Gas pipeline health management method based on dynamic fault tree and multi-source data fusion

Abstract

The invention discloses a gas pipeline health management method based on dynamic fault tree and multi-source data fusion, and relates to the technical field of gas pipeline health management, and the method comprises the following steps: S1, collecting multi-source heterogeneous data required by gas pipeline health management; s2, based on the multi-source heterogeneous data, constructing a gas pipeline dynamic fault tree; s3, the fault diagnosis model performs gas pipeline fault diagnosis based on the multi-source heterogeneous data and the dynamic fault tree; and S4, based on a gas pipeline fault diagnosis result of the fault diagnosis model, carrying out gas pipeline fault alarm and early warning. According to the method, the multi-source heterogeneous data of the gas pipeline are monitored and analyzed in real time, so that potential gas pipeline faults are identified, diagnosed and predicted. An advanced data analysis technology is utilized, and historical operation data and real-time data of a gas pipeline are combined, so that an intelligent method capable of automatically detecting abnormal modes and fault symptoms is constructed. The reliability and the safety of the gas pipeline are improved, and the accidental shutdown time is shortened.

Description

Technical Field

[0001] The present invention relates to the technical field of gas pipeline health management, and more particularly to a gas pipeline health management method based on dynamic fault tree and multi-source data fusion. Background Art

[0002] Gas pipeline health management is a systematic, scientific approach designed to ensure safe, reliable, and efficient operation of gas pipelines through monitoring, assessment, and maintenance throughout their entire lifecycle (design, construction, operation, maintenance, and decommissioning). Its core objectives are to prevent accidents (such as leaks and explosions), extend pipeline life, reduce operating costs, and ensure public safety and environmental sustainability.

[0003] In existing technologies, gas pipeline health management often involves manual inspections, visual checks, and leak detection using handheld devices. These methods are inefficient, have limited coverage, rely on personnel experience, and are unable to predict future failures. Summary of the Invention

[0004] To overcome the shortcomings of the aforementioned prior art, the present invention discloses a gas pipeline health management method based on a dynamic fault tree and multi-source data fusion. This invention aims to address the inefficiencies, limited coverage, reliance on personnel experience, and inability to predict future failures of the prior art methods of leak detection, such as manual inspections, visual checks, and handheld devices. Based on a dynamic fault tree and multi-source data fusion, the present invention utilizes advanced data analysis techniques, combined with historical and real-time equipment operating data, to construct an intelligent method capable of automatically detecting abnormal patterns and fault signs, thereby improving the reliability and safety of gas pipelines, reducing unplanned downtime, and optimizing maintenance plans.

[0005] In order to achieve the above objectives, the present invention adopts the following technical solutions:

[0006] The gas pipeline health management method based on dynamic fault tree and multi-source data fusion includes the following steps:

[0007] 1. Collecting Multi-Source Heterogeneous Data

[0008] S1. Collect multi-source heterogeneous data required for gas pipeline health management;

[0009] Preferably, in step S1, the multi-source heterogeneous data includes basic sensor data, text record data and environmental meteorological data;

[0010] in:

[0011] The basic sensor data are various parameters generated by the gas pipeline pressure regulating equipment during operation, including: inlet pressure, outlet pressure, flow, temperature, and the switch status of the relief valve and the shut-off valve;

[0012] The text record data includes: long-term historical records of pressure, flow, temperature, and text records of maintenance work orders, accident reports, and inspection logs;

[0013] The environmental meteorological data includes: temperature, humidity, rainfall, wind speed, and extreme weather events. The environmental meteorological data is obtained from the National Meteorological Administration and is used to determine the impact of corrosion, stress changes, and external shocks on gas pipelines and to construct a dynamic fault tree, including:

[0014] Corrosion rate: High humidity + high temperature accelerates the oxidation of metal gas pipelines;

[0015] Soil moisture: Rainfall affects soil expansion and contraction, leading to stress changes in gas pipelines;

[0016] Extreme weather: Heavy rains cause ground subsidence, and typhoons increase the risk of external mechanical shocks.

[0017] 2. Building a dynamic fault tree

[0018] S2. Construct a dynamic fault tree for gas pipelines based on multi-source heterogeneous data;

[0019] Preferably, in step S2, when constructing the dynamic fault tree of the gas pipeline, a basic fault tree is first established, and then the dynamic fault tree of the gas pipeline is established by combining the basic fault tree, text record data in the multi-source heterogeneous data, and environmental meteorological data;

[0020] The data recorded in combination with text includes:

[0021] Perform text cleaning and word segmentation and part-of-speech tagging on text record data, calculate term frequency-inverse document frequency using the TF-IDF algorithm, and filter high-frequency fault keywords in the text record data;

[0022] Use pre-trained models to identify faulty equipment components and types corresponding to high-frequency fault keywords;

[0023] Dynamically integrate the identified faulty equipment components and types with the basic fault tree, and dynamically update the fault event library to obtain a dynamic fault tree, including:

[0024] Add the identified faulty equipment components and types as new event nodes to the basic fault tree;

[0025] Adjust the fault event weights based on the frequency of occurrence of faulty equipment components and types in text record data;

[0026] Among them, combined with environmental meteorological data include:

[0027] The environmental meteorological data of the address where the gas pipeline is located is obtained through the China Meteorological Administration, and the environmental meteorological data is used to build an environmental risk model, which is then integrated with the dynamic fault tree.

[0028] Preferably, in step S2, the environmental risk model includes a dynamic corrosion model, and the dynamic corrosion model includes:

[0029] R(t)=K·T(t)·H(t)

[0030] Where R(t) is the environmental risk model; K is the soil corrosion coefficient, obtained from the regional environmental protection report; T(t) is the real-time temperature, °C; H(t) is the real-time humidity, %.

[0031] In the application of the dynamic corrosion model, the accuracy of the integral algorithm is adjusted according to specific needs, time correlation is added to the corrosion rate function, and an exception handling mechanism is added.

[0032] Preferably, in step S2, the integration of the environmental risk model with the dynamic fault tree includes: heavy rain triggers soil settlement, if the rainfall for three consecutive hours is greater than 50 mm, the probability of a soil settlement event increases by 20%; high temperature accelerates aging, if the temperature is greater than 40°C, the weight of the sealing ring aging event increases by 30%.

[0033] 3. Fault diagnosis

[0034] S3. The fault diagnosis model performs gas pipeline fault diagnosis based on multi-source heterogeneous data and a dynamic fault tree, including fixed-point analysis, trend analysis, and predictive analysis of the gas pipeline. The fixed-point analysis includes determining whether the gas pipeline is abnormal by analyzing multi-source heterogeneous data at a specific time point. The trend analysis includes detecting abnormalities in the gas pipeline by analyzing the time-varying trends of key parameters in the multi-source heterogeneous data. The predictive analysis includes predicting the future status and possible faults of the gas pipeline by analyzing the multi-source heterogeneous data.

[0035] 3.1 Fixed point analysis

[0036] Preferably, in step S3, the fixed-point analysis is achieved through data processing and statistical analysis, including:

[0037] Calculate the mean, median, or other statistical indicators at a specific time point;

[0038] Analyze the relationship between two variables using linear regression, implemented with the Apache Commons Math library.

[0039] Preferably, in step S3, in the linear regression, a multivariate linear regression model includes:

[0040] Y i =β0+β1X i1 +β2X i2 +...+β p X ip +ε i ,i=1,...,n

[0041] Among them, Y i is the regressor of the linear regression model; X i1 ,X i2 ...X ip is the regressor of the linear regression model; β0,β1,β2...β p is the regression coefficient of the linear regression model; ε i is the error term of the linear regression model.

[0042] 3.2 Trend Analysis

[0043] Preferably, in step S3, the trend analysis includes:

[0044] Pipeline blockage risk prediction: Calculate the N-day moving average of the inlet pressure. If it shows a continuous upward trend and the flow rate shows a downward trend during the same period, it is determined that the pipeline is blocked and a maintenance inspection work order is triggered;

[0045] Corrosion risk prediction: Key parameters include corrosion rate and soil moisture. Linear regression is used to fit the change in corrosion depth over time. Combined with rainfall data, the corrosion model weight is dynamically adjusted to predict the corrosion depth for the next N days. The corrosion depth is then compared with the corrosion depth threshold. If the threshold is greater than the threshold, corrosion risk exists, triggering the early arrangement of coating repairs.

[0046] Preferably, in step S3, the algorithm used for trend analysis includes a moving average algorithm, which smooths short-term fluctuations and highlights long-term trends by calculating the moving average of data points. The moving average algorithm includes:

[0047]

[0048] Among them, MA t is the moving average at time point t; n is the window size of the moving average, the number of periods; x t-i is the actual observation value at time point ti.

[0049] Preferably, in step S3, the algorithm used for trend analysis includes an exponential smoothing algorithm, which processes time series data and is implemented using the Apache Commons Math library. The exponential smoothing algorithm includes:

[0050] S t =αX t+(1-α)S t-1

[0051] Among them, S t is the smoothed value at time point t; X t is the actual observation value at time point t; S t-1 is the smoothing value at time point t-1; α is the smoothing parameter, which ranges from 0 to 1 and is determined by the user based on the data set and prediction target.

[0052] 3.3 Predictive Analysis

[0053] Preferably, in step S3, the predictive analysis includes:

[0054] The time window is input through the LSTM model, and the regression prediction outputs the parameter prediction value at the future time point;

[0055] If the predicted value exceeds the safety threshold, the over-limit probability is calculated, and the fault confidence is calculated based on the prediction error distribution.

[0056] Preferably, in step S3, the fault confidence level includes:

[0057]

[0058] Where P(fault) is the fault confidence level, μ is the predicted mean, σ is the standard deviation, and Φ is the standard normal distribution function.

[0059] 4. Fault warning and alarm

[0060] S4. Based on the gas pipeline fault diagnosis results of the fault diagnosis model, gas pipeline fault alarms and early warnings are performed, and fault-specific solutions and maintenance suggestions are generated.

[0061] Preferably, in step S4, for the analyzed fault, a fault alarm is sent to the front end; for the predicted fault, a fault warning is sent to the front end; the staff is notified of the fault or warning that has occurred, and the staff notifies the maintenance personnel of the fault information by issuing a task order to perform maintenance based on the fault-specific solution and maintenance suggestions.

[0062] Beneficial effects of the present invention:

[0063] The gas pipeline health management method proposed in this paper identifies, diagnoses, and predicts potential gas pipeline failures through real-time monitoring and analysis of multi-source heterogeneous data. Leveraging advanced data analysis techniques and combining historical and real-time gas pipeline operational data, it develops an intelligent approach capable of automatically detecting abnormal patterns and signs of failure. This approach improves gas pipeline reliability and safety, reduces unplanned downtime, and optimizes maintenance schedules.

[0064] The gas pipeline health management method provided by this invention utilizes a dynamic fault tree to intuitively display the various possible pathways and causal relationships of fault occurrence, thereby understanding the fault's mechanism. Qualitative and quantitative analysis of the fault tree allows for rapid identification of key fault factors, enabling targeted measures to be taken, improving the reliability and safety of the present invention. The dynamic fault tree is constructed by utilizing textual record data and environmental meteorological data, and dynamically adjusted. The fault diagnosis model, based on the dynamic fault tree, performs gas pipeline fault diagnosis, resulting in more accurate fault diagnosis.

[0065] The gas pipeline health management method provided by this invention utilizes multi-source heterogeneous data, including basic sensor data, textual record data, and environmental meteorological data. The fault diagnosis model performs gas pipeline fault diagnosis based on this multi-source heterogeneous data. This integration of multiple data types results in more accurate fault diagnosis. Furthermore, fault diagnosis includes fixed-point analysis, trend analysis, and predictive analysis, offering multiple functions. This method not only determines whether a gas pipeline is abnormal by analyzing multi-source heterogeneous data at a specific point in time, but also identifies gas pipeline anomalies by analyzing the changing trends of key parameters in this multi-source heterogeneous data over time. Furthermore, this multi-source heterogeneous data can be used to predict the future status and potential faults of the gas pipeline. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 Schematic diagram of the gas pipeline health management method of the present invention;

[0067] Figure 2 This is the feature importance analysis of the random forest model of the present invention;

[0068] Figure 3 A fault tree for determining a series voltage fault according to the present invention;

[0069] Figure 4 The prediction effect of the dynamic corrosion model of the present invention;

[0070] Figure 5 This is a schematic diagram of a non-pressure fault of a top-level event in the present invention;

[0071] Figure 6 This is a schematic diagram of the dynamic branching of the equipment failure of the present invention;

[0072] Figure 7 This is a schematic diagram of the dynamic branching of a pressure gauge failure according to the present invention;

[0073] Figure 8 This is a schematic diagram of the dynamic branching of the valve with loose sealing according to the present invention;

[0074] Figure 9 The impact of the dynamic threshold adjustment on the false alarm rate of the present invention;

[0075] Figure 10This is a simple code example of the moving average of the present invention;

[0076] Figure 11 This is a simple code example of exponential smoothing of the present invention;

[0077] Figure 12 This is the prediction effect of the LSTM model of the present invention on the gas pipeline outlet pressure. DETAILED DESCRIPTION

[0078] The following will provide a clear and complete description of the concept, specific structure and technical effects of the present invention in conjunction with the embodiments and drawings, so as to fully understand the purpose, features and effects of the present invention.

[0079] Example 1

[0080] Gas pipeline health management method based on dynamic fault tree and multi-source data fusion, such as Figure 1 As shown, the following steps are included:

[0081] S1. Collect multi-source heterogeneous data required for gas pipeline health management;

[0082] S2. Construct a dynamic fault tree for gas pipelines based on multi-source heterogeneous data;

[0083] S3. The fault diagnosis model performs gas pipeline fault diagnosis based on multi-source heterogeneous data and a dynamic fault tree, including fixed-point analysis, trend analysis, and predictive analysis of the gas pipeline. The fixed-point analysis includes determining whether the gas pipeline is abnormal by analyzing multi-source heterogeneous data at a specific time point. The trend analysis includes detecting abnormalities in the gas pipeline by analyzing the time-varying trends of key parameters in the multi-source heterogeneous data. The predictive analysis includes predicting the future status and possible faults of the gas pipeline by analyzing the multi-source heterogeneous data.

[0084] S4. Based on the gas pipeline fault diagnosis results of the fault diagnosis model, gas pipeline fault alarms and early warnings are performed, and fault-specific solutions and maintenance suggestions are generated.

[0085] The fault diagnosis model in this invention is a comprehensive set of diagnostic tools and methods designed to identify, diagnose, and predict potential equipment failures by monitoring and analyzing key performance parameters of gas pressure regulating equipment in real time. The core of this model is the use of advanced data analysis techniques, combined with historical and real-time equipment operating data, to construct an intelligent method that can automatically detect abnormal patterns and fault signs.

[0086] The main purpose of this invention is to improve the reliability and safety of gas pressure regulating equipment, reduce unexpected downtime, and optimize maintenance plans. Its basic functions include:

[0087] Real-time data monitoring: The model can collect key parameters of the pressure regulating equipment in real time, such as inlet pressure, outlet pressure, flow rate, temperature, etc., as well as the operating status of the equipment, such as leakage concentration and the closing status of the shut-off valve.

[0088] Fault diagnosis: By analyzing the collected data, the model can identify whether the equipment has deviated from the normal operating range and diagnose the specific type of fault.

[0089] Fault prediction: A model is built using real historical data on equipment operation. The model can predict possible future equipment failures, allowing preventive measures to be taken in advance.

[0090] Fault alarm: Provides timely alarm information, targeted solutions and maintenance suggestions, and helps operators develop maintenance plans.

[0091] Example 2

[0092] This embodiment further elaborates on step S1, building on the previous example. Data and analysis are fundamental and critical components of fault diagnosis for gas pressure regulating equipment. This ensures accurate capture of equipment status and timely detection of anomalies, providing reliable data support for fault diagnosis and prediction.

[0093] For multi-source heterogeneous data, in addition to using traditional real-time sensor data, the model of the present invention also combines historical text and manual records (maintenance records, accident reports, inspection logs), historical equipment operation data, and environmental and meteorological data.

[0094] ① Basic sensor data: Real-time data monitoring is the first step in fault analysis. It involves continuously tracking various parameters generated by the pressure regulating equipment during operation. These parameters include but are not limited to: inlet pressure, outlet pressure, flow rate, temperature, relief valve, and shut-off valve closing status.

[0095] ② Text records: long-term historical records of pressure, flow, temperature, and text records such as maintenance work orders, accident reports, and inspection logs.

[0096] ③ Environmental and meteorological data: temperature, humidity, rainfall, wind speed, and extreme weather events. Simple meteorological data is collected through the National Meteorological Administration's public API to determine the degree of pipeline corrosion, stress changes, and the impact of external shocks, and to build a dynamic fault tree.

[0097] Data analysis involves processing and interpreting the collected data to identify potential equipment failures. The following are the data analysis methods involved in the present invention:

[0098] ① Fixed-point analysis: By analyzing the collected data at a certain time, determine whether the data is abnormal, such as the outlet pressure accuracy exceeds the standard.

[0099] ② Trend analysis: By analyzing the changing trends of key parameters over time, the degradation trend or abnormal fluctuations of the equipment can be discovered.

[0100] ③ Predictive analysis: Use time series analysis to predict the future status and possible failures of equipment.

[0101] ④ Historical text analysis: Through natural language processing (NLP) technology, key information is extracted from unstructured texts such as maintenance logs and accident reports to supplement the fault tree event library and assist in root cause location.

[0102] ⑤ Meteorological Data Analysis: Corrosion Rate: High humidity and high temperatures accelerate oxidation of metal pipes. Soil Moisture: Rainfall affects soil expansion and contraction, leading to stress changes in pipes. Extreme Weather: Heavy rains cause ground subsidence, and typhoons increase the risk of external mechanical shock.

[0103] The contribution of each feature (pressure, flow, temperature, etc.) in the random forest model to fault diagnosis verifies the effectiveness of multi-parameter joint analysis, such as Figure 2 shown.

[0104] Example 3

[0105] This embodiment further explains step S2 based on the above embodiment.

[0106] 1. Basic fault tree construction

[0107] Fault trees can intuitively display the various possible pathways and causal relationships of a fault, helping analysts gain a comprehensive and systematic understanding of the fault's mechanisms. Through qualitative and quantitative analysis of the fault tree, the key factors of the fault can be quickly identified, allowing targeted measures to be taken to improve system reliability and safety.

[0108] Therefore, before developing fault judgment, a fault tree for a certain fault can be established first, and then the algorithm model can be constructed through the fault tree model. Figure 3 This is the fault tree for determining the series voltage fault in the present invention.

[0109] 2. Dynamic Fault Tree

[0110] The construction of a dynamic fault tree involves historical text analysis and environmental meteorological data analysis.

[0111] Historical text analysis applications are implemented by performing text cleaning and word segmentation and part-of-speech tagging on large amounts of text data. Term frequency-inverse document frequency is then calculated using the TF-IDF algorithm to filter out high-frequency fault keywords (such as "corrosion" and "leakage"). Pre-trained models (such as BERT) are then used to identify equipment components (such as "valve" and "weld") and fault types (such as "crack" and "leak"), thereby constructing a knowledge graph. Finally, the system is dynamically integrated with the fault tree, and the event database is dynamically updated: extracted keywords (such as "weld crack") are added as new event nodes to the fault tree. Weighting can also be adjusted: event probabilities are adjusted based on the frequency of fault occurrences in the text (e.g., if a keyword appears ten times, the corresponding event weight is increased by 15%).

[0112] Environmental meteorological data analysis application implementation: First, obtain the data of the device address (hourly temperature, humidity, and rainfall) through the China Meteorological Administration API, and then use the data to build environmental risk models (such as "corrosion rate model" and "soil stress model").

[0113] Dynamic corrosion model calculation:

[0114] R(t)=K·T(t)·H(t)

[0115] Where R(t) is the environmental risk model; K is the soil corrosion coefficient, obtained through the regional environmental protection report; T(t) is the real-time temperature, °C; H(t) is the real-time humidity,%.

[0116] The dynamic corrosion model prediction effect of the present invention is as follows Figure 4 shown.

[0117] In practical applications, the accuracy of the integration algorithm (n value) will be adjusted according to specific needs, time correlation will be added to the corrosion rate function, and an exception handling mechanism (such as negative value checking) will be added.

[0118] Integration with dynamic faults: Heavy rain triggers soil subsidence. If the rainfall is >50mm for 3 consecutive hours, the probability of soil subsidence increases by 20%. High temperature accelerates aging. If the temperature is >40°C, the weight of the seal aging event increases by 30%.

[0119] Take the top-level event non-pressure fault as an example, Figure 5 As shown; among them, the dynamic branch of equipment failure is as follows Figure 6 As shown; the dynamic branch of pressure gauge failure is as follows Figure 7 As shown; the valve is not sealed tightly and the dynamic branch is as follows Figure 8 shown.

[0120] After dynamically adjusting the fault tree logic gate threshold, the false alarm rate of safety valve failure warning is reduced, such as Figure 9 shown.

[0121] Example 4

[0122] This embodiment further explains step S3 based on the above embodiment.

[0123] The present invention uses different analysis algorithms according to different fault types, with a total of three analysis methods: fixed-point analysis, trend analysis, and predictive analysis.

[0124] 1. Fixed point analysis

[0125] Fixed-point analysis typically involves analyzing data at a specific point in time. For example, if the pressure stabilization accuracy exceeds the specified value, the cutoff accuracy exceeds the specified value, or the release accuracy exceeds the specified value, the current cutoff pressure is calculated and compared with the equipment's set accuracy range. If it is within the range, the result is normal; otherwise, it is abnormal.

[0126] Its implementation is achieved through simple data processing and statistical analysis, such as calculating the mean, median or other statistical indicators at a specific time point. For more complex fixed-point analysis, linear regression can be used: used to analyze the relationship between two variables, implemented through the Apache Commons Math library.

[0127] Given a random sample, (Y i ,X i1 ,X i2 ...X ip ),i=1,...,n, a linear regression model assumes that the regressor Y i and the regressor X i1 ,X i2 ...X ip The relationship between is that in addition to the influence of X, there are other variables. We add an error term ε i (also a random variable) to capture all the i1 ,X i2 ...X ip Any other pair Y i Therefore, a multivariate linear regression model is expressed as follows:

[0128] Y i =β0+β1X i1 +β2X i2 +...+β p X ip +ε i ,i=1,...,n

[0129] Among them, Y i is the regressor of the linear regression model; X i1 ,X i2 ...X ip is the regressor of the linear regression model; β0,β1,β2...β pis the regression coefficient of the linear regression model; ε i is the error term of the linear regression model.

[0130] 2. Trend Analysis

[0131] Trend analysis focuses on the changing trends of data over time and is mainly used for abnormal early warning and risk prediction.

[0132] Implementation 1: Predict pipeline blockage risk. Calculate the 30-day moving average of the inlet pressure. If a continuous upward trend is observed while the flow rate is decreasing during the same period, the pipeline may be blocked, triggering a maintenance inspection work order.

[0133] Implementation 2: Corrosion risk prediction. Key parameters include corrosion rate and soil moisture. Linear regression is used to fit the time-varying corrosion depth (slope β = 0.02 mm / day). Combined with rainfall data, the corrosion model weights are dynamically adjusted to predict when corrosion depth will exceed the threshold within the next 90 days, allowing for preemptive coating repairs.

[0134] The present invention adopts the following trend analysis algorithms:

[0135] ① Moving average: A simple trend analysis method that can smooth short-term fluctuations and highlight long-term trends by calculating the moving average of data points.

[0136]

[0137] Among them, MA t is the moving average at time point t; n is the window size of the moving average, the number of periods; x t-i is the actual observation value at time point ti.

[0138] A simple code example is Figure 10 As shown in the following example. In this code block, the calculateMovingAverage method receives a data array and a window size as parameters, then calculates and returns a new array containing the moving average. The window size windowSize defines the number of consecutive data points to consider when calculating each moving average.

[0139] Exponential smoothing: Another method for trend analysis, particularly well-suited for time series data, can be implemented using the Apache Commons Math library.

[0140] The formula is as follows:

[0141] S t =αX t +(1-α)S t-1

[0142] Among them, S tis the smoothed value at time point t; X t is the actual observation value at time point t; S t-1 is the smoothing value at time point t-1; α is the smoothing parameter, which ranges from 0 to 1 and is determined by the user based on the data set and prediction target.

[0143] A simple code example is Figure 11 As shown in this code example, the simpleExponentialSmoothing method accepts a data array and a smoothing parameter, alpha, and then calculates and returns a new array containing the smoothed values. The first smoothed value is initialized to the first actual observation, and each subsequent smoothed value is calculated based on the previous smoothed value and the current actual observation. The key to this algorithm model is choosing an appropriate alpha value, which is crucial to the effectiveness of exponential smoothing. A larger alpha value gives more weight to recent observations, resulting in a more pronounced smoothing effect. Determining the optimal alpha value typically requires experimentation or optimization.

[0144] 3. Predictive Analytics

[0145] The main implementation of predictive analysis is status prediction and failure probability calculation.

[0146] Implementation 1: Input a time window (24 hours, 72 hours, etc.) through the LSTM model, and use regression prediction to output parameter values ​​(such as pressure and flow) at future time points. For example, if the outlet pressure is predicted to drop by 5% in the next three days, a leak risk warning will be triggered.

[0147] Implementation 2: Failure probability calculation: If the predicted value exceeds the safety threshold (e.g., pressure > 1.2 MPa), the over-limit probability is calculated. Based on the prediction error distribution (e.g., Gaussian distribution), the failure confidence level is calculated.

[0148]

[0149] Where P(fault) is the fault confidence level, μ is the predicted mean, σ is the standard deviation, and Φ is the standard normal distribution function.

[0150] This prediction uses LSTM neural network to predict pipeline pressure and flow trends, identify long-term dependencies, replace traditional statistical methods, and improve prediction accuracy and real-time performance. Figure 12 shown.

[0151] Example 5

[0152] This embodiment further explains step S4 based on the above embodiment.

[0153] Finally, there is the auxiliary fault warning and alarm, which relies on fault analysis. After the model analyzes or predicts a fault, an alarm will be sent to the front end to promptly notify the staff of the fault or warning that has occurred. The staff can notify the maintenance personnel of the fault information by issuing a task order for repair.

[0154] The above is a detailed description of the embodiments of the present invention, but the present invention is not limited to the embodiments. Those skilled in the art may make various equivalent modifications or substitutions without departing from the spirit of the present invention. These equivalents or substitutions are all included in the scope defined by the claims of the present invention.

Claims

1. A gas pipeline health management method based on dynamic fault tree and multi-source data fusion is characterized by: The following steps are involved: S1. Collect multi-source heterogeneous data required for gas pipeline health management; S2. Construct a dynamic fault tree for gas pipelines based on multi-source heterogeneous data; S3. The fault diagnosis model performs gas pipeline fault diagnosis based on multi-source heterogeneous data and a dynamic fault tree, including fixed-point analysis, trend analysis, and predictive analysis of the gas pipeline. The fixed-point analysis includes determining whether the gas pipeline is abnormal by analyzing multi-source heterogeneous data at a specific time point. The trend analysis includes detecting abnormalities in the gas pipeline by analyzing the time-varying trends of key parameters in the multi-source heterogeneous data. The predictive analysis includes predicting the future status and possible faults of the gas pipeline by analyzing the multi-source heterogeneous data. S4. Based on the gas pipeline fault diagnosis results of the fault diagnosis model, gas pipeline fault alarms and early warnings are performed, and fault-specific solutions and maintenance suggestions are generated.

2. The gas pipeline health management method based on dynamic fault tree and multi-source data fusion according to claim 1 is characterized in that: In step S1, the multi-source heterogeneous data includes basic sensor data, text record data and environmental meteorological data; in: The basic sensor data are various parameters generated by the gas pipeline pressure regulating equipment during operation, including: inlet pressure, outlet pressure, flow, temperature, and the switch status of the relief valve and the shut-off valve; The text record data includes: long-term historical records of pressure, flow, temperature, and text records of maintenance work orders, accident reports, and inspection logs; The environmental meteorological data includes: temperature, humidity, rainfall, wind speed, and extreme weather events. The environmental meteorological data is obtained from the National Meteorological Administration and is used to determine the impact of corrosion, stress changes, and external shocks on gas pipelines and to construct a dynamic fault tree, including: Corrosion rate: High humidity + high temperature accelerates the oxidation of metal gas pipelines; Soil moisture: Rainfall affects soil expansion and contraction, leading to stress changes in gas pipelines; Extreme weather: Heavy rains cause ground subsidence, and typhoons increase the risk of external mechanical shocks.

3. The gas pipeline health management method based on dynamic fault tree and multi-source data fusion according to claim 1 is characterized in that: In step S2, when constructing the dynamic fault tree of the gas pipeline, a basic fault tree is first established, and then the dynamic fault tree of the gas pipeline is established by combining the basic fault tree, text record data in multi-source heterogeneous data, and environmental meteorological data; The data recorded in combination with text includes: Perform text cleaning and word segmentation and part-of-speech tagging on text record data, calculate term frequency-inverse document frequency using the TF-IDF algorithm, and filter high-frequency fault keywords in the text record data; Use pre-trained models to identify faulty equipment components and types corresponding to high-frequency fault keywords; Dynamically integrate the identified faulty equipment components and types with the basic fault tree, and dynamically update the fault event library to obtain a dynamic fault tree, including: Add the identified faulty equipment components and types as new event nodes to the basic fault tree; Adjust the fault event weights based on the frequency of occurrence of faulty equipment components and types in text record data; Among them, combined with environmental meteorological data include: The environmental meteorological data of the address where the gas pipeline is located is obtained through the China Meteorological Administration, and the environmental meteorological data is used to build an environmental risk model, which is then integrated with the dynamic fault tree.

4. The gas pipeline health management method based on dynamic fault tree and multi-source data fusion according to claim 3 is characterized in that: In step S2, the environmental risk model includes a dynamic corrosion model, which includes: R(t)=K·T(t)·H(t) Where R(t) is the environmental risk model; K is the soil corrosion coefficient, obtained from the regional environmental protection report; T(t) is the real-time temperature, °C; H(t) is the real-time humidity, %. In the application of the dynamic corrosion model, the accuracy of the integral algorithm is adjusted according to specific needs, time correlation is added to the corrosion rate function, and an exception handling mechanism is added; In step S2, the integration of the environmental risk model with the dynamic fault tree includes: heavy rain triggers soil settlement. If the rainfall is greater than 50 mm for three consecutive hours, the probability of soil settlement events increases by 20%; high temperature accelerates aging. If the temperature is greater than 40°C, the weight of the sealing ring aging event increases by 30%.

5. The gas pipeline health management method based on dynamic fault tree and multi-source data fusion according to claim 1 is characterized in that: In step S3, the fixed-point analysis is achieved through data processing and statistical analysis, including: Calculate the mean, median, or other statistical indicators at a specific time point; Use linear regression to analyze the relationship between two variables, implemented using the Apache Commons Math library; In the linear regression, a multivariate linear regression model includes: Y i =β0+β1X i1 +β2X i2 +...+b p X ip +e i ,i=1,...,n Among them, Y i is the regressor of the linear regression model; X i1 ,X i2 ...X ip is the regressor of the linear regression model; β0,β1,β2...β p is the regression coefficient of the linear regression model; ε i is the error term of the linear regression model.

6. The gas pipeline health management method based on dynamic fault tree and multi-source data fusion according to claim 1 is characterized in that: In step S3, the trend analysis includes: Pipeline blockage risk prediction: Calculate the N-day moving average of the inlet pressure. If it shows a continuous upward trend and the flow rate shows a downward trend during the same period, it is determined that the pipeline is blocked and a maintenance inspection work order is triggered; Corrosion risk prediction: Key parameters include corrosion rate and soil moisture. Linear regression is used to fit the change in corrosion depth over time. Combined with rainfall data, the corrosion model weight is dynamically adjusted to predict the corrosion depth for the next N days. The corrosion depth is then compared with the corrosion depth threshold. If the threshold is greater than the threshold, corrosion risk exists, triggering the early arrangement of coating repairs.

7. The gas pipeline health management method based on dynamic fault tree and multi-source data fusion according to claim 1 is characterized in that: In step S3, the trend analysis algorithm includes a moving average algorithm, which smooths short-term fluctuations and highlights long-term trends by calculating the moving average of data points. The moving average algorithm includes: Among them, MA t is the moving average at time point t; n is the window size of the moving average, the number of periods; x t-i is the actual observation value at time point ti.

8. The gas pipeline health management method based on dynamic fault tree and multi-source data fusion according to claim 1 is characterized in that: In step S3, the algorithm used for trend analysis includes an exponential smoothing algorithm. The exponential smoothing algorithm processes time series data and is implemented using the Apache Commons Math library. The exponential smoothing algorithm includes: S t =αX t +(1-α)S t-1 Among them, S t is the smoothed value at time point t; X t is the actual observation value at time point t; S t-1 is the smoothing value at time point t-1; α is the smoothing parameter, which ranges from 0 to 1 and is determined by the user based on the data set and prediction target.

9. The gas pipeline health management method based on dynamic fault tree and multi-source data fusion according to claim 1 is characterized in that: In step S3, the prediction analysis includes: The time window is input through the LSTM model, and the regression prediction outputs the parameter prediction value at the future time point; If the predicted value exceeds the safety threshold, the over-limit probability is calculated, and based on the prediction error distribution, the fault confidence level is calculated. The fault confidence level includes: Where P(fault) is the fault confidence level, μ is the predicted mean, σ is the standard deviation, and Φ is the standard normal distribution function.

10. The gas pipeline health management method based on dynamic fault tree and multi-source data fusion according to claim 1, characterized in that: In step S4, for the analyzed fault, a fault alarm is sent to the front end; for the predicted fault, a fault warning is sent to the front end; Notify staff of the fault or warning that has occurred. Staff will notify maintenance personnel of the fault information by issuing a task order based on the fault-specific solution and maintenance suggestions for repair.

Images