An oil pipeline leakage monitoring method, device, equipment and medium

Abstract

The application discloses an oil pipeline leakage monitoring method, device, equipment and medium, and relates to the technical field of pipeline monitoring. The method comprises the following steps: constructing a graph model of a pipeline monitoring network, wherein each node in the graph model is a measuring point on a pipeline; collecting pipeline data of each measuring point in real time, wherein the pipeline data comprises temperature, pressure and flow data; performing feature extraction on the pipeline data to obtain a plurality of feature data; determining a pipeline working condition for the plurality of feature data corresponding to each measuring point; adjusting the fusion weight of the fusion of various feature data in real time based on the pipeline working condition by using a reinforcement learning algorithm; performing feature fusion on the plurality of feature data according to the fusion weight to obtain multi-source fusion features; inputting the multi-source fusion features of each measuring point into a sensor fault recognition model to output a sensor fault label; and inputting the multi-source fusion features of each node carrying the sensor fault label into a leakage recognition and positioning model to output a leakage probability of each node. The application can improve the prediction accuracy.

Description

Technical Field

[0001] This application relates to the field of pipeline monitoring technology, and in particular to a method, device, equipment and medium for monitoring oil pipeline leaks. Background Technology

[0002] Traditional oil pipeline leak monitoring technologies primarily rely on data from single or simple combinations of sensors, using fixed rules for anomaly detection. Taking distributed fiber optic temperature measurement systems as an example, they typically trigger alarms based solely on single-point temperature thresholds or simple temperature gradient changes, supplemented by independent monitoring from pressure and flow sensors. Data fusion often involves signal superposition after threshold exceedances, lacking in-depth analysis of the spatiotemporal correlation of multi-source data. In data preprocessing, traditional methods use simple techniques like median filtering and linear interpolation to handle noise and anomalies, failing to effectively distinguish leak signals from environmental interference (such as localized temperature rises caused by sunlight). Leak identification and location depend on preset temperature change amplitudes or unidirectional correlations of pressure and flow, lacking a feature library containing collaborative patterns of multi-source parameters and an adaptive adjustment mechanism for pipeline operating conditions (such as oil type and transport load). For leak diffusion prediction, traditional solutions generally employ simplified dynamic models based on fluid mechanics, applicable only to regular pipeline networks and unable to integrate dynamic operating data in real time, leading to significant discrepancies between predicted results and actual diffusion. Summary of the Invention

[0003] The purpose of this application is to provide a method, apparatus, equipment, and medium for monitoring oil pipeline leaks, which can improve the accuracy of prediction.

[0004] To achieve the above objectives, this application provides the following solution:

[0005] Firstly, this application provides a method for monitoring oil pipeline leaks, including:

[0006] A graph model of a pipeline monitoring network is constructed, the graph model including multiple nodes, each node corresponding to a measurement point on the pipeline in the pipeline monitoring network that collects pipeline data;

[0007] Real-time acquisition of pipeline data at various measuring points, including temperature data, pressure data, and flow rate data;

[0008] Feature extraction is performed on the pipeline data to obtain various feature data;

[0009] For each measuring point, the pipeline condition is determined based on multiple feature data corresponding to the pipeline data. Based on the pipeline condition, a reinforcement learning algorithm is used to adjust the fusion weights of various feature data in real time.

[0010] Based on the fusion weights, feature fusion is performed on multiple feature data to obtain multi-source fused features;

[0011] The multi-source fusion features of each measuring point are input into the sensor fault identification model, and the sensor fault marker is output. The sensor fault marker is used to indicate whether the pipeline data of the measuring point is data collected under sensor fault conditions. The sensor fault identification model is obtained by training a support vector machine.

[0012] The multi-source fusion features carrying sensor fault markers at each node are input into the leak identification and localization model, which outputs the leak probability of each node; the leak identification and localization model is obtained by training a graph neural network.

[0013] Optionally, feature extraction is performed on the pipeline data to obtain various feature data, specifically including:

[0014] The temperature data at each measuring point is denoised, and the denoised temperature data at each measuring point is used to construct a time series of temperature data;

[0015] A temperature feature extraction model is used to extract the trend and periodic features of the time series of temperature data; the temperature feature extraction model is obtained by training a first LSTM network.

[0016] Based on the denoised temperature data in the spatial neighborhood of each measuring point, the spatial temperature gradient matrix of each measuring point is calculated.

[0017] Perform outlier detection on both pressure and flow data;

[0018] An outlier repair model is used to repair detected pressure and flow anomalies; the outlier repair model is obtained by training a conditional generative adversarial network.

[0019] Using the pressure and flow data after outlier repair, construct pressure and flow data time series corresponding to the temperature data time series;

[0020] Wavelet transform is performed on the time series of pressure data, and the fluctuation energy of the set high-frequency band is used as the frequency domain feature of the pressure data.

[0021] After applying Kalman filtering to the time series of traffic flow data, the traffic flow deviation rate after Kalman filtering is used as the trend of traffic flow data.

[0022] Optionally, the temperature data at each measuring point is denoised, and a time series of temperature data is constructed using the denoised temperature data at each measuring point, specifically including:

[0023] A spatiotemporal attention filtering algorithm is used to denoise the temperature data at each measuring point;

[0024] The denoised temperature data from each measuring point are processed by a sliding window, and the temperature data after sliding window processing constitutes a temperature data time series.

[0025] Optionally, an outlier repair model is used to repair the detected pressure and flow anomalies, specifically including:

[0026] The outlier is input into the outlier repair model along with the timestamp corresponding to the outlier, the non-outlier data within the adjacent time period of the outlier measurement point, and the operating parameters corresponding to the timestamp. The repaired pressure or flow rate value is then output. The outlier is either a pressure outlier or a flow rate outlier.

[0027] The physical constraints satisfied by the generator output in the conditional generative adversarial network are expressed as follows:

[0028] , ;

[0029] in, The pressure or flow rate value after repair. and The operating parameters are as follows: hour The minimum and maximum values, For fluctuation threshold, This represents the non-abnormal data within adjacent time periods of the outlier measurement point, where t represents the time step. Indicates a time period.

[0030] Optionally, after inputting the multi-source fusion features carrying sensor fault markers at each node into the leak identification and location model, and outputting the leak probability of each node, the oil pipeline leak monitoring method further includes:

[0031] By combining the physical coordinate encoding of each node, a spatial attention mechanism is used to locate node clusters with a probability exceeding a set threshold, and these node clusters with a probability exceeding the set threshold are identified as the core locations of the leakage points.

[0032] Optionally, after inputting the multi-source fusion features carrying sensor fault markers at each node into the leak identification and location model, and outputting the leak probability of each node, the oil pipeline leak monitoring method further includes:

[0033] The node features of each node at the current time are input into the time series encoder, which outputs a diffusion state prediction vector for multiple future time steps. The diffusion state prediction vector includes the leakage concentration and diffusion rate of each node. The time series encoder is obtained by training a second LSTM network.

[0034] Optionally, the negative sample data used when training the support vector machine comes from a cross-sensor correlation verification mechanism;

[0035] The cross-sensor correlation verification mechanism is as follows: when the data of the first sensor at a measuring point is abnormal and the data of the second and third sensors are normal, the pipeline data at that measuring point is determined to be data collected under sensor failure conditions; the first sensor, the second sensor, and the third sensor are one of the following: temperature sensor, pressure sensor, and flow sensor, respectively.

[0036] Secondly, this application provides an oil pipeline leak detection device, which applies the oil pipeline leak detection method, and the oil pipeline leak detection device includes:

[0037] The graph model construction module is used to construct a graph model of the pipeline monitoring network. The graph model includes multiple nodes, and each node corresponds to a measurement point on the pipeline in the pipeline monitoring network that collects pipeline data.

[0038] The data acquisition module is used to collect pipeline data at various measuring points in real time. The pipeline data includes temperature data, pressure data, and flow rate data.

[0039] The feature extraction module is used to extract features from the pipeline data to obtain various feature data;

[0040] The fusion weight determination module is used to determine the pipeline operating condition for each measuring point based on multiple feature data corresponding to the pipeline data, and to adjust the fusion weight of various feature data in real time based on the pipeline operating condition using a reinforcement learning algorithm.

[0041] The feature fusion module is used to fuse multiple feature data according to each fusion weight to obtain multi-source fused features;

[0042] The sensor fault labeling module is used to input the multi-source fusion features of each measuring point into the sensor fault identification model and output a sensor fault label; the sensor fault label is used to indicate whether the pipeline data of the measuring point is data collected under sensor fault conditions; the sensor fault identification model is obtained by training a support vector machine.

[0043] The leak identification and localization module is used to input the multi-source fusion features of each node carrying sensor fault markers into the leak identification and localization model and output the leakage probability of each node; the leak identification and localization model is obtained by training a graph neural network.

[0044] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the oil pipeline leakage monitoring method described in any one of the above.

[0045] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the oil pipeline leakage monitoring method described above.

[0046] According to the specific embodiments provided in this application, the following technical effects are disclosed:

[0047] This application provides a method, device, equipment, and medium for monitoring oil pipeline leaks. Based on pipeline operating conditions, a reinforcement learning algorithm is used to adjust the fusion weights of various feature data in real time, realizing adaptive adjustment of feature fusion according to operating conditions and improving the accuracy of leak identification. In addition, a sensor fault identification model is used to identify whether the pipeline data is collected under sensor fault conditions. The multi-source fusion features carrying sensor fault markers at each node are input into the leak identification and location model, taking into account the impact of sensor faults on leak identification, reducing the possibility of false alarms, and further improving the accuracy of leak identification. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 This is a flowchart illustrating an oil pipeline leakage monitoring method provided in one embodiment of this application.

[0050] Figure 2 This is a schematic diagram of the principle architecture of an oil pipeline leakage monitoring method provided in one embodiment of this application.

[0051] Figure 3 This is a layered schematic diagram of an oil pipeline leakage monitoring process provided in an embodiment of this application.

[0052] Figure 4 This is a schematic diagram of the cross-sensor correlation verification process decision provided in an embodiment of this application.

[0053] Figure 5 This is a schematic diagram of the functional modules of an oil pipeline leakage monitoring device provided in one embodiment of this application.

[0054] Figure 6 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0055] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0056] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0057] In one exemplary embodiment, this application provides a method for monitoring oil pipeline leaks, such as... Figures 1-3 As shown, the oil pipeline leakage monitoring method includes steps 101-107.

[0058] Step 101: Construct a graph model of the pipeline monitoring network. The graph model includes multiple nodes, and each node corresponds to a measurement point on the pipeline in the pipeline monitoring network that collects pipeline data.

[0059] In the graphical model, edges are pipe segments connecting nodes, and edge characteristics include the spatial distance between pipe segments, the thermal conductivity of the material, etc.

[0060] Step 102: Collect pipeline data at each measuring point in real time. The pipeline data includes temperature data, pressure data, and flow rate data.

[0061] Step 103: Extract features from the pipeline data to obtain various feature data.

[0062] Step 104: For each measuring point, determine the pipeline operating condition based on the various feature data corresponding to the pipeline data, and use a reinforcement learning algorithm to adjust the fusion weights of various feature data in real time based on the pipeline operating condition.

[0063] Step 105: Perform feature fusion on multiple feature data according to each fusion weight to obtain multi-source fused features.

[0064] Step 106: Input the multi-source fusion features of each measuring point into the sensor fault identification model and output a sensor fault marker; the sensor fault marker is used to indicate whether the pipeline data of the measuring point is data collected under sensor fault conditions; the sensor fault identification model is obtained by training a support vector machine.

[0065] Step 107: Input the multi-source fusion features of each node carrying sensor fault markers into the leakage identification and localization model, and output the leakage probability of each node; the leakage identification and localization model is obtained by training a graph neural network.

[0066] In an exemplary embodiment, the function of step 102 is implemented by the data acquisition module, which specifically collects pipeline data through a distributed optical fiber temperature measurement system, a pressure sensor, and a flow sensor. The temperature measurement optical cable is laid along the pipeline to collect temperature data, and the pressure sensor and flow sensor are installed at key locations in the pipeline to collect operating data. All pipeline data are synchronously sent to the data preprocessing module.

[0067] The functions of steps 103-106 are implemented by the data preprocessing module and the multi-source data fusion analysis module.

[0068] After the data preprocessing module denoises the temperature data and verifies and repairs the pressure and flow data, it outputs standardized data to the multi-source data fusion analysis module.

[0069] Multi-source data fusion analysis module: ① Employs a spatiotemporal feature fusion algorithm to extract time-series features of temperature data (such as the rate of temperature change and duration of anomalies extracted by the first LSTM network) and spatial gradient features (temperature difference matrix between adjacent measuring points) to construct a leakage signal feature library containing spatiotemporal dimensions; ② Determines the current transportation condition (oil type, load status) through the operating condition identification submodule, dynamically allocates the fusion weights of temperature, pressure, and flow data based on reinforcement learning, and establishes an adaptive data association model; The analysis results are sent to the leakage identification and location module.

[0070] The function of step 107 is implemented by the leak identification and location module. Based on the fusion analysis results and the location information of the temperature measuring optical cable, the leak identification and location module identifies the leak point and reconstructs the leak diffusion prediction mechanism by introducing a graph neural network (GNN).

[0071] In an exemplary embodiment, step 103 specifically includes steps 201-208.

[0072] Step 201: Denoise the temperature data at each measuring point, and construct a time series of temperature data using the denoised temperature data at each measuring point. This function is implemented by the spatiotemporal attention filtering module.

[0073] Step 202: Use a temperature feature extraction model to extract the trend and periodic features of the temperature data time series; the temperature feature extraction model is obtained by training the first LSTM network.

[0074] Step 203: Calculate the spatial temperature gradient matrix of each measuring point based on the denoised temperature data in the spatial neighborhood of each measuring point.

[0075] Step 204: Perform outlier detection on the pressure data and flow data respectively, and output the detected pressure outliers and flow outliers.

[0076] Both pressure and flow data have their own operating thresholds. Pressure and flow data that exceed the corresponding operating thresholds are considered outliers.

[0077] Step 205: Use an outlier repair model to repair the detected pressure and flow anomalies. The outlier repair model (Conditional Generative Adversarial Repair Module) is trained from a Conditional Generative Adversarial Network (cGAN).

[0078] Step 206: Using the pressure and flow data after outlier repair, construct pressure and flow data time series corresponding to the temperature data time series.

[0079] Step 207: Perform wavelet transform on the pressure data time series and use the fluctuation energy of the set high-frequency band as the frequency domain feature of the pressure data.

[0080] The high-frequency band is defined as 10Hz-50Hz. The fluctuation energy of the high-frequency band is expressed as: , Used to reflect the characteristics of pressure pulses caused by leakage; These are the wavelet transform coefficients corresponding to frequency f after performing wavelet transform on the repaired pressure data time series.

[0081] Step 208: After performing Kalman filtering on the time series of flow data, the flow deviation rate after Kalman filtering is used as the trend of flow data change.

[0082] Traffic flow data trend characteristics: Estimating the baseline traffic flow value using Kalman filtering Calculate the deviation rate Identify abnormal fluctuations in traffic. The data represents the traffic flow at time step t.

[0083] In an exemplary embodiment, step 201 specifically includes: using a spatiotemporal attention filtering algorithm to denoise the temperature data of each measuring point; performing sliding window processing on the denoised temperature data of each measuring point, and the temperature data after sliding window processing constitutes a temperature data time series.

[0084] The temperature data processed by the spatiotemporal attention filtering algorithm is then subjected to sliding window processing, with a window length of [value missing]. Corresponding time neighborhood , t represents the number of sampling periods, and t represents the time step.

[0085] The trend characteristics are the rate of temperature rise or fall, and the periodic characteristics are the baseline temperature curve under the influence of diurnal temperature difference.

[0086] The time series of single-point temperature data is processed using the first LSTM network. Modeling and extracting trend features (rate of temperature rise / fall) and periodic characteristics , This represents the deviation of the diurnal temperature range baseline curve. This is the historical average for the same period. This is a time series of temperature data at measurement point i at time step t. This represents the change in the temperature data time series at time step t compared to the temperature data time series at time step t-1. This represents the change in time between two time steps.

[0087] For continuous pipeline There are 1 measurement point, and the spatial neighborhood is represented as: Calculate the spatial temperature gradient matrix for the number of adjacent measuring points. ,in , The distance between measuring points, [ , ], This represents the abrupt change between measurement point j and measurement point i. Used to identify single-point mutations (a certain (Sudden increase) or regional gradient anomaly, where a regional gradient anomaly refers to at least 3 consecutive increases. Exceeding limits.

[0088] The temperature data preprocessing stage employs a spatiotemporal attention filtering algorithm, using a self-attention model to calculate spatiotemporal weighted correction values. This process suppresses environmental noise while preserving the temperature surge signal caused by the leak. The filtered result serves as input to step 201, enabling the temperature feature extraction model and spatial gradient calculation to be based on the denoised, high-quality data, resulting in a signal-to-noise ratio improvement of over 30%.

[0089] ;

[0090] in, This is a time series of temperature data after spatiotemporal weighting correction. The temperature data of measuring point j at time k. These are the weighting coefficients. It is determined by the self-attention model in the spatiotemporal attention filtering algorithm, thereby reflecting the correlation between the spatiotemporal neighborhood measurement points and the current measurement point.

[0091] This application employs a spatiotemporal attention filtering algorithm to denoise the temperature data, and calculates the correlation weights of spatiotemporal neighborhood measurement points using a self-attention model. This method preserves the spatiotemporally consistent temperature change signal caused by the leak and suppresses environmental noise (such as sunlight and electromagnetic interference). Unlike traditional single-point filtering or fixed-window smoothing, this method associates the temporal neighborhood (sliding window) with the spatial neighborhood (gradient of adjacent measurement points) to achieve directional enhancement of the leak temperature signal, thereby improving the signal-to-noise ratio by more than 30% and providing a high-quality data foundation for subsequent feature analysis.

[0092] Specifically, step 205 includes: storing the timestamp t corresponding to the outlier, and the non-outlier data within the adjacent time period of the outlier measurement point. and the operating condition parameters corresponding to the timestamp Input an outlier repair model and output the repaired pressure or flow rate value; the outlier value is either a pressure outlier or a flow rate outlier. Specifically Time slot to Non-abnormal data between time slots. The repaired pressure or flow values ​​replace the corresponding abnormal values. The repaired data generated by this application using conditional generative adversarial networks is more consistent with physical laws. This indicates the current pipeline's pressure rating.

[0093] The physical constraints satisfied by the generator output in the conditional generative adversarial network are expressed as follows:

[0094] , ;

[0095] in, The pressure or flow rate value after repair. and The operating parameters are as follows: hour The minimum and maximum values, For fluctuation threshold, This represents the non-abnormal data within adjacent time periods of the outlier measurement point, where t represents the time step. This indicates the time period. This physical constraint avoids logical contradictions caused by traditional interpolation (such as pseudo-data where pressure drops suddenly but flow remains unchanged).

[0096] Figure 3In the Conditional Generative Adversarial Network (CGN), D2 and D3 refer to two discriminator modules. In the pressure / flow data repair process, the CGN employs an adversarial process: a generator produces repaired data, and a discriminator verifies the data's authenticity. This achieves accurate repair of missing or abnormal pressure / flow data. D2 and D3, as different branches or stages of the discriminator, verify the authenticity of the repaired data from different dimensions, ensuring that the output repaired pressure / flow data accurately reflects the actual operating status of the pipeline, providing reliable foundational data for subsequent feature extraction and leak analysis.

[0097] This application inputs spatiotemporal feature parameters, along with the frequency domain features of pressure data (pressure fluctuation energy after wavelet transform) and the changing trends of flow data (flow deviation rate after Kalman filtering), into a support vector machine to construct a dynamically updated leakage signal feature library. The spatiotemporal feature parameters include the trend characteristics of temperature data. Periodic characteristics and spatial temperature gradient matrix .

[0098] Sensor fault markers are introduced during support vector machine training. Semi-supervised learning is used to distinguish between leakage anomalies and single-sensor faults, thereby reducing the false alarm rate. The sensor fault markers are derived from the verification results of a cross-sensor correlation verification mechanism.

[0099] This application uses the K-means clustering algorithm to analyze historical operating data (oil viscosity). , conveying pressure Ambient temperature The system classifies the operating conditions to obtain various pipeline operating conditions, and drives the reinforcement learning algorithm to adjust the data fusion weights in real time to form leakage judgment rules that are adapted to different operating conditions.

[0100] For each measuring point, the pipeline condition is determined based on various feature data corresponding to the pipeline data. Specifically, this includes matching the feature data corresponding to the measuring point with various pipeline conditions obtained by the K-means clustering algorithm, calculating the similarity between the feature data and the feature data corresponding to each pipeline condition, and taking the pipeline condition with the highest similarity as the pipeline condition of the measuring point.

[0101] In the process of adjusting the data fusion weights, the environment of the reinforcement learning algorithm is the oil pipeline operation scenario. The state includes the current pipeline oil viscosity, delivery pressure, ambient temperature and other operating parameters. The actions include adjusting the fusion weights of temperature, pressure and flow data. The reward is to minimize the classification error of historical samples in the leak detection.

[0102] Pipeline operating conditions, specifically, such as full-load operation of high-viscosity oil in winter and peak shaving of low-viscosity oil in summer.

[0103] In step 104, a reinforcement learning algorithm is used to dynamically adjust the weights of the multi-source data fusion based on the current working conditions. , As a weight for temperature feature fusion, Weights are fused to represent pressure features. Weights are fused based on traffic characteristics.

[0104] ;

[0105] in, For the first The temperature, pressure, and / or flow rate characteristics of each sample, where n corresponds to temperature T, pressure P, and flow rate Q, respectively, and K is the total number of samples. The label is 0 or 1, representing the leakage. By minimizing the classification error of historical samples, leakage determination rules adapted to the current operating conditions are generated.

[0106] In an exemplary embodiment, after step 107, the oil pipeline leakage monitoring method further includes: combining the physical coordinate codes of each node, using a spatial attention mechanism to locate the node cluster with a probability exceeding a set threshold, and taking the node cluster with a probability exceeding the set threshold as the core location of the leakage point.

[0107] In an exemplary embodiment, in step 106, the negative sample data used when training the support vector machine comes from a cross-sensor correlation verification mechanism.

[0108] The cross-sensor correlation verification mechanism is as follows: when the data of the first sensor at a measuring point is abnormal and the data of the second and third sensors are normal, the pipeline data at that measuring point is determined to be data collected under sensor failure conditions; the first sensor, the second sensor, and the third sensor are one of the following: temperature sensor, pressure sensor, and flow sensor, respectively.

[0109] For example, such as Figure 4 As shown, when the pressure data is abnormal at a certain moment but the temperature at the corresponding measuring point does not change abruptly and the flow rate does not change in tandem, the triple verification logic is triggered:

[0110] 1) Check if the temperature gradient between adjacent measuring points exceeds the limit;

[0111] 2) Verify the historical failure probability of the pressure sensor;

[0112] 3) Compare the fluctuation patterns of data from the same type of pipeline during the same period.

[0113] If a sensor malfunction is identified, the data point is marked and synchronized to the multi-source data fusion analysis module to update the "sensor anomaly" negative samples in the leakage feature library.

[0114] In an exemplary embodiment, after step 107, the oil pipeline leakage monitoring method further includes: inputting the node features of each node at the current moment into a time series encoder, and outputting a diffusion state prediction vector for multiple future time steps; the diffusion state prediction vector includes the leakage concentration and diffusion rate of each node; the time series encoder is obtained by training a second LSTM network.

[0115] This application utilizes the spatiotemporal feature fusion of a multi-source data fusion analysis module to reconstruct a leakage propagation prediction mechanism through a graph neural network. The specific rules are as follows: Based on the constructed graph model (graph structure data) of the pipeline monitoring network, temperature measurement points, pressure sensors, and flow sensors are used as nodes. Node features include real-time temperature, pressure, and flow data, as well as oil physical parameters (viscosity, volatility). Edge features are defined as connection attributes such as spatial distance of pipeline segments and thermal conductivity of materials, forming a graph structure data that includes the spatial topology and operating parameters of the pipeline network. These edge features also serve as inputs to the graph neural network. Parameters output by the multi-source data fusion analysis module, such as the temperature change rate, spatial gradient matrix, pressure frequency domain energy, and flow deviation rate, are used as dynamic feature inputs to the graph neural network. Through multi-layer graph convolution operations, the spatial propagation pattern of leakage signals in the pipeline network and the dependencies between nodes are captured.

[0116] In the leak identification and localization stage, the graph neural network (Graph Neural Network) determines the leak probability of each measuring point through a node classification task. Combined with the physical coordinate encoding of the temperature-sensing optical cable, it uses a spatial attention mechanism to locate node clusters with probability values ​​exceeding a threshold (e.g., 0.85), thus determining the core location of the leak point (location error ≤ ±2m) and the surrounding three-level influence range (core leak area, diffusion warning area, and environmentally sensitive monitoring area). Based on the leak probability of each measuring point output by the Graph Neural Network, the geographic coordinate embedding vector of the temperature-sensing optical cable, the node clusters located by the spatial attention mechanism with a probability ≥ 0.85, and pipeline operation parameters and environmentally sensitive area data, the three areas of the core leak area, diffusion warning area, and environmentally sensitive monitoring area are determined. In the predictive alarm stage, the Graph Neural Network further models the leak diffusion process through an edge prediction task: using the node features at the current moment as the initial state, it inputs historical leak dynamic features extracted by a time-series encoder (e.g., LSTM) to generate diffusion state prediction vectors for multiple future time steps, outputting parameters such as leak concentration and diffusion rate for each node. The system constructs a three-dimensional diffusion heat map based on the prediction results, and assesses the impact level of the leak on the surrounding area by combining data on environmentally sensitive areas (such as water sources and residential areas). It automatically generates intelligent alarm signals that include the precise coordinates of the leak point, the predicted diffusion path for the next 24 hours, and a graded emergency response plan.

[0117] Temporal features such as temperature change rate, pressure frequency domain energy, and flow deviation rate are directly used as dynamic feature inputs to the nodes of the graph neural network. These features are extracted from the original data through LSTM network, wavelet transform, etc., reflecting the changing trend of parameters over time.

[0118] Spatial distribution characteristics: The spatial temperature gradient matrix is ​​transformed into the correlation weights between nodes, which guides the graph neural network to construct node dependencies and capture the spatial propagation pattern of leakage signals.

[0119] Oil physical parameters (viscosity, volatility) are used as static features of nodes to participate in the construction of the graph model. These parameters have been used in the working condition identification submodule to classify different transportation scenarios. The graph neural network further learns the association rules between working conditions and diffusion patterns on this basis.

[0120] This application utilizes a reinforcement learning algorithm to generate dynamic fusion weights, which are implicitly integrated into the attention mechanism of a graph neural network, enabling the leak detection and localization module to automatically focus on the most relevant sensor data under different operating conditions.

[0121] The leakage signal feature library constructed by support vector machines provides initial classification rules for graph neural networks. Graph neural networks expand and refine these rules in terms of spatial dimensions through graph convolution operations.

[0122] The fault-labeled data generated by the cross-sensor correlation verification mechanism is used for noise filtering in the training of graph neural networks, improving the sensitivity of the leak identification and localization model to real leak signals.

[0123] The denoised data processed by the spatiotemporal attention filtering algorithm is used as input to the graph neural network to ensure that the leak identification and localization model learns the spatiotemporal consistent changes caused by the real leak.

[0124] After the time-series data fragments generated by sliding window processing are encoded by the temperature feature extraction model, they are fused with the spatial features output by the graph neural network using tensors to achieve bidirectional enhancement of spatiotemporal information.

[0125] This application achieves a significant leap in leak monitoring accuracy through the deep integration of spatiotemporal correlation analysis and intelligent data processing technologies. It solves the problems of traditional methods relying on single-point data threshold judgments, being susceptible to environmental noise interference, and failing to capture the spatiotemporal evolution characteristics of leaks. This application employs a spatiotemporal attention filtering algorithm to denoise temperature data, and learns the correlation of spatiotemporal neighborhood measurement points through a self-attention model, effectively suppressing noise such as sunlight and electromagnetic interference while retaining the spatiotemporally consistent temperature change signal caused by the leak, thus improving the data signal-to-noise ratio by more than 30%. Based on this, it combines an LSTM network to extract time-series features such as temperature change rate and diurnal temperature difference deviation, identifies abnormal distribution patterns by calculating the spatial temperature gradient matrix of continuous measurement points, and integrates multi-source parameters such as pressure frequency domain energy and flow deviation rate into the feature vector, constructing a dynamic feature library that includes spatiotemporal dimensions and multi-sensor collaborative patterns. This technological breakthrough, from single-point data to spatiotemporal correlation and from single parameters to multi-source fusion, enables the system to accurately identify micro-leaks and early signs, reducing the false alarm rate from 30%-50% to below 2% compared to traditional methods, significantly improving the accuracy and reliability of leak signal identification.

[0126] This application overcomes the limitations of traditional fixed-rule monitoring and upgrades predictive capabilities. Addressing the differences in leakage characteristics under various oil product transportation and load conditions, typical operating conditions are categorized using K-means clustering. Reinforcement learning is used to dynamically adjust the fusion weights of temperature, pressure, and flow data. For example, the weight of the flow parameter is automatically increased during the transportation of high-viscosity oils, enabling the monitoring model to adapt to complex operating conditions and avoid missed detections caused by fixed rules. Monitoring reliability in complex scenarios is improved by more than 50%. Simultaneously, a conditional generative adversarial network is introduced to repair abnormal pressure / flow data, ensuring that the repair results conform to the physical constraints of the operating conditions and providing reliable input for feature fusion. In the leakage propagation prediction stage, a graph neural network is used to construct graph-structured data of the pipeline network topology and operating parameters. Graph convolution operations are used to capture the spatial propagation patterns of leakage signals in complex pipeline networks, overcoming the assumptions of traditional dynamic models regarding regular pipeline networks and reducing the diffusion prediction error from 30% to less than 15%. By combining historical leakage dynamic features extracted by LSTM, the system can generate a diffusion state prediction vector for the next 24 hours, construct a three-dimensional diffusion heat map and assess the environmental impact level, achieving a leap from immediate alarm to future risk projection, and reserving sufficient time for emergency decision-making.

[0127] This application achieves a fully intelligent upgrade across the entire chain, from leak detection to tiered decision support, and possesses system-level collaborative advantages. By combining precise leak point location, diffusion path prediction, and data from environmentally sensitive areas, a three-dimensional risk assessment system encompassing safety, environment, and economy is constructed. Based on the prediction results, differentiated emergency response plans are automatically generated. For example, for Level 1 risks, automatic switching of backup pipelines and linkage with drone inspections are triggered; for Level 2 risks, suggestions for manual inspection routes and material allocation are pushed, improving emergency response efficiency by over 60% and significantly reducing losses and risks caused by leaks. Furthermore, sensor layout is optimized through digital twin models, encrypting temperature measurement points in leak-prone areas and matching them with the spatiotemporal neighborhood parameters of the algorithm, reducing monitoring blind spots by over 70%, forming a deep synergy between hardware layout and algorithm design. The system also has online learning capabilities. When the pipeline operating environment changes, it dynamically updates the leakage feature database and graph neural network (leakage identification and location model) using new data. The update cycle does not exceed 24 hours, ensuring that the long-term operating accuracy remains stable at over 98%. It has built a highly accurate, highly adaptable, predictable, and easy-to-determine oil pipeline leakage monitoring technology system, achieving breakthrough improvements in core indicators such as leakage identification accuracy, false alarm rate control, and emergency response efficiency.

[0128] Based on the same inventive concept, this application also provides an oil pipeline leak monitoring device for implementing the oil pipeline leak monitoring method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations of one or more embodiments of the oil pipeline leak monitoring device provided below can be found in the limitations of the oil pipeline leak monitoring method described above, and will not be repeated here.

[0129] In one exemplary embodiment, such as Figure 5 As shown, an oil pipeline leak detection device is provided. This device applies the aforementioned oil pipeline leak detection method and includes:

[0130] The graph model construction module is used to construct a graph model of the pipeline monitoring network. The graph model includes multiple nodes, each node corresponding to a measurement point on the pipeline in the pipeline monitoring network that collects pipeline data.

[0131] The data acquisition module is used to collect pipeline data at various measuring points in real time. The pipeline data includes temperature data, pressure data, and flow rate data.

[0132] The feature extraction module is used to extract features from the pipeline data to obtain various feature data.

[0133] The fusion weight determination module is used to determine the pipeline operating condition for each measuring point based on various feature data corresponding to the pipeline data, and to adjust the fusion weight of various feature data in real time based on the pipeline operating condition using a reinforcement learning algorithm.

[0134] The feature fusion module is used to fuse multiple feature data according to each fusion weight to obtain multi-source fused features.

[0135] The sensor fault labeling module is used to input the multi-source fusion features of each measuring point into the sensor fault identification model and output a sensor fault label; the sensor fault label is used to indicate whether the pipeline data of the measuring point is the data collected under the condition of sensor failure; the sensor fault identification model is obtained by training a support vector machine.

[0136] The leak identification and localization module is used to input the multi-source fusion features of each node carrying sensor fault markers into the leak identification and localization model and output the leakage probability of each node; the leak identification and localization model is obtained by training a graph neural network.

[0137] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 6 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores oil pipeline leak monitoring data. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements an oil pipeline leak monitoring method.

[0138] Those skilled in the art will understand that Figure 6The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0139] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0140] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0141] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0142] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be, but are not limited to, general-purpose processors, central processing units, graphics processors, digital signal processors, programmable logic units (PLCs), or other data processing logic units.

[0143] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0144] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method of monitoring a pipeline for leaks of an oil product, the method comprising: The oil product pipeline leakage monitoring method comprises: A graph model of the pipeline monitoring network is constructed, the graph model comprising a plurality of nodes, each node corresponding to a measurement point on a pipeline in the pipeline monitoring network for collecting pipeline data; Pipeline data of each measurement point is collected in real time, the pipeline data comprising temperature data, pressure data and flow data; Feature extraction is performed on the pipeline data to obtain a plurality of feature data; For each measurement point, a pipeline working condition is determined according to the plurality of feature data corresponding to the pipeline data, and a reinforcement learning algorithm is used to adjust fusion weights of the various feature data for fusion in real time based on the pipeline working condition; Feature fusion is performed on the plurality of feature data according to the fusion weights to obtain multi-source fusion features; The multi-source fusion features of each measurement point are input into a sensor fault identification model to output a sensor fault label; the sensor fault label is used to indicate whether the pipeline data of the measurement point is collected in a sensor fault condition; the sensor fault identification model is obtained by training a support vector machine; The multi-source fusion features of each node carrying the sensor fault label are input into a leakage identification and positioning model to output a leakage probability of each node; the leakage identification and positioning model is obtained by training a graph neural network.

2. The oil product pipeline leak monitoring method of claim 1, wherein Feature extraction is performed on the pipeline data to obtain a plurality of feature data, specifically comprising: Denoising is performed on the temperature data of each measurement point, and a temperature data time sequence is constructed using the denoised temperature data of each measurement point; Trend features and periodic features of the temperature data time sequence are extracted using a temperature feature extraction model; the temperature feature extraction model is obtained by training a first LSTM network; A spatial temperature gradient matrix of each measurement point is calculated based on the denoised temperature data in the spatial neighborhood of each measurement point; Anomaly value detection is performed on the pressure data and the flow data respectively; An anomaly value repair model is used to repair the detected pressure anomaly values and flow anomaly values; the anomaly value repair model is obtained by training a conditional generative adversarial network; Pressure data time sequences and flow data time sequences corresponding to the temperature data time sequence are constructed using the pressure data and the flow data after anomaly value repair; Wavelet transform is performed on the pressure data time sequence, and fluctuation energy of a set high frequency band is taken as a frequency domain feature of the pressure data; After Kalman filtering is performed on the flow data time sequence, a flow deviation rate after Kalman filtering is taken as a change trend of the flow data.

3. The oil product pipeline leak monitoring method of claim 2, wherein, Denoising is performed on the temperature data of each measurement point using a spatio-temporal attention filtering algorithm; The denoised temperature data of each measurement point is subjected to sliding window processing, and the temperature data after sliding window processing constitutes a temperature data time sequence. The anomaly value repair model is used to repair the detected pressure anomaly values and flow anomaly values, specifically comprising:

4. The oil product pipeline leak monitoring method of claim 2, wherein A timestamp corresponding to the anomaly value, non-anomaly data in a neighboring time period of the anomaly value measurement point and working condition parameters corresponding to the timestamp are input into the anomaly value repair model to output a repaired pressure value or flow value; the anomaly value is a pressure anomaly value or a flow anomaly value; ​ The physical constraint satisfied by the output of the generator in the conditional generative adversarial network is represented as: ， ； wherein, is a repaired pressure value or flow value, and are minimum and maximum values, respectively, for the operating parameter when is the minimum and maximum value, respectively, for the operating parameter is a fluctuation threshold value, denotes non-anomalous data in the adjacent time period for the anomalous value measurement point, t denotes a time step, denotes a time period.

5. The oil product pipeline leak monitoring method of claim 1, wherein, After inputting the multi-source fusion features of each node carrying the sensor failure marker into the leakage identification and positioning model and outputting the leakage probability of each node, the oil product pipeline leakage monitoring method further comprises: In combination with the physical coordinate coding of each node, a spatial attention mechanism is used to locate the node cluster whose probability exceeds the set threshold, and the node cluster whose probability exceeds the set threshold is taken as the core position of the leakage point.

6. The oil product pipeline leak monitoring method of claim 1, wherein, After inputting the multi-source fusion features of each node carrying the sensor failure marker into the leakage identification and positioning model and outputting the leakage probability of each node, the oil product pipeline leakage monitoring method further comprises: The node features of each node at the current time are input into a time sequence encoder to output a diffusion state prediction vector at a plurality of future time steps; the diffusion state prediction vector comprises the leakage concentration and diffusion velocity of each node; and the time sequence encoder is obtained by training a second LSTM network.

7. The oil product pipeline leak monitoring method of claim 1, wherein The negative sample data used for training the support vector machine is derived from a cross-sensor correlation verification mechanism; The cross-sensor correlation verification mechanism is configured to determine that the pipeline data of a measuring point is collected under a sensor failure condition when the first sensor data of the measuring point is abnormal and the second sensor and the third sensor are normal; and the first sensor, the second sensor and the third sensor are one of a temperature sensor, a pressure sensor and a flow sensor.

8. An oil pipeline leak monitoring apparatus, characterized by, The oil product pipeline leakage monitoring device applies the oil product pipeline leakage monitoring method of any one of claims 1-7, and comprises: a graph model construction module configured to construct a graph model of the pipeline monitoring network, wherein the graph model comprises a plurality of nodes, and each node corresponds to a measuring point on the pipeline of the pipeline monitoring network which collects pipeline data; a data acquisition module configured to acquire pipeline data of each measuring point in real time, wherein the pipeline data comprises temperature data, pressure data and flow data; a feature extraction module configured to extract features from the pipeline data to obtain a plurality of feature data; a fusion weight determination module configured to determine a pipeline working condition for each measuring point based on the plurality of feature data corresponding to the pipeline data, and adjust fusion weights of the plurality of feature data for fusion in real time based on the pipeline working condition by using a reinforcement learning algorithm; a feature fusion module configured to fuse the plurality of feature data based on the fusion weights to obtain multi-source fusion features; a sensor failure marker module configured to input the multi-source fusion features of each measuring point into a sensor failure identification model to output a sensor failure marker; the sensor failure marker is used to indicate whether the pipeline data of the measuring point is collected under a sensor failure condition; and the sensor failure identification model is obtained by training a support vector machine; a leakage identification and positioning module configured to input the multi-source fusion features of each node carrying the sensor failure marker into a leakage identification and positioning model to output a leakage probability of each node; and the leakage identification and positioning model is obtained by training a graph neural network.

9. A computer device comprising: A memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that the processor executes the computer program to implement the oil pipeline leakage monitoring method of any one of claims 1-7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the oil pipeline leakage monitoring method of any one of claims 1-7.

Images