Multi-Sensor Based Oil and Gas Pipeline Fault Location Method and System

Abstract

The present invention discloses a method and system for fault location of oil and gas pipelines based on multi-sensors. The method includes collecting pressure, flow rate, and vibration data; performing data cleaning and feature extraction to obtain low-dimensional features; constructing a Bayesian network model to analyze the conditional probability between features and generate a feature correlation matrix; when the conditional probability is lower than a threshold, weighted updating the feature values; combining with a pipeline geographic information database to match the topological structure and operation years, and using a weighted voting algorithm to preliminarily locate the fault area; using an LSTM network to predict the fault probability distribution; verifying the location result through the Hausdorff distance, and finally determining the fault location. This method realizes accurate location of the pipeline fault location, improves the pipeline operation and maintenance efficiency, and extends the service life of the pipeline.

Description

Technical Field

[0001] The present invention relates to the technical field of pipeline fault locating technology, and in particular to a multi-sensor based oil and gas pipeline fault locating method and system. Background Art

[0002] At present, oil and gas pipelines are the core infrastructure for energy transportation, and their safe operation is directly related to energy supply and social stability. However, pipelines have been in complex environments for a long time, and are affected by factors such as geological changes and material aging. They are prone to failures such as leakage and cracks, which may cause significant economic losses and environmental risks. Traditional fault detection methods mainly rely on single sensor data, but the pipeline system covers a wide range and has changeable operating conditions. It is difficult for a single data source to fully capture fault characteristics, and it is easily affected by noise interference or data loss, resulting in insufficient positioning accuracy. In addition, the operating status of the pipeline is dynamically affected by multi-dimensional factors such as flow, pressure, and vibration. The fault characteristics are closely related to the pipeline topology, which further increases the difficulty of accurate positioning. Therefore, how to achieve high-precision and high-reliability fault positioning in a complex environment has become a key issue that needs to be urgently solved in the field of pipeline operation and maintenance.

[0003] In one existing technology, a single type of pressure sensor is first deployed at intervals along the pipeline to collect internal pressure data of the pipeline at a fixed sampling frequency. When the pressure fluctuation exceeds the preset threshold, the system triggers the fault detection process: the original pressure data is processed by sliding average filtering, and the peak detection algorithm is used to identify abnormal pressure points, and then the fault location is calculated based on the theoretical model of pressure wave propagation velocity. For periods of missing data, the system uses linear extrapolation of the data of the previous period to complete the data. Finally, the calculated fault location coordinates are matched with the mileage pile number in the pipeline GIS system, and the pipeline pile number positioning result of the fault point is output.

[0004] The following problems exist in the existing technology: a single data source cannot cope with the problem of missing data caused by sensor failure or signal interference, and a simple linear extrapolation completion method is difficult to restore the true operation status of the pipeline. The propagation speed of the pressure wave is easily affected by factors such as the pipeline material and medium characteristics, and significant errors will occur under complex working conditions. The coupling relationship between multiple physical quantities such as pressure, flow, and vibration in the pipeline system is not well considered, resulting in incomplete extraction of fault features. The positioning process lacks effective use of the historical operation data and topological structure of the pipeline, making it difficult to identify the aging risk of a specific pipe section. Therefore, the existing technology cannot accurately locate the location of pipeline faults. Summary of the invention

[0005] The present invention provides a multi-sensor based oil and gas pipeline fault location method and system to accurately locate the pipeline fault position.

[0006] In a first aspect, to solve the above technical problems, the present invention provides a multi-sensor-based oil and gas pipeline fault location method, including:

[0007] Collect sensor data during pipeline operation, where the sensor data includes pressure, flow rate, and vibration data;

[0008] Perform data cleaning and feature extraction on the sensor data to obtain low-dimensional fault features;

[0009] Construct a Bayesian network model based on the low-dimensional fault features, and calculate the conditional probabilities between features according to the Bayesian network model and generate a feature correlation matrix;

[0010] When the conditional probability is lower than a preset probability threshold, update the corresponding feature value weighted according to the low-dimensional fault features to obtain updated fault features;

[0011] Calculate the causal values and fusion scores between the updated fault features according to the feature correlation matrix and the Bayesian network model, and generate pipeline fault features in combination with the updated fault features;

[0012] Match the pipeline fault features with the topological structure and operation years in a pre-established pipeline geographic information database, and use a weighted voting algorithm to determine a candidate set of fault locations near the connection points of aging pipelines;

[0013] According to the updated fault features, use a long short-term memory network model to predict the probability distribution of pipeline faults to obtain high-probability fault locations;

[0014] Calculate the average Hausdorff distance according to the candidate set of fault locations and the high-probability fault locations. When the average Hausdorff distance is lower than a preset distance threshold, re-determine the candidate set of fault locations to obtain the final fault location.

[0015] In an optional implementation manner, the performing data cleaning and feature extraction on the sensor data to obtain low-dimensional fault features includes:

[0016] Perform data completion on the sensor data using a linear interpolation algorithm to obtain multi-dimensional data;

[0017] Perform dimensionality reduction processing and feature extraction on the multi-dimensional data using a principal component analysis method to obtain low-dimensional fault features.

[0018] In an optional implementation manner, the constructing a Bayesian network model based on the low-dimensional fault features, and calculating the conditional probabilities between features according to the Bayesian network model and generating a feature correlation matrix includes:

[0019] Based on the low-dimensional fault features, perform structure learning through the PC algorithm to determine the dependence relationship between features, adopt the maximum likelihood estimation method for parameter learning, determine the conditional probability between features, and construct a Bayesian network;

[0020] Calculate the Pearson correlation coefficient matrix according to the Bayesian network to obtain the feature correlation matrix.

[0021] In an alternative embodiment, when the conditional probability is lower than a preset probability threshold, the corresponding eigenvalue is weighted and updated according to the low-dimensional fault features to obtain updated fault features, including:

[0022] When the conditional probability between features is lower than the preset probability threshold, use the linear regression algorithm to determine the weight coefficient of the low-dimensional fault features and perform weighted update of the corresponding eigenvalue;

[0023] Modify the low-dimensional fault features according to the updated eigenvalue, and optimize the modified low-dimensional fault features by the Kalman filter method to obtain updated fault features.

[0024] In an alternative embodiment, calculating the causal value and fusion score between the updated fault features according to the feature correlation matrix and the Bayesian network model, and generating pipeline fault features in combination with the updated fault features, including:

[0025] Extract the non-diagonal values of the feature correlation matrix. When the non-diagonal value is greater than a preset correlation threshold, analyze the causal value between features through the Bayesian network model to determine the high-causal feature combination;

[0026] According to the high-causal feature combination, fuse the corresponding sensor data by the weighted average method to obtain the fusion score;

[0027] When the fusion score is greater than a preset fusion threshold, regard the high-causal feature combination as the high-fusion feature combination;

[0028] Extract pipeline fault features according to the updated fault features corresponding to the high-fusion feature combination by the principal component analysis method.

[0029] In an alternative embodiment, matching the pipeline fault features with the topological structure and operation years in the pre-established pipeline geographic information database, and using the weighted voting algorithm to determine the candidate set of fault positions near the connection points of aging pipelines, including:

[0030] Obtain topological data and operation years from the pre-established pipeline geographic information database;

[0031] Calculate the topological matching degree between the pipeline fault features and the topological data by the cosine similarity;

[0032] Calculate the aging coefficient using the linear weighted method based on the topological matching degree and the operating years, and estimate the failure probability near the pipeline connection point through the logistic regression algorithm;

[0033] Use the weighted voting algorithm to perform weight assignment and weighted calculation on the failure probability and the topological matching degree to obtain the failure location score;

[0034] When the failure location score is greater than the preset score threshold, include the corresponding location in the preliminary failure location candidate set;

[0035] According to the preliminary failure location candidate set, perform position calibration through the pre-stored pipeline coordinates and calibration parameters, and optimize the positioning accuracy using the least squares method to obtain the failure location candidate set.

[0036] In an alternative embodiment, calculating the average Hausdorff distance based on the failure location candidate set and the high-probability failure location, and when the average Hausdorff distance is lower than the preset distance threshold, re-determine the failure location candidate set to obtain the final failure location, including:

[0037] Calculate the average Hausdorff distance based on the failure location candidate set and the high-probability failure location. When the average Hausdorff distance is lower than the preset distance threshold, use the logistic regression algorithm to adjust the weights of the weighted voting algorithm to generate the second weights of the updated failure probability and topological matching degree;

[0038] Recalculate the failure location score according to the second weights and re-analyze the failure location candidate set;

[0039] When iteratively updating until the average Hausdorff distance is higher than or equal to the distance threshold, determine the failure location candidate set of the last iteration as the final failure location candidate set;

[0040] Select the failure location with the highest failure location score from the final failure location candidate set as the final failure location.

[0041] In a second aspect, the present invention provides an oil and gas pipeline fault location system based on multi-sensors, including:

[0042] A data acquisition module for collecting sensor data during pipeline operation, where the sensor data includes pressure, flow rate, and vibration data;

[0043] A data processing module for performing data cleaning and feature extraction based on the sensor data to obtain low-dimensional fault features;

[0044] A feature association module, configured to construct a Bayesian network model based on the low-dimensional fault features, and calculate the conditional probability between features according to the Bayesian network model and generate a feature correlation matrix;

[0045] A feature update module, configured to, when the conditional probability is lower than a preset probability threshold, perform weighted update on the corresponding feature values according to the low-dimensional fault features to obtain updated fault features;

[0046] A fault feature analysis module, configured to calculate the causal value and fusion score between the updated fault features according to the feature correlation matrix and the Bayesian network model, and generate pipeline fault features in combination with the updated fault features;

[0047] A fault location analysis module, configured to match the pipeline fault features with the topological structure and operation years in a pre-established pipeline geographic information database, and use a weighted voting algorithm to determine a candidate set of fault locations near the connection points of aging pipelines;

[0048] A fault probability analysis module, configured to, according to the updated fault features, use a long short-term memory network model to predict the probability distribution of pipeline faults occurring to obtain high-probability fault locations;

[0049] A fault location determination module, configured to calculate the average Hausdorff distance according to the candidate set of fault locations and the high-probability fault locations, and when the average Hausdorff distance is lower than a preset distance threshold, re-determine the candidate set of fault locations to obtain the final fault location.

[0050] In a third aspect, the present invention further provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, where when the processor executes the computer program, the method for locating oil and gas pipeline faults based on multi-sensors described in any one of the above is implemented.

[0051] In a fourth aspect, the present invention further provides a computer-readable storage medium, where the computer-readable storage medium includes a stored computer program, and wherein when the computer program runs, it controls the device where the computer-readable storage medium is located to execute the method for locating oil and gas pipeline faults based on multi-sensors described in any one of the above.

[0052] Compared with the prior art, the present invention has the following beneficial effects:

[0053] (1) Through multi-sensor (pressure, flow, vibration) data fusion and Bayesian network modeling, the problem that a single data source is vulnerable to interference or missing, resulting in incomplete feature extraction, is solved, and the comprehensiveness and reliability of fault feature analysis are significantly improved.

[0054] (2) By combining the topological structure and operation years of the pipeline geographic information database, a weighted voting algorithm is used to dynamically match the fault candidate areas, solving the technical defect that the traditional method ignores the pipeline aging risk and spatial correlation, and realizing the accurate preliminary screening of the fault location.

[0055] (3) Using the LSTM network to predict the fault probability distribution and iteratively optimizing the candidate set through the Hausdorff distance, solving the mispositioning problem caused by static threshold determination, and still maintaining high-robustness fault location ability under complex working conditions.

[0056] (4) Dynamically updating the feature weights through the feature correlation matrix and causal analysis, and combining the Kalman filter to optimize the data quality, solving the influence of noise interference and feature redundancy on the positioning accuracy, and realizing the coordinated improvement of the pipeline fault location efficiency and accuracy. Description of the Drawings

[0057] Figure 1 is a schematic flow chart of the oil and gas pipeline fault location method based on multi-sensors provided by the first embodiment of the present invention;

[0058] Figure 2 is a schematic structural diagram of the oil and gas pipeline fault location system based on multi-sensors provided by the second embodiment of the present invention. Specific Embodiments

[0059] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0060] Refer to Figure 1 The first embodiment of the present invention provides an oil and gas pipeline fault location method based on multi-sensors, including the following steps:

[0061] S11, collecting sensor data during pipeline operation, where the sensor data includes pressure, flow rate, and vibration data;

[0062] S12, performing data cleaning and feature extraction according to the sensor data to obtain low-dimensional fault features;

[0063] S13, constructing a Bayesian network model according to the low-dimensional fault features, and calculating the conditional probability between features and generating a feature correlation matrix according to the Bayesian network model;

[0064] S14. When the conditional probability is lower than a preset probability threshold, update the corresponding eigenvalue by weighting according to the low-dimensional fault feature to obtain an updated fault feature.

[0065] S15. Calculate the causal value and fusion score between the updated fault features according to the feature correlation matrix and the Bayesian network model, and generate pipeline fault features in combination with the updated fault features.

[0066] S16. Match the pipeline fault features with the topological structure and operation years in the pre-established pipeline geographic information database, and use a weighted voting algorithm to determine a candidate set of fault locations near the connection points of aging pipelines.

[0067] S17. According to the updated fault features, use a long short-term memory network model to predict the probability distribution of pipeline faults to obtain high-probability fault locations.

[0068] S18. Calculate the average Hausdorff distance according to the candidate set of fault locations and the high-probability fault locations. When the average Hausdorff distance is lower than a preset distance threshold, re-determine the candidate set of fault locations to obtain the final fault location.

[0069] In step S11, sensor data during pipeline operation is collected, where the sensor data includes pressure, flow, and vibration data.

[0070] Specifically, pressure sensors deployed along the oil and gas pipeline measure the internal pressure change of the pipeline in real time. The flow sensor uses the ultrasonic time-difference method to detect the fluid velocity and calculate the volume flow. The vibration sensor collects the three-dimensional vibration signal of the pipeline through a MEMS accelerometer. The three types of sensors synchronously transmit the original data to the central processing unit through the industrial bus protocol to ensure that the timestamps of each monitoring point are strictly aligned.

[0071] This collection scheme enables multi-source sensors to work together, laying a foundation for the comprehensive analysis of subsequent pressure, flow, and vibration data, effectively avoiding monitoring blind spots that may be caused by a single data source, and providing reliable data guarantee for accurate fault location. Among them, the pressure data is used to analyze the pipeline pressure wave characteristics, the flow data is used to monitor the medium transportation state, and the vibration data is used to identify abnormal changes in the pipeline mechanical structure.

[0072] In step S12, data cleaning and feature extraction are performed according to the sensor data to obtain low-dimensional fault features.

[0073] In a specific implementation manner, the data cleaning and feature extraction according to the sensor data to obtain low-dimensional fault features includes:

[0074] Based on the sensor data, a linear interpolation algorithm is used to complete the data, obtaining multi-dimensional data;

[0075] Based on the multi-dimensional data, a principal component analysis method is used for dimensionality reduction processing and feature extraction, obtaining low-dimensional fault features.

[0076] Specifically, first, for the parts with missing values in the pressure data, flow data, and vibration data, a linear interpolation algorithm is used for completion. The linear interpolation algorithm calculates the data value at the missing position according to the adjacent valid data points before and after the missing value, ensuring the continuity of the time series. The pressure data, flow data, and vibration data after completion constitute multi-dimensional data.

[0077] Then, a principal component analysis method is used for dimensionality reduction processing on the completed multi-dimensional data. The three-dimensional data set composed of the pressure data, flow data, and vibration data completed by linear interpolation is used as the input matrix. First, calculate the covariance matrix of this matrix, which characterizes the linear correlation degree between the data of each dimension. Then, solve the eigenvalues and eigenvectors of the covariance matrix. The magnitude of the eigenvalues reflects the amount of original data information contained in the corresponding principal components. Arrange the eigenvalues in descending order, and the corresponding eigenvectors are the directions of the new orthogonal bases.

[0078] The cumulative contribution rate threshold is determined by analyzing the dimensionality reduction effect of historical data. The specific method is: statistically analyze the cumulative contribution rate distribution of each principal component on the historical data set, and select the number of principal components required to make the cumulative contribution rate reach 85% as the dimension threshold. For example, in most historical cases, the cumulative contribution rate of the first two principal components can meet this threshold requirement.

[0079] Select the eigenvectors corresponding to the first k largest eigenvalues to construct a projection matrix, where k is the smallest integer that makes the cumulative contribution rate reach the threshold. Multiply the original three-dimensional data matrix by the projection matrix to obtain the dimensionality-reduced two-dimensional principal component feature matrix. This matrix contains two dimensions: the pressure principal component feature and the flow principal component feature. Among them, the pressure principal component feature mainly reflects the comprehensive information of pipeline pressure fluctuations, and the flow principal component feature characterizes the overall trend of fluid state changes. The finally output low-dimensional fault features include the pressure principal component feature, the flow principal component feature, and the vibration principal component feature. These features not only retain the main information of the original data but also reduce the data dimension, providing an effective feature representation for subsequent fault analysis. This step solves the problems of sensor data missing and dimensionality disaster through data completion and dimensionality reduction processing, laying a foundation for accurately identifying pipeline faults.

[0080] In step S13, a Bayesian network model is constructed based on the low-dimensional fault features, and the conditional probabilities between the features are calculated according to the Bayesian network model to generate a feature correlation matrix.

[0081] In a specific embodiment, constructing a Bayesian network model according to the low-dimensional fault features and calculating the conditional probabilities between the features and generating a feature correlation matrix according to the Bayesian network model includes:

[0082] Based on the low-dimensional fault features, perform structure learning through the PC algorithm to determine the dependence relationship between the features, and use the maximum likelihood estimation method for parameter learning to determine the conditional probabilities between the features, and construct a Bayesian network;

[0083] Calculate the Pearson correlation coefficient matrix according to the Bayesian network to obtain the feature correlation matrix.

[0084] Specifically, in step S13, the specific process of constructing a Bayesian network model based on three types of low-dimensional fault features, namely pressure principal component features, flow principal component features, and vibration principal component features, is as follows: First, use the PC algorithm for network structure learning. In the initial stage, construct a fully connected candidate graph structure for the pressure, flow, and vibration principal component features, and then iteratively delete redundant edges through conditional independence tests. For each pair of feature combinations (such as the pressure principal component feature and the flow principal component feature), perform a partial correlation test under the condition of a given third feature (such as the vibration principal component feature). By calculating the correlation coefficient between the residual sequences and converting it into a statistic, the probability value representing the independence between the features is obtained. The preset significance threshold is determined by analyzing the independence test results of all feature pairs under historical normal operating conditions. The specific method is to statistically analyze the probability value distribution of truly independent feature pairs in normal data, and select the critical value with a misjudgment rate lower than 5% as the screening criterion. When the probability value obtained from the test is lower than this threshold, it is determined that there is a conditional dependence relationship between the features and the connection edge is retained. After completing the structure learning, use the maximum likelihood estimation method to statistically analyze the conditional probability distribution of each feature node under the value combination of its parent nodes. For example, calculate the occurrence probability of the pressure principal component feature when the flow principal component feature is in the high-fluctuation interval and the vibration principal component feature is in the low-frequency band. Based on the constructed Bayesian network structure, for the linear correlation degree between the pressure, flow, and vibration principal component features pairwise, calculate the Pearson correlation coefficient through the ratio of the covariance to the product of the feature standard deviations, and finally generate a feature correlation matrix containing three rows and three columns. Each element in this matrix represents the correlation coefficient of the corresponding feature pair. For example, the value in the second row and the third column represents the linear correlation degree between the flow principal component feature and the vibration principal component feature. This step quantifies the dynamic association pattern between multi-sensor features through a probabilistic graph model, eliminates the interference of redundant features. Its core value lies in constructing a feature dependence network that can accurately reflect the pipeline operation state, providing a probabilistic inference framework for subsequent weighted feature update and fault candidate region screening, thereby improving the accuracy of fault location.

[0085] The Bayesian network model established in this step can accurately characterize the probabilistic dependence relationship among pressure, flow rate, and vibration characteristics. The generated feature correlation matrix provides a probabilistic inference basis for subsequent fault feature analysis and location, realizing the mapping from data features to fault causal relationships.

[0086] In step S14, when the conditional probability is lower than a preset probability threshold, the corresponding eigenvalue is weighted and updated according to the low-dimensional fault feature to obtain an updated fault feature.

[0087] In a specific implementation manner, the step of when the conditional probability is lower than a preset probability threshold, weighting and updating the corresponding eigenvalue according to the low-dimensional fault feature to obtain an updated fault feature includes:

[0088] When the conditional probability between features is lower than a preset probability threshold, a linear regression algorithm is used to determine the weight coefficient of the low-dimensional fault feature and weight and update the corresponding eigenvalue;

[0089] Modify the low-dimensional fault feature according to the updated eigenvalue, and optimize the modified low-dimensional fault feature using the Kalman filter method to obtain an updated fault feature.

[0090] Specifically, when the conditional probability between the pressure principal component feature and the flow rate principal component feature, the conditional probability between the pressure principal component feature and the vibration principal component feature, or the conditional probability between the flow rate principal component feature and the vibration principal component feature is lower than a preset probability threshold, weighted update processing is performed on the corresponding low-dimensional fault feature. The probability threshold is determined by analyzing the conditional probability distribution of various feature combinations in historical normal operation data, and specifically, the 5th percentile of this distribution is used as the threshold boundary value.

[0091] The specific implementation process is as follows: First, take the pressure principal component feature, the flow rate principal component feature, and the vibration principal component feature as inputs. When it is detected that the conditional probability between any two features is lower than the preset threshold, the feature update process is triggered. Taking the update of the pressure principal component feature and the flow rate principal component feature as an example, a linear regression algorithm is used to establish a quantitative relationship between the two. The flow rate principal component feature is used as the explanatory variable, and the pressure principal component feature is used as the response variable. The regression coefficient of the flow rate principal component feature is calculated by minimizing the sum of squared residuals, and this coefficient is the weight coefficient. Use this weight coefficient to weightedly adjust the original flow rate principal component feature value to obtain a temporarily updated flow rate principal component feature value.

[0092] Subsequently, the temporarily updated traffic principal component eigenvalue is input into the Kalman filter together with the original pressure principal component eigenvalue and the vibration principal component eigenvalue. Based on the feature state estimation at the previous moment and the current observation value, the Kalman filter iteratively calculates through two stages of prediction and update, and finally outputs the updated traffic principal component eigenvalue, the updated pressure principal component eigenvalue, and the updated vibration principal component eigenvalue after noise suppression. These updated eigenvalues not only retain the main information of the original data but also eliminate abnormal interference, forming a more reliable updated fault feature set.

[0093] This step triggers feature update through conditional probability and uses a method combining linear regression and Kalman filtering to achieve dynamic correction of abnormal feature relationships, ensuring the accuracy and reliability of fault feature expression, and providing a more accurate feature basis for subsequent fault location. This adaptive feature update strategy based on probability threshold effectively solves the problem of false fault judgment caused by fixed features in traditional methods.

[0094] In step S15, according to the feature correlation matrix and the Bayesian network model, calculate the causal value and fusion score between the updated fault features, and generate pipeline fault features in combination with the updated fault features.

[0095] In a specific embodiment, the calculating the causal value and fusion score between the updated fault features according to the feature correlation matrix and the Bayesian network model, and generating pipeline fault features in combination with the updated fault features includes:

[0096] Extract the non-diagonal values of the feature correlation matrix. When the non-diagonal value is greater than a preset correlation threshold, analyze the causal value between features through the Bayesian network model to determine high-causal feature combinations;

[0097] According to the high-causal feature combinations, fuse the corresponding sensor data using the weighted average method to obtain a fusion score;

[0098] When the fusion score is greater than a preset fusion threshold, regard the high-causal feature combinations as high-fusion feature combinations;

[0099] According to the updated fault features corresponding to the high-fusion feature combinations, extract pipeline fault features using the principal component analysis method.

[0100] Specifically, first, the correlation coefficients between the pressure principal component features and the flow principal component features, between the pressure principal component features and the vibration principal component features, and between the flow principal component features and the vibration principal component features are extracted from the feature correlation matrix. The extracted correlation coefficients are the non-diagonal values of the feature correlation matrix. The correlation threshold is determined by taking the 95th percentile through analyzing the distribution characteristics of the correlation coefficients of each feature combination in the historical normal operation data. When the correlation coefficient of a certain feature combination exceeds this threshold, the causal analysis process is triggered.

[0101] Causal analysis process: Calculate the causal values of the highly correlated feature combinations through the Bayesian network model. Taking the pressure principal component features and the flow principal component features as an example, calculate the conditional probability change rate of the pressure principal component features with respect to the flow principal component features based on the network structure. This change rate is the causal value. According to the historical data analysis, mark the feature combinations with causal values greater than the 90th percentile in their distribution as highly causal feature combinations.

[0102] Data fusion is performed on the highly causal feature combinations using the weighted average method. The weight coefficients are determined after normalizing the causal values of each feature. Divide the causal values of the pressure principal component features, the flow principal component features, and the vibration principal component features by their sum respectively to obtain the fusion weights of each feature. Use these weights to perform weighted summation on the corresponding updated fault features to calculate the fusion score. The fusion threshold is determined by taking the intersection point of the two distributions by statistically analyzing the fusion score distributions of the normal conditions and the fault conditions in the historical data.

[0103] When the fusion score exceeds the threshold, mark this feature combination as a highly fused feature combination. Finally, perform principal component analysis on the updated pressure principal component features, the updated flow principal component features, and the updated vibration principal component features corresponding to the highly fused feature combination, and retain the principal components with a cumulative contribution rate reaching 85% as the final pipeline fault features.

[0104] The process of obtaining the cumulative contribution rate is as follows: Based on the updated pressure principal component features, updated flow principal component features, and updated vibration principal component features corresponding to the high-fusion feature combination, a three-dimensional data matrix is constructed. First, the covariance matrix of this matrix is calculated to characterize the linear correlation degree among the features. Subsequently, the eigenvalues of the covariance matrix are solved and arranged in descending order. The proportion of the variance amount explained by each principal component corresponding to each eigenvalue to the total variance amount is the contribution rate of this principal component. For example, if the three sorted eigenvalues are 2.8, 0.9, and 0.3 respectively, their contribution rates are 70%, 22.5%, and 7.5% respectively. The contribution rates are accumulated step by step from large to small. When the cumulative value first reaches or exceeds the preset 85% threshold (such as when the 70% of the first principal component and 22.5% of the second principal component are accumulated to 92.5%), the accumulation stops and the corresponding number of principal components is selected. This threshold is determined by analyzing the coverage ability of the principal components for the original feature information in historical fault cases. The specific method is to statistically analyze the improvement curve of the fault feature recognition accuracy under different cumulative contribution rates, and select the critical value corresponding to the inflection point of the accuracy increase (for example, when the cumulative contribution rate increases from 80% to 85%, the accuracy increase amplitude decreases significantly), so as to achieve a balance between the dimensionality reduction effect and information retention. Finally, the principal components with the cumulative contribution rate meeting the standard are retained as the pipeline fault features.

[0105] This step realizes the transformation from the original features to the fault features through multi-level threshold screening and feature fusion, which not only retains the key fault information but also reduces the data dimension, providing an effective feature representation for subsequent fault location. The entire process automatically determines each threshold based on historical data statistics without manual intervention, ensuring the objectivity and repeatability of the method.

[0106] In step S16, the pipeline fault features are matched with the topological structure and operation years in the pre-established pipeline geographic information database, and the weighted voting algorithm is used to determine the candidate set of fault locations near the connection points of the aging pipelines.

[0107] In a specific implementation manner, the matching of the pipeline fault features with the topological structure and operation years in the pre-established pipeline geographic information database and the use of the weighted voting algorithm to determine the candidate set of fault locations near the connection points of the aging pipelines include:

[0108] Obtain the topological data and operation years from the pre-established pipeline geographic information database;

[0109] Use the cosine similarity to calculate the topological matching degree between the pipeline fault features and the topological data;

[0110] According to the topological matching degree and the operation years, use the linear weighting method to calculate the aging coefficient, and estimate the fault probability near the pipeline connection points through the logistic regression algorithm;

[0111] The weighted voting algorithm is used to assign weights and perform weighted calculations on the failure probability and the topology matching degree to obtain a failure location score.

[0112] When the failure location score is greater than a preset score threshold, the corresponding location is included in the preliminary failure location candidate set.

[0113] According to the preliminary failure location candidate set, position calibration is performed through pre-stored pipeline coordinates and calibration parameters, and the least squares method is used to optimize the positioning accuracy to obtain a failure location candidate set.

[0114] Specifically, first, the topological connection relationship, spatial coordinate information, and operation years data of each pipe section are extracted from the pipeline geographic information database. When calculating the cosine similarity between the pipeline failure characteristics and the pipe section topological feature vector, first, the pressure principal component feature, flow principal component feature, and vibration principal component feature are combined into a three-dimensional failure feature vector (denoted as F), and at the same time, the connectivity, curvature, and slope of the target pipe section are combined into a three-dimensional topological feature vector (denoted as T). The connectivity represents the number of connections of the pipe section to other pipe sections (for example, the connectivity of a tee pipe section is 3), the curvature is obtained by calculating the ratio of the actual length of the pipe section to the straight-line length (for example, the ratio of a curved pipe is 1.2), and the slope is expressed as the percentage of the elevation change of the pipe section to the horizontal projection length (for example, a 2% slope). Both types of feature vectors need to be standardized to eliminate the dimension difference. The specific method is to perform min-max normalization on each feature dimension separately and map the values to the [0,1] interval. Subsequently, the cosine similarity between F and T is calculated. Its mathematical essence is the cosine value of the angle between two vectors in the standardized space, which is specifically realized by summing the products of the corresponding dimension values of the two vectors and then dividing by the product of the norms of their respective vectors. Since all features are non-negative after normalization, the calculation result ranges from [0,1], and it is directly used as the topology matching degree output, where 1 indicates that the failure characteristics are completely matched with the pipe section topology, and 0 indicates no association. This process quantifies the correlation strength between dynamic failure characteristics and static pipeline structures through geometric space projection, provides a comparable similarity index for the subsequent weighted voting algorithm, and thus deeply integrates sensor monitoring data with pipeline physical properties to improve the accuracy of screening the failure location candidate set.

[0115] Based on the operation years data of pipe segments, the linear weighted method is used to calculate the aging coefficient. During the calculation of the aging coefficient, first, according to the historical pipeline maintenance records and the statistics of fault events, the frequency distribution of faults occurring in pipe segments in different operation years intervals is analyzed. For example, it is statistically found that the number of faults occurring in pipe segments with an operation time exceeding 20 years accounts for 65% of all faults, those with 10 - 20 years account for 28%, and those with less than 10 years account for 7%. Based on this distribution law, through normalization, the fault frequency is converted into a weight coefficient, with the sum of weights being 1. The specific mapping relationship is as follows: pipe segments with an operation time exceeding 20 years are assigned a weight of 0.7, those with 10 - 20 years are assigned 0.4, and those with less than 10 years are assigned 0.1. The weight value is positively correlated with the fault frequency. In the training stage of the logistic regression model, each pipe segment sample in the historical data includes the topological matching degree (such as 0.85), the aging coefficient (such as 0.7), and the label (0 or 1) indicating whether the pipe segment has a fault. The model uses an iterative optimization algorithm to find the optimal parameters, so that the combination of the topological matching degree and the aging coefficient can best fit the actual probability of fault occurrence. For newly input pipe segment data, the topologically matching degree calculated in real time (such as 0.92) and the aging coefficient (such as 0.4) are combined into a two-dimensional feature vector and input into the model. Inside the model, through the mapping calculation of a non-linear function, the fault probability value (such as 0.83) near the connection point of the pipe segment is calculated. The higher this value is, the greater the fault risk under the current feature combination. Through quantifying the correlation between static attributes and dynamic features throughout the process, the physical state of the pipeline is deeply integrated with real-time monitoring data to achieve accurate estimation of the fault probability.

[0116] The weighted voting algorithm is used to comprehensively score the fault probability and topological matching degree. During the weight assignment process of the weighted voting algorithm, first, the optimal weight combination of the fault probability and topological matching degree is determined by the grid search cross-validation method. Specifically, when implementing, the historical data set is divided into a training set and a validation set in chronological order, and different weight combinations are traversed (for example, the weight of the fault probability ranges from 0.5 to 0.7 with a step of 0.05, and the weight of the topological matching degree correspondingly ranges from 0.5 to 0.3). For each weight combination, the recall rate and precision rate of the fault location candidate set in the validation set are calculated, and the weight parameters that maximize the harmonic mean (F1 value) of the two are selected (such as finally determining the fault probability weight of 0.6 and the topological matching degree weight of 0.4). In the comprehensive scoring calculation stage, for each pipe segment connection point, the fault probability value output by its logistic regression model is multiplied by 0.6, and at the same time, the topological matching degree obtained by calculating the cosine similarity is multiplied by 0.4, and the two are added to obtain a comprehensive score value within the range of 0 to 1. The preset scoring threshold is determined by statistically analyzing the comprehensive score distribution of all pipe segment connection points in historical fault events. The specific method is as follows: collect the comprehensive score data of the pipe segments where actual faults occurred in the past five years before the fault occurred, and calculate the 80th percentile value of these score values (for example, when the 80% percentile of the scores of historical fault pipe segments is 0.72, set the threshold to 0.72). When the real-time calculated comprehensive score exceeds this threshold, it is determined that this position has high-risk characteristics and is included in the preliminary candidate set.

[0117] In the position calibration and optimization stage, the system calls the coordinate calibration parameters stored in the pipeline geographic information database (including the coordinate system transformation matrix, measurement device error coefficient, etc.) to correct the error of each position coordinate in the preliminary candidate set. Based on the known accurate coordinates of three monitoring points adjacent to the target position (such as the upstream monitoring point A, the downstream monitoring point B, and the lateral reference point C), a three-dimensional error model including the plane coordinates X, Y, and elevation Z is established. The spatial residuals between the target position coordinates and the three reference points are calculated iteratively by the least squares method, and the coordinate offset is dynamically adjusted until the sum of the squared residuals reaches the minimum value, and finally, the accurate coordinates after spatial error compensation are output. This process effectively eliminates the systematic positioning errors caused by factors such as sensor installation deviation and cumulative geological settlement, and improves the positioning accuracy of the fault location candidate set from the ten-meter level to the meter level.

[0118] This step realizes the preliminary positioning of the fault location by fusing the static attributes and dynamic monitoring features of the pipeline, providing a reliable candidate area for subsequent precise positioning. The parameters and thresholds determined based on historical data statistics ensure the objectivity and accuracy of the method, and the least squares optimization further improves the positioning accuracy. The whole process makes full use of the spatial topology information and operation history data of the pipeline, effectively solving the limitations of relying solely on sensor data for fault location.

[0119] In step S17, according to the updated fault characteristics, a long short-term memory network model is used to predict the probability distribution of pipeline fault occurrence, and the high-probability fault location is obtained.

[0120] Specifically, the updated pressure principal component features, updated flow principal component features, and updated vibration principal component features arranged in time series are used as model inputs to construct a multi-layer LSTM network architecture. This network is obtained through training with historical monitoring data. During the training process, a sliding window method is used to process the time series data, and the window length covers a complete working condition change cycle. The monitoring data corresponding to each time step is input into the network after being normalized.

[0121] The model training adopts an optimization strategy with an early stopping mechanism, and the training process is terminated when the performance index of the validation set no longer improves. In the prediction stage, the updated feature data collected in real time is input into the trained LSTM model after passing through the same preprocessing process, and the fault occurrence probability of each pipeline section is output. The probability threshold is determined by analyzing the performance of the model on an independent validation set, and the optimal critical value that can balance the detection sensitivity and false alarm rate is selected.

[0122] During the training process of the LSTM model, the implementation process of the optimization strategy with an early stopping mechanism is as follows: First, the preprocessed historical data set is divided into a training set, a validation set, and a test set. The training set is used to update the model parameters, and the validation set is used to monitor the generalization performance of the model. At the beginning of each training epoch, the model sequentially reads the training set samples (such as the updated pressure principal component features, flow principal component features, and vibration principal component feature sequences in the past 72 hours), calculates the predicted failure probability through forward propagation, and updates the network weights based on the difference between the prediction result and the true failure label. After each epoch, the model calculates the performance metrics (such as the validation set loss value or F1 score) on the validation set and records the change of the current metric relative to the historical best value. The preset patience parameter determines the maximum number of consecutive epochs allowed for the validation set performance to stagnate (for example, patience = 10). When the validation set loss does not decrease within 10 consecutive epochs, the early stopping mechanism is triggered to terminate the training. During the training process, a copy of the model parameters with the best performance on the validation set is always saved (such as the model state at the 50th epoch). Even if the performance decreases in subsequent epochs, this best state is still retained for the final prediction. The setting of the early stopping threshold is determined by analyzing the convergence characteristics of the validation set loss curve during the historical training process. For example, the failure detection recall rate and false alarm rate of the model on the independent test set are statistically analyzed under different patience values, and the critical value (such as patience = 10) that makes the recall rate not less than 90% and the false alarm rate not higher than 5% is selected. This mechanism terminates the training before overfitting occurs by dynamically judging the upper limit of the model's learning ability, ensuring that the model maintains the generalization ability for unknown data in the fault prediction task and providing a stable and reliable prediction result basis for subsequent probability threshold optimization.

[0123] For the pipeline sections whose predicted probability exceeds the threshold, extract their spatial location information as high-probability fault regions. At the same time, save the characteristic change patterns in the corresponding time periods, including the fluctuation amplitude of the pressure principal component features, the offset trend of the flow principal component features, and the abnormal frequency band distribution of the vibration principal component features, to form a complete fault characteristic description. This prediction process makes full use of the LSTM network's modeling ability for time series features to achieve early identification and spatial location of potential faults.

[0124] In step S18, calculate the average Hausdorff distance according to the candidate fault location set and the high-probability fault location. When the average Hausdorff distance is lower than the preset distance threshold, re-determine the candidate fault location set to obtain the final fault location.

[0125] In a specific implementation manner, calculating the average Hausdorff distance according to the candidate set of fault positions and the high-probability fault positions, and when the average Hausdorff distance is lower than a preset distance threshold, re-determining the candidate set of fault positions to obtain the final fault position, including:

[0126] Calculating the average Hausdorff distance according to the candidate set of fault positions and the high-probability fault positions, and when the average Hausdorff distance is lower than a preset distance threshold, using a logistic regression algorithm to adjust the weights of the weighted voting algorithm to generate updated second weights for the fault probability and topological matching degree;

[0127] Recalculating the fault position score according to the second weights and re-analyzing the candidate set of fault positions;

[0128] When iteratively updating until the average Hausdorff distance is higher than or equal to the distance threshold, determining the candidate set of fault positions of the last iteration as the final candidate set of fault positions;

[0129] Taking the fault position with the highest fault position score from the final candidate set of fault positions as the final fault position.

[0130] Specifically, first calculate the Hausdorff distance between each candidate position in the candidate set of fault positions and the high-probability fault position. When calculating the average Hausdorff distance between the candidate set of fault positions and the high-probability fault position, the specific process is as follows: The Hausdorff distance is essentially a measure of the maximum and minimum distances between two point sets. First, for each candidate coordinate point in the candidate set of fault positions (for example, point A with coordinates (x1, y1)), calculate its Euclidean distance to all points in the high-probability fault position set, and select the minimum distance value from them (for example, the distance from point A to the nearest point B in the high-probability position set is 15 meters). After traversing all points in the candidate set, obtain a set of minimum distance values D1 (for example, containing 10 minimum distance values corresponding to 10 points in the candidate set). Similarly, for each point in the high-probability fault position set, calculate its minimum distance to all points in the candidate set of fault positions, forming another set of minimum distance values D2. The final average Hausdorff distance is calculated as the arithmetic mean of all distance values in sets D1 and D2 (for example, the average value of D1 is 18 meters, the average value of D2 is 20 meters, then the average Hausdorff distance is 19 meters). The preset distance threshold is determined by statistically analyzing the distribution of the average Hausdorff distance between the correctly diagnosed positions and the false alarm positions in historical fault cases. For example, by analyzing that the average distance of correctly matched cases is concentrated within 25 meters, while the average distance of false alarm cases is mostly distributed above 40 meters, 30 meters is selected as the discrimination threshold. When the average Hausdorff distance calculated in real time is lower than this threshold, it indicates that the spatial distribution consistency between the candidate set and the high-probability position is insufficient, and the weight adjustment process is triggered. This measurement method effectively avoids the interference of single-point abnormal matching on the overall evaluation through spatial distance averaging processing.

[0131] When the average Hausdorff distance is lower than this distance threshold, the weight adjustment process is triggered. The logistic regression algorithm is used to re-analyze the contribution degrees of the fault probability and the topological matching degree to the positioning result, and the weight coefficients of these two features are updated based on the verification result of the current iteration. Among them, the new weight coefficient of the fault probability is determined by its spatial correlation with the high-probability fault position, and the new weight coefficient of the topological matching degree is determined by its spatial consistency with the candidate position.

[0132] Use the updated weight coefficients to recalculate the comprehensive scores of each candidate position, generating a new candidate set of fault positions. This process is iteratively executed, and the average Hausdorff distance between the new candidate set and the high-probability fault position is recalculated after each iteration. The iteration termination condition is set as the average Hausdorff distance reaching or exceeding the preset threshold. At this time, it is considered that the candidate position and the predicted position have met the spatial consistency requirement.

[0133] Finally, the position with the highest comprehensive score is selected from the candidate set that meets the distance requirement as the determined fault position. The spatial coordinates of this position are combined with the mileage stake number information in the pipeline geographic information database to output the positioning result including the pipeline section number and the accurate stake number. Through spatial distance measurement and dynamic weight adjustment in the entire verification process, the consistency between the positioning result and multi-source prediction evidence is ensured, effectively improving the reliability of fault location.

[0134] Referring Figure 2 , the second embodiment of the present invention provides a multi-sensor-based oil and gas pipeline fault location system, including:

[0135] A data acquisition module for acquiring sensor data during pipeline operation, where the sensor data includes pressure, flow, and vibration data;

[0136] A data processing module for performing data cleaning and feature extraction based on the sensor data to obtain low-dimensional fault features;

[0137] A feature association module for constructing a Bayesian network model based on the low-dimensional fault features, and calculating the conditional probability between features and generating a feature correlation matrix according to the Bayesian network model;

[0138] A feature update module for, when the conditional probability is lower than a preset probability threshold, performing weighted update on the corresponding feature values according to the low-dimensional fault features to obtain updated fault features;

[0139] A fault feature analysis module for calculating the causal value and fusion score between the updated fault features according to the feature correlation matrix and the Bayesian network model, and generating pipeline fault features in combination with the updated fault features;

[0140] A fault location analysis module for matching the pipeline fault features with the topological structure and operation years in the pre-established pipeline geographic information database, and using a weighted voting algorithm to determine a candidate set of fault positions near the connection points of aging pipelines;

[0141] A fault probability analysis module for predicting the probability distribution of pipeline faults using a long short-term memory network model according to the updated fault features to obtain high-probability fault positions;

[0142] A fault location determination module for calculating the average Hausdorff distance between the candidate set of fault positions and the high-probability fault positions. When the average Hausdorff distance is lower than a preset distance threshold, re-determining the candidate set of fault positions to obtain the final fault position.

[0143] It should be noted that a multi-sensor-based oil and gas pipeline fault location device provided by an embodiment of the present invention is used to execute all process steps of a multi-sensor-based oil and gas pipeline fault location method in the above embodiment. The working principles and beneficial effects of the two correspond one by one, and thus will not be elaborated here.

[0144] An embodiment of the present invention further provides an electronic device. The electronic device includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a multi-sensor-based oil and gas pipeline fault location program. When the processor executes the computer program, it implements the steps in each of the above embodiments of the multi-sensor-based oil and gas pipeline fault location method, such as Figure 1 the step S11 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in each of the above device embodiments, such as a multi-sensor-based oil and gas pipeline fault location module.

[0145] Exemplarily, the computer program may be divided into one or more modules / units. The one or more modules / units are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing specific functions, and these instruction segments are used to describe the execution process of the computer program in the electronic device.

[0146] The electronic device may be a computing device such as a desktop computer, a notebook, a palm computer, and a smart tablet. The electronic device may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the above components are only examples of the electronic device and do not constitute a limitation on the electronic device. It may include more or fewer components than the above, or combine some components, or different components. For example, the electronic device may further include input / output devices, network access devices, a bus, etc.

[0147] The so-called processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The processor is the control center of the electronic device and connects various parts of the entire electronic device through various interfaces and circuits.

[0148] The memory can be used to store the computer programs and / or modules. The processor realizes various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory, and by calling the data stored in the memory. The memory mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for at least one function (such as a sound playback function, an image playback function, etc.); the data storage area can store data created according to the use of the mobile phone (such as audio data, phone book, etc.). In addition, the memory can include high-speed random access memory, and can also include non-volatile memory, such as a hard disk, memory, plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, at least one magnetic disk storage device, flash memory device, or other volatile solid-state storage devices.

[0149] Among them, if the modules / units integrated in the electronic device are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, to implement all or part of the processes in the above-described embodiment methods of the present invention, it can also be completed by a computer program instructing relevant hardware. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above-described various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0150] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, in the attached drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that they have a communication connection, which can be specifically implemented as one or more communication buses or signal lines. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0151] The specific embodiments described above have further elaborated on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above description is only the specific embodiments of the present invention and is not used to limit the protection scope of the present invention. It is particularly pointed out that for those skilled in the art, any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A multi-sensor based oil and gas pipeline fault location method, characterized in that, Including: Collecting sensor data during the operation of the pipeline; Performing data cleaning and feature extraction based on the sensor data to obtain low-dimensional fault features; Constructing a Bayesian network model based on the low-dimensional fault features, and calculating the conditional probability between features and generating a feature correlation matrix according to the Bayesian network model; When the conditional probability is lower than a preset probability threshold, updating the corresponding feature value by weighting according to the low-dimensional fault features to obtain updated fault features; Calculating the causal value and fusion score between the updated fault features according to the feature correlation matrix and the Bayesian network model, and generating pipeline fault features in combination with the updated fault features; Matching the pipeline fault features with the topological structure and operation years in the pre-established pipeline geographic information database, and using a weighted voting algorithm to determine a candidate set of fault positions near the connection points of the aging pipeline; According to the updated fault features, using a long short-term memory network model to predict the probability distribution of pipeline faults to obtain high-probability fault positions; Calculating the average Hausdorff distance according to the candidate set of fault positions and the high-probability fault positions. When the average Hausdorff distance is lower than a preset distance threshold, re-determining the candidate set of fault positions to obtain the final fault position; Among them, the calculating the causal value and fusion score between the updated fault features according to the feature correlation matrix and the Bayesian network model, and generating pipeline fault features in combination with the updated fault features includes: Extracting the non-diagonal values of the feature correlation matrix. When the non-diagonal values are greater than a preset correlation threshold, analyzing the causal value between features through the Bayesian network model to determine a high-causal feature combination; Fusing the corresponding sensor data by using a weighted average method according to the high-causal feature combination to obtain a fusion score; When the fusion score is greater than a preset fusion threshold, taking the high-causal feature combination as a high-fusion feature combination; Extracting pipeline fault features by using the principal component analysis method according to the updated fault features corresponding to the high-fusion feature combination; Among them, the matching the pipeline fault features with the topological structure and operation years in the pre-established pipeline geographic information database, and using a weighted voting algorithm to determine a candidate set of fault positions near the connection points of the aging pipeline includes: Obtaining topological data and operation years from the pre-established pipeline geographic information database; Calculating the topological matching degree between the pipeline fault features and the topological data by using cosine similarity; Calculating an aging coefficient by using a linear weighting method according to the topological matching degree and the operation years, and estimating the fault probability near the pipeline connection points through a logistic regression algorithm; Using a weighted voting algorithm to perform weight assignment and weighted calculation on the fault probability and the topological matching degree to obtain a fault position score; When the fault position score is greater than a preset score threshold, including the corresponding position in the preliminary candidate set of fault positions; According to the preliminary candidate set of fault positions, performing position calibration through the pre-stored pipeline coordinates and calibration parameters, and optimizing the positioning accuracy by using the least squares method to obtain a candidate set of fault positions.

2. The multi-sensor based oil and gas pipeline fault location method according to claim 1, wherein Performing data cleaning and feature extraction based on the sensor data to obtain low-dimensional fault features, including: Based on the sensor data, using a linear interpolation algorithm to complete data filling to obtain multi-dimensional data; Based on the multi-dimensional data, using the principal component analysis method to perform dimensionality reduction processing and feature extraction to obtain low-dimensional fault features.

3. The multi-sensor-based oil and gas pipeline fault location method according to claim 1, wherein Constructing a Bayesian network model based on the low-dimensional fault features, and calculating the conditional probability between features and generating a feature correlation matrix according to the Bayesian network model, including: Based on the low-dimensional fault features, performing structure learning through the PC algorithm to determine the dependence relationship between features, using the maximum likelihood estimation method to perform parameter learning, determining the conditional probability between features, and constructing a Bayesian network; Calculating the Pearson correlation coefficient matrix according to the Bayesian network to obtain the feature correlation matrix.

4. The multi-sensor-based oil and gas pipeline fault location method according to claim 1, characterized in that, When the conditional probability is lower than a preset probability threshold, updating the corresponding feature value by weighting according to the low-dimensional fault features to obtain updated fault features, including: When the conditional probability between features is lower than a preset probability threshold, using a linear regression algorithm to determine the weight coefficient of the low-dimensional fault features, and performing weighted update of the corresponding feature value; Modifying the low-dimensional fault features according to the updated feature value, and optimizing the modified low-dimensional fault features by the Kalman filtering method to obtain updated fault features.

5. The multi-sensor based oil and gas pipeline fault location method according to claim 1, characterized in that Calculating the average Hausdorff distance according to the fault location candidate set and the high-probability fault location. When the average Hausdorff distance is lower than a preset distance threshold, re-determining the fault location candidate set to obtain the final fault location, including: Calculating the average Hausdorff distance according to the fault location candidate set and the high-probability fault location. When the average Hausdorff distance is lower than a preset distance threshold, using a logistic regression algorithm to adjust the weight of the weighted voting algorithm to generate a second weight for the updated fault probability and topological matching degree; Recalculating the fault location score according to the second weight, and re-analyzing the fault location candidate set; When iteratively updating repeatedly until the average Hausdorff distance is higher than or equal to the distance threshold, determining the fault location candidate set of the last iteration as the final fault location candidate set; Taking the fault location with the highest fault location score in the final fault location candidate set as the final fault location.

6. A multi-sensor-based oil and gas pipeline fault location system, characterized in that, Used to implement the multi-sensor-based oil and gas pipeline fault location method as described in any one of claims 1 to 5, including: A data acquisition module for acquiring sensor data during pipeline operation, where the sensor data includes pressure, flow, and vibration data; A data processing module for performing data cleaning and feature extraction based on the sensor data to obtain low-dimensional fault features; A feature association module for constructing a Bayesian network model based on the low-dimensional fault features, and calculating the conditional probability between features and generating a feature correlation matrix according to the Bayesian network model; A feature update module for updating the corresponding feature value by weighting according to the low-dimensional fault features when the conditional probability is lower than a preset probability threshold to obtain updated fault features; A fault feature analysis module, configured to calculate the causal value and the fusion score between the updated fault features according to the feature correlation matrix and the Bayesian network model, and generate pipeline fault features in combination with the updated fault features; A fault location analysis module, configured to match the pipeline fault features with the topological structure and the operation years in a pre-established pipeline geographic information database, and determine a fault location candidate set near the connection point of the aging pipeline by using a weighted voting algorithm; A fault probability analysis module, configured to predict the probability distribution of pipeline faults by using a long short-term memory network model according to the updated fault features, and obtain high-probability fault locations; A fault location determination module, configured to calculate the average Hausdorff distance according to the fault location candidate set and the high-probability fault locations, and re-determine the fault location candidate set when the average Hausdorff distance is lower than a preset distance threshold, so as to obtain the final fault location.

7. An electronic device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the multi-sensor-based oil and gas pipeline fault location method according to any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein when the computer program runs, it controls the device where the computer-readable storage medium is located to execute the multi-sensor-based oil and gas pipeline fault location method according to any one of claims 1 to 5.

Images