregularly changed, it indicates that there is mechanical stagnation in the actuator or a problem with the electrical control system. The strain gauge sensor is installed at the connection between the valve stem and the valve disc, and mainly monitors the structural deformation characteristics. By measuring the strain data at the joint, the structural integrity and stress state of the valve can be evaluated. For example, abnormal strain data indicates that the valve is subjected to mechanical stress beyond the design range, posing a potential structural safety hazard.

[0089] In the multi-mode sensor group of the metering station, the spectral analysis sensor is installed on the downstream side of the pipeline bypass sampling line. It can analyze the medium composition of natural gas in real time and detect whether there are impurities, moisture or other abnormal components. For example, if hydrogen sulfide, moisture or other gas components that should not be present are detected, an alarm can be immediately issued to prevent unqualified natural gas from entering the transmission system. The ultrasonic flow sensor is installed on the straight section of the pipeline to accurately monitor the flow change and pipeline status by obtaining the disturbance acoustic characteristic data of the fluid. For example, sudden flow fluctuations or irregular acoustic characteristics indicate pipeline blockage, leakage or other abnormal conditions, providing a basis for timely preventive maintenance. The electromagnetic interference detection sensor is installed in the electrical connection terminal area of ​​the control cabinet housing to obtain electromagnetic interference characteristic data. It can monitor the electrical safety status of the control system. For example, if an abnormal electromagnetic interference signal is detected, it indicates that there is a fault in electrical equipment or external electromagnetic interference nearby, affecting the normal operation of the metering station.

[0090] In some embodiments, the multi-mode sensor module further comprises: The multi-mode sensor group corresponding to the sub-transmission station includes: an eddy current sensor installed at a fluid diameter reducing section or elbow of a branch of a pipeline network of the sub-transmission station, used to obtain flow eddy current characteristic data of the branch of the pipeline network; a thermocouple installed on the outer surface of the pipe wall of the branch of the pipeline network of the sub-transmission station and located in a heated area, used to obtain thermodynamic coupling characteristic data of the branch of the pipeline network; a pressure spectrum analysis sensor installed on a straight section of a branch of the pipeline network of the sub-transmission station, used to obtain pressure fluctuation spectrum characteristic data of the branch of the pipeline network; a fiber grating strain sensor installed at a connection of a branch of the pipeline network of the sub-transmission station, used to obtain real-time stress and strain data at the connection of the pipeline network; The multi-mode sensor group corresponding to the compressor station includes: an acoustic emission sensor installed on the surface of the compressor blade or the area around the compressor blade, for obtaining the acoustic characteristic data of the wear of the compressor blade; a vibration acceleration sensor installed on the compressor bearing support seat or near the compressor bearing, for obtaining the vibration spectrum characteristic data of the compressor bearing; an electromagnetic field strength sensor installed on the compressor rotor casing or the area around the compressor rotor, for obtaining the electromagnetic radiation characteristic data of the compressor rotor; a flow sensor installed on the straight section of the compressor inlet and outlet pipes, for obtaining the flow characteristic data of the medium inside the compressor.

[0091] Specifically, in the multi-mode sensor group of the distribution station, the eddy current sensor is installed at the fluid reducer or elbow of the pipeline branch, which is specially used to capture the eddy current characteristics during the fluid flow process. For example, when natural gas flows at the elbow of the pipeline, complex eddy current patterns are generated. These eddies cause pipeline vibration, pressure fluctuations and even local corrosion. By accurately measuring the eddy current characteristics, the stability of pipeline flow and potential structural stress can be evaluated.

[0092] Specifically, thermocouples are installed in the heated areas of the outer surface of the pipe wall of the branch pipe network, mainly monitoring the thermodynamic coupling characteristics of the pipeline. For example, under direct sunlight or drastic changes in ambient temperature, the pipeline will produce thermal expansion and temperature gradients. Thermocouples can monitor these temperature changes in real time, help predict thermal stress and material fatigue risks, and ensure the safe operation of the pipeline under extreme temperature conditions.

[0093] Specifically, the pressure spectrum analysis sensor is installed on the straight section of the pipeline branch to obtain the spectrum characteristic data of pressure fluctuations. For example, natural gas will have pulsating pressure during transportation, and this pressure fluctuation may be caused by the compressor, valve opening and closing, or pipeline geometry changes. By analyzing the pressure spectrum, abnormal pressure fluctuation patterns can be identified to warn of pipeline vibration, resonance, or potential structural damage risks.

[0094] Specifically, fiber Bragg grating strain sensors are installed at the joints of pipe network branches to monitor stress and strain data at the joints in real time. For example, pipe joints are key locations where stress is concentrated, and local strain is generated due to welding defects, material inhomogeneity, or external loads. Fiber Bragg grating sensors can accurately measure tiny strain changes, detect potential crack initiation or structural weakening in a timely manner, and prevent serious pipeline failures.

[0095] Specifically, in the multi-mode sensor group of the compressor station, the acoustic emission sensor is installed on the surface of the compressor blade or the surrounding area, specifically used to obtain the acoustic characteristics of blade wear. For example, as the compressor runs for a long time, the blade will experience severe mechanical wear, corrosion or microcracks. The acoustic emission sensor can capture the tiny acoustic emission signals generated on the blade surface. These signals can reflect the degree of damage and development trend of the blade, providing key information for predictive maintenance.

[0096] Specifically, the vibration acceleration sensor is installed near the compressor bearing support or bearing to obtain the vibration spectrum characteristic data of the bearing. For example, the bearing is a key component of the compressor, and its health directly affects the operating reliability of the equipment. By analyzing the vibration spectrum of the bearing, problems such as bearing wear, imbalance, poor centering or insufficient lubrication can be identified, and maintenance can be carried out in time to avoid serious equipment failures.

[0097] Specifically, the electromagnetic field strength sensor is installed on the compressor rotor housing or surrounding area to monitor the electromagnetic radiation characteristic data of the rotor. For example, the rotor generates an abnormal electromagnetic field during high-speed rotation, which is a sign of bearing damage, winding failure or electrical system imbalance. By monitoring the electromagnetic field strength and distribution in real time, potential electrical faults can be detected early to prevent serious equipment damage.

[0098] Specifically, the flow sensor is installed on the straight section of the compressor inlet and outlet pipes to obtain the flow characteristic data of the medium inside the compressor. For example, the performance of the compressor depends largely on the flow characteristics of the fluid inside it. By accurately measuring the flow, velocity and pressure changes at the inlet and outlet, the working efficiency of the compressor can be evaluated, throttling or blockage problems can be detected, and the operating parameters of the compressor can be optimized.

[0099] In some embodiments, the multi-mode sensor module comprises: The multi-mode sensor group corresponding to the pressure regulating station includes: a displacement sensor installed on the valve stem of the pressure regulating valve, used to obtain dynamic response characteristic data of the pressure regulating valve in the pressure regulating station; a thermodynamic sensor installed in the expansion section, the downstream section of the regulating valve, the expansion section and the elbow area of ​​the gas flow pipeline inside the pressure regulating station, used to obtain the characteristic data of gas expansion entropy change inside the pressure regulating station; an electromagnetic shielding effectiveness sensor installed in the outer shell or shielding cover of the pressure regulating station, used to obtain the characteristic data of electromagnetic shielding effectiveness of the pressure regulating station; an acoustic resonance sensor installed on the pipeline wall or valve body surface before and after the pressure regulating valve, used to obtain the pressure gradient acoustic resonance characteristic data of the pressure regulating valve in the pressure regulating station.

[0100] Specifically, in the multi-mode sensor group of the pressure regulating station, the displacement sensor is installed on the valve stem of the pressure regulating valve to obtain the dynamic response characteristic data of the pressure regulating valve. For example, the pressure regulating valve needs to accurately control the valve opening to adjust the natural gas pressure during operation. The displacement sensor can monitor the tiny displacement changes of the valve stem in real time, including the opening angle, response speed and position stability. By analyzing these dynamic response characteristics, the control accuracy and mechanical wear degree of the valve can be evaluated, and valve failure can be predicted to ensure the safe and stable operation of the pressure regulating station.

[0101] Specifically, thermodynamic sensors are installed in the expansion section, downstream section of the regulating valve, expansion section and elbow area of ​​the gas flow pipeline inside the pressure regulating station to obtain the entropy change characteristic data of gas expansion inside the pressure regulating station. For example, when natural gas passes through the pressure regulating valve, it will experience obvious pressure changes and temperature changes, resulting in complex thermodynamic processes. Thermodynamic sensors can accurately capture the entropy change characteristics of the gas expansion process, including temperature gradient, pressure change rate and local heat exchange characteristics. These data can not only help optimize the energy efficiency of the pressure regulating station, but also warn of structural thermal stress and material fatigue risks.

[0102] Specifically, the electromagnetic shielding effectiveness sensor is installed on the casing or shielding cover of the pressure regulating station to obtain the electromagnetic shielding effectiveness characteristic data of the pressure regulating station. For example, as a key natural gas pipeline network equipment, the pressure regulating station needs to prevent the influence of external electromagnetic interference on the control system. The electromagnetic shielding effectiveness sensor can monitor the electromagnetic shielding performance of the casing or shielding cover in real time, including shielding attenuation, electromagnetic field penetration and frequency response characteristics. By analyzing these data, the electromagnetic protection capability of the pressure regulating station can be evaluated, the deterioration of shielding materials or installation defects can be discovered in time, and the electromagnetic compatibility and reliability of the control system can be ensured.

[0103] Specifically, the acoustic resonance sensor is installed on the pipeline wall or valve body surface before and after the pressure regulating valve to obtain the pressure gradient acoustic resonance characteristic data of the pressure regulating valve in the pressure regulating station. For example, when natural gas passes through the pressure regulating valve, complex pressure gradients and acoustic oscillations are generated. The acoustic resonance sensor can accurately capture the acoustic resonance characteristics caused by these pressure changes, including resonance frequency, amplitude, and attenuation characteristics. By analyzing these acoustic resonance data, the fluid dynamics characteristics of the pressure regulating station can be evaluated, and the risks of pressure fluctuations, pipeline vibrations or resonances can be identified, providing key information for optimizing the operating parameters of the pressure regulating station and preventing potential failures. Embodiment 2

[0104] like Figure 7 As shown, an embodiment of the present invention also provides a distributed natural gas pipeline network safety monitoring system based on edge computing, which includes: A plurality of monitoring units are respectively arranged at a plurality of target nodes of a natural gas pipeline network, each monitoring unit is configured with a first deep learning model, and each monitoring unit is used to collect node operation data of a local natural gas pipeline network node; perform real-time multi-dimensional anomaly detection processing on the node operation data according to the first deep learning model to identify safety risk events of the local natural gas pipeline network node; and upload the safety risk events and node operation data of the local natural gas pipeline network node to a cloud monitoring center; The cloud monitoring center is used to summarize and analyze the node operation data uploaded by multiple monitoring units to obtain the global operation data of the pipeline network; based on the global operation data of the pipeline network and the safety risk events uploaded by multiple monitoring units, a global safety risk identification result is obtained through a second deep learning model; a control instruction is generated according to the global safety risk identification result, and the control instruction is issued to one or more target monitoring units; The execution unit is connected to the target monitoring unit and is used to execute corresponding safety actions according to the control instructions forwarded by the target monitoring unit.

[0105] In some embodiments, a monitoring unit is correspondingly set for each target node; the monitoring unit includes: Multi-mode sensor module, used to collect node operation data of local natural gas pipeline network nodes; An edge computing module, configured to perform real-time multi-dimensional anomaly detection processing on the node operation data according to the first deep learning model, and identify safety risk events of local natural gas pipeline network nodes; The communication module is used for data transmission between the multi-mode sensor module and the edge computing module, and data transmission between the edge computing module and the cloud monitoring center.

[0106] In some embodiments, the edge computing module is specifically configured as follows: Collect node operation data of local natural gas pipeline network nodes through the multi-mode sensor module to obtain a multi-dimensional data set of node operation data; Performing data optimization processing on the multidimensional data set of the node operation data to obtain optimized node operation data; The optimized node operation data is input into a first deep learning model, and the multidimensional data set is subjected to real-time multidimensional anomaly detection processing by the first deep learning model to obtain an anomaly detection result for the node operation data; wherein the first deep learning model includes a convolutional neural network and a long short-term memory network; the convolutional neural network is used to extract the spatial distribution characteristics of the optimized node operation data; and the long short-term memory network is used to extract the time series characteristics of the optimized node operation data; Based on the anomaly detection results, security risk events of local natural gas pipeline network nodes are identified in combination with an event classification rule base; Generate corresponding node security risk event description data for each security risk event.

[0107] In some embodiments, the cloud monitoring center is specifically configured as follows: Receive node operation data and corresponding safety risk events uploaded by multiple monitoring units, and summarize and analyze the received node operation data to obtain the global operation data of the pipeline network; Inputting the global operation data of the pipeline network and the safety risk events of multiple local natural gas pipeline network nodes into a second deep learning model to obtain a global safety risk identification result; When a global security risk is identified, safety enforcement measures are determined based on the level of the global security risk, the scope of impact, the global operation data of the pipeline network, and the cloud event decision rule library; A control instruction is generated according to the security execution measure, and the control instruction is sent to one or more corresponding target monitoring units.

[0108] In some embodiments, the plurality of target nodes include: a valve station, a metering station, a distribution station, a compressor station, and a pressure regulating station of a natural gas pipeline network.

[0109] In some embodiments, the multi-mode sensor module includes dedicated sensor groups for different pipe network nodes.

[0110] In some embodiments, the multi-mode sensor module comprises: The multi-mode sensor group corresponding to the valve station includes: a piezoelectric acceleration sensor, installed at the connection between the valve actuator and the valve stem, used to obtain the mechanical wear vibration characteristic data of the valve of the valve station; an acoustic emission sensor, installed at the sealing interface between the valve body and the valve seat, used to obtain the sealing electro-acoustic coupling characteristic data of the valve of the valve station; a photoelectric encoder, installed on the rotating shaft of the valve actuator, used to obtain the action response characteristic data of the valve of the valve station; a strain gauge sensor, installed at the connection between the valve stem and the valve disc, used to obtain the surface deformation characteristic data of the valve of the valve station; The multi-mode sensor group corresponding to the metering station includes: a spectral analysis sensor, which is installed on the bypass sampling pipeline of the metering station pipeline and is located on the downstream side of the sampling valve, and is used to obtain the medium composition in the metering station pipeline; an ultrasonic flow sensor, which is installed on the straight section of the metering station pipeline, and is used to obtain the disturbance acoustic characteristic data of the fluid in the metering station pipeline; an electromagnetic interference detection sensor, which is installed in the electrical connection terminal area of ​​the metering station control cabinet housing, and is used to obtain the electromagnetic interference characteristic data of the metering station pipeline.

[0111] In some embodiments, the multi-mode sensor module comprises: The multi-mode sensor group corresponding to the sub-transmission station includes: an eddy current sensor installed at a fluid diameter reducing section or elbow of a branch of a pipeline network of the sub-transmission station, used to obtain flow eddy current characteristic data of the branch of the pipeline network; a thermocouple installed on the outer surface of the pipe wall of the branch of the pipeline network of the sub-transmission station and located in a heated area, used to obtain thermodynamic coupling characteristic data of the branch of the pipeline network; a pressure spectrum analysis sensor installed on a straight section of a branch of the pipeline network of the sub-transmission station, used to obtain pressure fluctuation spectrum characteristic data of the branch of the pipeline network; a fiber grating strain sensor installed at a connection of a branch of the pipeline network of the sub-transmission station, used to obtain real-time stress and strain data at the connection of the pipeline network; The multi-mode sensor group corresponding to the compressor station includes: an acoustic emission sensor installed on the surface of the compressor blade or the area around the compressor blade, for obtaining the acoustic characteristic data of the wear of the compressor blade; a vibration acceleration sensor installed on the compressor bearing support seat or near the compressor bearing, for obtaining the vibration spectrum characteristic data of the compressor bearing; an electromagnetic field strength sensor installed on the compressor rotor casing or the area around the compressor rotor, for obtaining the electromagnetic radiation characteristic data of the compressor rotor; a flow sensor installed on the straight section of the compressor inlet and outlet pipes, for obtaining the flow characteristic data of the medium inside the compressor.

[0112] In some embodiments, the multi-mode sensor module comprises: The multi-mode sensor group corresponding to the pressure regulating station includes: a displacement sensor installed on the valve stem of the pressure regulating valve, used to obtain dynamic response characteristic data of the pressure regulating valve in the pressure regulating station; a thermodynamic sensor installed in the expansion section, the downstream section of the regulating valve, the expansion section and the elbow area of ​​the gas flow pipeline inside the pressure regulating station, used to obtain the characteristic data of gas expansion entropy change inside the pressure regulating station; an electromagnetic shielding effectiveness sensor installed in the outer shell or shielding cover of the pressure regulating station, used to obtain the characteristic data of electromagnetic shielding effectiveness of the pressure regulating station; an acoustic resonance sensor installed on the pipeline wall or valve body surface before and after the pressure regulating valve, used to obtain the pressure gradient acoustic resonance characteristic data of the pressure regulating valve in the pressure regulating station.

[0113] In some embodiments, the edge computing module is also configured to: when the event risk level of a security risk event of a local natural gas pipeline node exceeds a preset level, generate an emergency response signal based on the event risk level and a local risk response rule base, and control the execution unit to perform a security action corresponding to the emergency response signal.

[0114] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0115] An embodiment of the present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the program is executed by a processor, the distributed natural gas pipeline network safety monitoring method based on edge computing as described above is implemented.

[0116] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0117] The present invention also provides an electronic device. The electronic device of an embodiment of the present invention includes: one or more processors; a storage device for storing one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors implement the distributed natural gas pipeline network safety monitoring method based on edge computing provided by the present invention.

[0118] Reference below Figure 8 , which shows a schematic diagram of the structure of a computer system 800 of an electronic device suitable for implementing an embodiment of the present invention. Figure 8 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0119] like Figure 8 As shown, the computer system 800 includes a central processing unit (CPU) 801, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 802 or a program loaded from a storage part 808 to a random access memory (RAM) 803. In the RAM 803, various programs and data required for the operation of the computer system 800 are also stored. The CPU 801, the ROM 802, and the RAM 803 are connected to each other through a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0120] The following components are connected to the I / O interface 805: an input section 806 including a keyboard, a mouse, etc.; an output section 807 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 808 including a hard disk, etc.; and a communication section 809 including a network interface card such as a LAN card, a modem, etc. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the I / O interface 805 as needed. A removable medium 811, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 810 as needed so that a computer program read therefrom is installed into the storage section 808 as needed.

[0121] In particular, according to the embodiments disclosed in the present invention, the process described in the main step diagram above can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the main step diagram. In the above embodiment, the computer program can be downloaded and installed from the network through the communication part 809, and / or installed from the removable medium 811. When the computer program is executed by the central processing unit 801, the above functions defined in the system of the present invention are executed.

[0122] It should be noted that the computer-readable medium shown in the present invention may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0123] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may occur depending on design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A distributed natural gas pipeline network security monitoring method based on edge computing, characterized in that: The following steps are involved: S10: Monitoring units for performing edge computing are respectively set at multiple target nodes of the natural gas pipeline network, each of the monitoring units collects node operation data of a local natural gas pipeline network node, performs real-time multi-dimensional anomaly detection processing on the node operation data according to a first deep learning model, and identifies security risk events of the local natural gas pipeline network node; S20: Each of the monitoring units uploads the safety risk events and node operation data of the local natural gas pipeline network node to the cloud monitoring center; S30: The cloud monitoring center aggregates and analyzes the node operation data uploaded by the plurality of monitoring units to obtain the global operation data of the pipe network; Obtaining a global safety risk identification result based on the second deep learning model according to the global operation data of the pipeline network and the safety risk events uploaded by the plurality of monitoring units; Generate corresponding control instructions according to the global security risk identification result, and send the control instructions to one or more target monitoring units; S40: The execution unit connected to the target monitoring unit executes a corresponding safety action according to the control instruction forwarded by the target monitoring unit.

2. The distributed natural gas pipeline network safety monitoring method based on edge computing according to claim 1 is characterized in that: The monitoring unit includes a multi-mode sensor module, an edge computing module and a communication module; The multi-mode sensor module is used to collect node operation data of the local natural gas pipeline network node; The edge computing module is used to perform real-time multi-dimensional anomaly detection processing on the node operation data according to the first deep learning model to identify safety risk events of the local natural gas pipeline network node; The communication module is used for data transmission between the multi-mode sensor module and the edge computing module, and data transmission between the edge computing module and the cloud monitoring center.

3. The distributed natural gas pipeline network safety monitoring method based on edge computing according to claim 2 is characterized in that: Step S10 specifically includes: S101: collecting node operation data of local natural gas pipeline network nodes through the multi-mode sensor module to obtain a multi-dimensional data set of the node operation data; S102: performing data optimization processing on the multidimensional data set of the node operation data to obtain optimized node operation data; S103: Input the optimized node operation data into a first deep learning model, and perform real-time multi-dimensional anomaly detection processing on the multi-dimensional data set through the first deep learning model to obtain an anomaly detection result for the node operation data; wherein the first deep learning model includes a convolutional neural network and a long short-term memory network; the convolutional neural network is used to extract the spatial distribution characteristics of the optimized node operation data; and the long short-term memory network is used to extract the time series characteristics of the optimized node operation data; S104: Based on the anomaly detection result, identify the safety risk events of the local natural gas pipeline network nodes in combination with the event classification rule base; S105: Generate corresponding node security risk event description data for each of the security risk events, wherein the node security risk event description data includes event type, event impact scope, event occurrence time, and event risk level.

4. The distributed natural gas pipeline network safety monitoring method based on edge computing according to claim 1 is characterized in that: Step S30 specifically includes: S301: The cloud monitoring center receives the node operation data and corresponding security risk events uploaded by each monitoring unit, and summarizes and analyzes the received node operation data to obtain the global operation data of the pipe network; S302: The cloud monitoring center inputs the global operation data of the pipeline network and the safety risk events of the multiple local natural gas pipeline network nodes into a second deep learning model to obtain a global safety risk identification result, wherein the global safety risk identification result includes whether there is a global safety risk, the level of the global safety risk and the impact range of the global safety risk; S303: When a global security risk is identified, the cloud monitoring center determines a security execution measure according to the level of the global security risk, the impact scope of the global security risk, the global operation data of the pipeline network and the cloud event decision rule library; S304: The cloud monitoring center generates a control instruction according to the security execution measure, wherein the control instruction includes the security action to be executed, the execution priority and the identification of the target monitoring unit, and sends the control instruction to the corresponding one or more target monitoring units.

5. The distributed natural gas pipeline network safety monitoring method based on edge computing according to claim 2 is characterized in that: The multiple target nodes specifically include: valve stations, metering stations, distribution stations, compressor stations and pressure regulating stations of the natural gas pipeline network.

6. The distributed natural gas pipeline network safety monitoring method based on edge computing according to claim 5 is characterized in that: The multi-mode sensor module comprises: The multi-mode sensor group corresponding to the valve station includes: The piezoelectric acceleration sensor is installed at the connection between the valve actuator and the valve stem to obtain the mechanical wear and vibration characteristic data of the valve in the valve station; Acoustic emission sensor, installed at the sealing interface between the valve body and the valve seat, is used to obtain the sealing electro-acoustic coupling characteristic data of the valve at the valve station; Photoelectric encoder, installed on the rotating shaft of the valve actuator, is used to obtain the action response characteristic data of the valve in the valve station; The strain gauge sensor is installed at the connection between the valve stem and the valve disc to obtain the surface deformation characteristic data of the valve at the valve station; The multi-mode sensor group corresponding to the metering station includes: The spectral analysis sensor is installed on the bypass sampling pipeline of the metering station pipeline and is located on the downstream side of the sampling valve to obtain the medium composition in the metering station pipeline; Ultrasonic flow sensor, installed on the straight section of the metering station pipeline, used to obtain disturbance acoustic characteristic data of the fluid in the metering station pipeline; The electromagnetic interference detection sensor is installed in the electrical connection terminal area of ​​the metering station control cabinet housing to obtain the electromagnetic interference characteristic data of the metering station pipeline.

7. The distributed natural gas pipeline network safety monitoring method based on edge computing according to claim 5 is characterized in that: The multi-mode sensor module comprises: The multi-mode sensor group corresponding to the sub-transmission station includes: The eddy current sensor is installed at the fluid diameter reducing section or elbow of the branch pipe network of the sub-transmission station to obtain the flow eddy current characteristic data of the branch pipe network; Thermocouples are installed on the outer surface of the pipe wall of the branch pipe network of the distribution station and are located in the heated area, and are used to obtain thermodynamic coupling characteristic data of the branch pipe network; The pressure spectrum analysis sensor is installed on the straight section of the branch network of the distribution station to obtain the pressure fluctuation spectrum characteristic data of the branch network; Fiber Bragg grating strain sensors are installed at the connection points of the branch pipe network at the sub-transmission station to obtain real-time stress and strain data at the connection points of the pipe network; The multi-mode sensor group corresponding to the compressor station includes: An acoustic emission sensor is installed on the surface of the compressor blade or the area around the compressor blade to obtain the wear acoustic characteristic data of the compressor blade; A vibration acceleration sensor is installed on the compressor bearing support seat or near the compressor bearing to obtain vibration spectrum characteristic data of the compressor bearing; An electromagnetic field intensity sensor is installed on the compressor rotor housing or the area around the compressor rotor to obtain electromagnetic radiation characteristic data of the compressor rotor; The flow sensor is installed on the straight section of the compressor inlet and outlet pipes to obtain the flow characteristic data of the medium inside the compressor.

8. The distributed natural gas pipeline network safety monitoring method based on edge computing according to claim 5 is characterized in that: The multi-mode sensor module comprises: The multi-mode sensor group corresponding to the pressure regulating station includes: A displacement sensor, installed on the valve stem of the pressure regulating valve, is used to obtain dynamic response characteristic data of the pressure regulating valve in the pressure regulating station; Thermodynamic sensors are installed in the expansion section, downstream section of the regulating valve, expansion section and elbow area of ​​the gas flow pipeline inside the pressure regulating station to obtain the entropy change characteristic data of gas expansion inside the pressure regulating station; An electromagnetic shielding effectiveness sensor is installed in the housing or shielding cover of the voltage regulating station to obtain electromagnetic shielding effectiveness characteristic data of the voltage regulating station; The acoustic resonance sensor is installed on the pipe wall or valve body surface before and after the pressure regulating valve, and is used to obtain the pressure gradient acoustic resonance characteristic data of the pressure regulating valve in the pressure regulating station.

9. The distributed natural gas pipeline network safety monitoring method based on edge computing according to claim 3 is characterized in that: The method further comprises: S50: When the event risk level of the security risk event of the local natural gas pipeline network node exceeds the preset level, the edge computing module generates an emergency response signal according to the event risk level and the local risk response rule base, and controls the execution unit to perform a security action corresponding to the emergency response signal.

10. A distributed natural gas pipeline network safety monitoring system based on edge computing, characterized in that: include: A plurality of monitoring units are respectively arranged at a plurality of target nodes of a natural gas pipeline network, each monitoring unit is configured with a first deep learning model, and each monitoring unit is used to collect node operation data of a local natural gas pipeline network node; real-time multi-dimensional anomaly detection processing is performed on the node operation data according to the first deep learning model to identify safety risk events of the local natural gas pipeline network node; Uploading the safety risk events and node operation data of the local natural gas pipeline network node to the cloud monitoring center; The cloud monitoring center is used to aggregate and analyze the node operation data uploaded by the multiple monitoring units to obtain the global operation data of the pipe network; Based on the global operation data of the pipeline network and the safety risk events uploaded by the plurality of monitoring units, a global safety risk identification result is obtained through a second deep learning model; Generate a control instruction according to the global security risk identification result, and send the control instruction to one or more target monitoring units; An execution unit is connected to the target monitoring unit and is used to execute corresponding safety actions according to the control instructions forwarded by the target monitoring unit.

Images