A dynamic leakage prediction method and system for old oil and gas pipelines

Abstract

This application discloses a dynamic leakage prediction method and system for old oil and gas pipelines. The method includes: real-time acquisition of pressure, flow, and temperature parameters through multi-source sensors, combined with time series analysis and frequency domain feature extraction to identify abnormal fluctuations; calculation of pipeline status assessment results based on material degradation models; establishment of a leakage prediction model that integrates the wall thickness degradation dynamics equation and the LSTM neural network, using the Monte Carlo method to calculate the leakage probability and generate the diffusion velocity and concentration gradient through CFD numerical simulation; when the diffusion prediction exceeds the safety threshold, the control strategy is optimized through fuzzy logic and genetic algorithms; real-time adjustment and feedback verification model is used, and online learning is carried out through Bayesian optimization; finally, the model is calibrated using experimental data and deployed to generate risk warnings. The system includes modules such as a multi-source sensor array and a data processing platform, realizing closed-loop control from perception to warning.

Description

Technical Field

[0001] The present application relates to the technical field of pipeline leakage prediction, and in particular to a method and system for dynamic leakage prediction of old oil and gas pipelines. Background Art

[0002] As critical infrastructure for energy transportation, the safe operation of urban oil and gas pipelines is directly related to the security of urban energy supply and public safety. With the rapid development of urbanization in my country, a large number of oil and gas pipelines have entered the aging stage, significantly increasing the risk of pipeline leakage.

[0003] Current pipeline leakage prediction methods primarily rely on static risk assessments and periodic inspections, which struggle to accurately reflect the dynamic characteristics of pipelines during actual operation. Existing technologies often overlook the impact of process parameter fluctuations on the leakage process and lack a deep understanding of the evolution of leakage over the entire pipeline lifecycle. This results in low prediction accuracy and an inability to meet the practical needs of aging pipeline safety management.

[0004] Chinese patent publication CN119129436B discloses a "Method and System for Predicting Gas Pipeline Leakage Based on Improved Lattice Boltzmann Method." This method uses a GPU-accelerated lattice Boltzmann method to generate flow field simulation data, combining dimensionality reduction with a beta variational autoencoder and an iTransformer model to predict methane concentration fields. While this solution achieves dynamic flow field prediction for oil and gas pipelines, it still suffers from the following drawbacks: it fails to consider the impact of degradation characteristics such as pipeline material strength loss and corrosion depth on flow field evolution, resulting in a disconnect between the prediction model and the actual operating conditions of aging pipelines. Summary of the Invention

[0005] In order to address the deficiencies of the existing technology, the purpose of this application is to provide a dynamic leakage prediction method and system for old oil and gas pipelines, which integrates pipeline aging models, dynamic process parameters and control intervention verification to achieve accurate prediction of leakage behavior under dynamic pipeline conditions.

[0006] To achieve the above objectives, this application adopts the following technical solutions:

[0007] The present application provides a method for predicting dynamic leakage of old oil and gas pipelines, which includes:

[0008] S101: Real-time collection of pipeline operating pressure, flow, and temperature distribution parameters. Pressure fluctuation characteristics are extracted using a time series analysis algorithm. When the characteristic value exceeds a preset threshold, the pipeline condition assessment result is calculated by combining the pipeline material degradation model and operating time.

[0009] S102, establishing a leakage prediction model based on the pipeline status assessment results. When the pipeline aging degree exceeds a critical value, the oil and gas leakage probability is calculated by integrating historical data, and the diffusion velocity and concentration gradient are obtained through numerical simulation;

[0010] S103: Calculate the oil and gas diffusion range under different pressure and flow combinations using a leakage prediction model, and classify and predict the diffusion rate using a support vector machine algorithm. If the predicted value exceeds a safety threshold, dynamically optimize the control strategy based on the temperature distribution and material degradation parameters, and determine the pipeline control parameters.

[0011] S104: Adjust pipeline control parameters in real time and verify the stability of operating data and the downward trend of corrosion indicators through feedback algorithms. If the conditions are met, update the leakage prediction model weights and perform online optimization.

[0012] S105, using experimental data to calibrate the accuracy of the leakage prediction model, deploying it to the monitoring system after verification, and generating risk warning information based on real-time operation data.

[0013] As a preferred technical solution, in S101, the time series analysis algorithm uses the ARIMA model to analyze the pressure data to predict the pressure fluctuation characteristics in the future period and calculate the fluctuation standard deviation. When the fluctuation standard deviation exceeds the preset threshold, it is judged as a fluctuation anomaly; at the same time, the flow and temperature distribution parameters are Fourier transformed to extract frequency characteristics and identify periodic abnormal patterns.

[0014] As a preferred technical solution, in S101, the pipeline status assessment result is calculated by combining the pipeline material degradation model and the operating time, including: calculating the tensile strength value at the current moment based on the initial tensile strength value of the pipeline material, the preset annual tensile strength decrease rate value, and the pipeline operating time, wherein the current tensile strength value shows an exponential decay trend with the operating time; based on the preset annual corrosion depth reference value, the total cumulative corrosion amount of the pipeline is accumulated according to the operating years; and comprehensively combining the current tensile strength value and the cumulative total corrosion amount to generate a status assessment result including an aging grade grading index and a corrosion risk grade grading index.

[0015] As a preferred technical solution, in S102, after the leakage prediction model is established, a long short-term memory neural network is also used to train the leakage prediction model, wherein the input features include the pipeline wall thickness measurement value, pipeline pressure value, flow value and temperature value; based on several sets of historical operation data, the training set and the test set are constructed into a data set in proportion; and the optimizer is used to iteratively update the weights at a set learning rate.

[0016] As a preferred technical solution, in S103, the oil and gas diffusion range under different pressure and flow combination conditions is calculated by using the leakage prediction model, which includes: establishing a pipeline oil and gas flow model based on the leakage diffusion model, and the pipeline oil and gas flow model includes the partial differential equation of viscous fluid motion and the mass conservation equation; the pipeline oil and gas flow model adopts Turbulence model, setting turbulence effect compensation mechanism and no-slip wall boundary conditions; calculating the oil and gas diffusion range under different pressure and flow combination inputs.

[0017] As a preferred technical solution, in S103, the diffusion velocity is classified and predicted using the support vector machine algorithm, including: setting three safety levels of low speed, medium speed, and high speed; input features include pressure gradient, flow dynamic value and ambient humidity; output diffusion velocity level classification results; in S103, the control strategy is dynamically optimized based on the temperature distribution and material degradation parameters, and the control parameters are determined, including: adjusting the control strategy strength through a fuzzy logic algorithm, the input of the fuzzy logic algorithm is the temperature distribution state and material performance degradation parameters, and the output is the control strategy strength; and optimizing the control parameters through a genetic algorithm, the objective function is to minimize the oil and gas diffusion range and energy consumption, setting the population size and number of iterations, and obtaining the optimal control parameter combination.

[0018] As a preferred technical solution, in S104, the conditions for updating the weight parameters of the leakage prediction model are as follows: whether the standard deviation of the parameter fluctuation after the intervention is smaller than the standard deviation of the fluctuation before the intervention, and whether the corrosion degree assessment index decreases.

[0019] As a preferred technical solution, in S104, the Bayesian optimization algorithm is used to perform online optimization of the leakage prediction model, with the weighted sum of the prediction error and the calculation delay as the optimization target, the weights of the prediction error and the calculation delay are set, and the time of each optimization calculation is limited.

[0020] As a preferred technical solution, in S105, the accuracy of the leakage prediction model is calibrated using experimental data, including: inputting the experimental data set into the revised prediction model, generating simulation results through the random forest algorithm to obtain a risk prediction value, and calculating the prediction deviation between the simulation results and the experimental data based on the mean square error formula; if the prediction deviation is lower than the preset threshold, the logistic regression algorithm is used to quantify the model accuracy evaluation value; in S105, risk warning information is generated based on real-time operation data, including: real-time collection of pressure change, flow dynamics and temperature distribution data from the deployment system; performing denoising and normalization operations through the data preprocessing module to generate standardized monitoring data; using the support vector machine algorithm to analyze the standardized data and output the pipeline leakage risk level; generating a risk warning signal according to the risk level value.

[0021] The present application also provides a dynamic leakage prediction system for old oil and gas pipelines, which is used to implement the above-mentioned dynamic leakage prediction method for old oil and gas pipelines. The system includes: a multi-source sensor array, which includes a distributed optical fiber pressure sensor, an ultrasonic flowmeter and an infrared thermal imaging temperature monitoring unit, and is used to collect pressure change characteristics, flow dynamic data and temperature distribution status of pipeline operation in real time; a data processing platform, which is used to execute time series analysis algorithms, pipeline aging and corrosion degree assessments, establish leakage evolution dynamics equations, train leakage development trend prediction models, perform numerical simulations, execute support vector machine algorithm classification predictions, activate control intervention response mechanisms, perform feedback control and model online learning optimization; a control execution unit, which is used to adjust pipeline operation parameters in real time according to the control parameter combination generated by the control intervention response mechanism; an early warning module, which is used to generate early warning information according to the leakage risk level; a model verification unit, which is used to verify and calibrate the revised prediction model using similar experimental data; and a model deployment unit, which is used to deploy the verified model into the actual pipeline monitoring system.

[0022] Compared with the prior art, the present invention has the following advantages:

[0023] This application proposes a four-dimensional dynamic leakage prediction system of "perception-prediction-control-verification". Through five major technological breakthroughs, including multi-source sensing network (pressure / flow / temperature), physical-AI fusion modeling (wall thickness degradation equation + LSTM), closed-loop control optimization (genetic algorithm + SVM classification) and online verification and calibration, it realizes the full-process prevention and control of leakage risks in aging oil and gas pipelines.

[0024] Therefore, compared to traditional pipeline leak prediction, this application improves prediction accuracy, achieves breakthroughs in micro-leak identification, and significantly increases the detection rate of very early hidden dangers. This application also shortens early warning response time, triggering intervention at the earliest stages of leak risk. This application also achieves precise suppression of diffuse hazards, significantly reducing the scope of accident impact through dynamic control strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This application provides a flow chart of a method for dynamic leakage prediction of old oil and gas pipelines. DETAILED DESCRIPTION

[0026] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the specific implementation of the present application will be clearly and completely described below in conjunction with the drawings in the implementation of the present application.

[0027] like Figure 1 As shown, the present application provides a method for dynamic leakage prediction of old oil and gas pipelines, which includes:

[0028] S101 collects the pressure, flow and temperature distribution parameters of pipeline operation in real time, extracts pressure fluctuation characteristics through time series analysis algorithm, and calculates pipeline status assessment results by combining pipeline material degradation model and operation time when the characteristic value exceeds the preset threshold.

[0029] In this application, multi-sensor data fusion technology is used to collect the pressure, flow and temperature distribution parameters of pipeline operation in real time. Specifically, the pressure sensor, flow meter and temperature sensor are used to collect data respectively:

[0030] The pressure sensor records once per second, with a pressure range of 0.5 to 2.0 MPa; the flow meter records once per minute, with a flow range of 10 to 50 cubic meters per hour; and the temperature sensor records once every 10 seconds, with a temperature range of -10 to 60 degrees Celsius. A weighted average algorithm is used for fusion, with weights of 0.4, 0.3, and 0.3, respectively, to ensure that the data comprehensively reflects the pipeline status. Weights are assigned based on the contribution of pipeline leakage risk factors, with pressure sensitivity accounting for the highest proportion, followed by the synergistic impact of flow and temperature.

[0031] Furthermore, in S101, the time series analysis algorithm uses the ARIMA model (autoregressive integrated moving average) to analyze pressure data and predict pressure fluctuation characteristics over a period of time. The ARIMA model uses autoregression to reflect the influence of historical values, differencing to eliminate nonstationarity, and moving average to address random noise. This allows for precise quantification of the dynamic evolution of pipeline pressure, laying the technical foundation for leak prediction. A model selection criterion dynamically determines the order of autoregression, differencing, and moving average to adapt to the pressure evolution patterns of different operating conditions.

[0032] And calculate the standard deviation of pressure fluctuations:

[0033] ;

[0034] in is the pressure sampling value, is the mean, is the window data size, is the number of sampling times.

[0035] This application solves the static monitoring blind spots of oil and gas pipelines by extracting process parameter fluctuation characteristics through multi-sensor fusion and time series analysis.

[0036] In this application, the ARIMA model is used to analyze the pressure data, predict the pressure fluctuation characteristics in the next 10 minutes, and calculate the standard deviation of the pressure parameter fluctuation. If the fluctuation standard deviation exceeds the preset threshold of 0.1MPa, the fluctuation is considered abnormal.

[0037] When the standard deviation of the fluctuation exceeds the preset threshold, it is determined to be a fluctuation anomaly. At the same time, the flow and temperature distribution parameters are Fourier transformed to extract frequency characteristics and identify periodic abnormal patterns:

[0038] ;

[0039] .

[0040] in, For time, Time-varying flow parameters, is the frequency characteristic of the flow parameter, is the position distribution vector, Temperature distribution parameters that vary with time, is the frequency characteristic of the temperature distribution parameters.

[0041] The cause of high-frequency flow oscillation may be valve seal failure, etc., and the cause of low-frequency temperature field pulsation may be local accumulation of corrosion products, etc.

[0042] Furthermore, when it is detected that the parameter pressure fluctuation amplitude exceeds a threshold, for example, the pressure fluctuation standard deviation reaches 0.15 MPa, the pipeline aging and corrosion degree assessment module is triggered. The pipeline aging and corrosion degree assessment module is based on a material performance degradation model.

[0043] Furthermore, in S101, the pipeline condition assessment results calculated by combining the pipeline material degradation model with the operating time include:

[0044] The current tensile strength is calculated based on the initial tensile strength of the pipeline material, the preset annual tensile strength decline rate, and the pipeline's operating time. The current tensile strength decays exponentially with operating time. Specifically, current tensile strength = initial tensile strength × (1 - annual decline rate) ^ operating years. In this application, the pipeline material is carbon steel, with an initial tensile strength of 500 MPa. It degrades at a rate of 2% per year. Combined with 10 years of operation, the calculated current tensile strength is 400 MPa.

[0045] The pipeline aging and corrosion assessment module incorporates a corrosion rate model. Based on a preset annual corrosion depth baseline, the model accumulates the total amount of corrosion in the pipeline by year of operation. Furthermore, the current tensile strength value and the total amount of corrosion are combined to generate a condition assessment result that includes aging grade and corrosion risk grade indicators.

[0046] In this application, five corrosion risk levels are set based on the cumulative corrosion depth: cumulative corrosion depth < 0.5mm, corrosion risk level 1; cumulative corrosion depth 0.5-1.0mm, corrosion risk level 2; cumulative corrosion depth 1.0-3.0mm, corrosion risk level 3; cumulative corrosion depth 3.0-4.0mm, corrosion risk level 4; cumulative corrosion depth ≥ 4.0mm, corrosion risk level 5. In this application, if the annual corrosion depth of the pipeline is 0.2mm and the cumulative corrosion depth is 2mm, the pipeline condition assessment result is moderate aging and the corrosion risk level is 3.

[0047] The pipeline condition assessment results are automatically fed back to the maintenance scheduling system. If the corrosion risk level exceeds Level 2, optimized operating parameter recommendations are generated, such as reducing the pressure to 1.2 MPa to slow aging. This entire process is handled entirely by the system, with data analysis and assessment results updated in real time to ensure safe pipeline operation.

[0048] S102, establishing a leakage prediction model based on the assessment results; when the pipeline aging degree exceeds a critical value, the oil and gas leakage probability is calculated by integrating historical data, and the diffusion velocity is generated through numerical simulation.

[0049] Specifically, the leakage prediction model includes a dynamic model of the pipe wall thickness reduction over time:

[0050] ;

[0051] in, is the wall thickness, is the corrosion rate constant, is the pressure in the pipeline, is the activation energy, is the gas constant, is the temperature. In this application, Set to 0.01mm / year, Assuming 10MPa, is 50 kJ / mol, is 8.314 J / mol·K, The corrosion rate constant is determined based on the standard carbon steel corrosion test data, and the activation energy adopts the typical activation energy parameters of carbon steel in the Materials Science Handbook.

[0052] A numerical solution is used to calculate the evolution of pipe wall thickness over a period of time. In this application, the Runge-Kutta method (fourth-order Runge-Kutta method) is used with a step size of 0.01 years to calculate the evolution of the wall thickness from 5 mm to 4.5 mm over 10 years. The Runge-Kutta method approximates the true solution through a weighted average of multiple slopes.

[0053] Furthermore, in S102 , after the leakage prediction model is established, a long short-term memory neural network (LSTM) is used to train the leakage prediction model.

[0054] The input features include the pipe wall thickness measurement value, pipe pressure value, flow value, and temperature value. In this application, the wall thickness measurement value is 4.5mm, the pipe pressure value is 10MPa, the flow value is 1000m³ / h, and the temperature value is 300K.

[0055] Based on several sets of historical operation data, the training set and test set are proportionally constructed into a data set. In this application, the data set is 1000 sets of historical operation data, of which the training data accounts for 80% and the test data accounts for 20%.

[0056] The optimizer is used to iteratively update the weights at a set learning rate. In this application, the optimizer is Adam, the learning rate is 0.001, and 1000 iterations are performed to predict the probability of leakage in the next year.

[0057] Furthermore, the LSTM network uses gradient clipping technology to prevent training divergence and sets an early stopping mechanism to prevent overfitting.

[0058] The leakage prediction model output shows that the leakage probability increases exponentially as the wall thickness decreases.

[0059] This application combines leakage evolution dynamics equations with neural network training to improve the accuracy of pipeline dynamic leakage trend prediction.

[0060] Furthermore, in S102, if the pipeline condition assessment result exceeds the critical level, the Monte Carlo method is used to calculate the leakage probability, combined with the historical data of process parameter fluctuations. In this application, if the pipeline depth exceeds 30%, the Monte Carlo method is used to perform 10,000 samplings based on the historical data of process parameter fluctuations: internal pressure fluctuations of ±1 MPa and flow fluctuations of ±100 m³ / h, and the leakage probability is calculated to be 0.15.

[0061] Furthermore, a quasi-Monte Carlo method based on the Sobol sequence is used to improve the spatial uniformity of low-probability event sampling.

[0062] Furthermore, numerical simulations used CFD (Computational Fluid Dynamics) to simulate leakage diffusion, setting the pipeline operating conditions to: internal pressure of 10 MPa, pore diameter of 2 mm, and natural gas with a density of 0.8 kg / m³. CFD methods discretize the continuous fluid domain into finite elements (a mesh) and simulate flow behavior by solving the governing equations (Navier-Stokes equations).

[0063] The Navier-Stokes equations include:

[0064] Continuity equation: ;

[0065] Momentum equation: ;

[0066] Energy equation: ;

[0067] in, is the fluid density, is the flow velocity vector, For pressure, is the viscous stress tensor, is the acceleration due to gravity, is the thermal conductivity coefficient, For temperature.

[0068] In this application, Fluent software was used to create a mesh of 1 million cells and calculate the diffusion velocity. The initial pipeline pressure was 10 MPa, the flow rate was 0.5 m³ / s, and the ambient temperature was 298 K. Using the Navier-Stokes equations combined with the mass conservation equation, with a meshing spatial step of 0.1 m and a time step of 0.01 s, the calculation showed that the oil and gas diffusion range reached a radius of 5 m within 10 seconds. The diffusion velocity and concentration gradient were calculated to be 2.5 m / s and 0.1 kg / m³·m.

[0069] This process forms a complete logical chain from dynamic modeling to probability calculation to diffusion simulation. Parameters are linked through wall thickness, pressure, etc. to ensure consistency between prediction and simulation results.

[0070] When pipeline condition assessment results exceed critical levels, this application significantly increases risk assessment confidence through a collaborative verification mechanism using Monte Carlo and CFD. Furthermore, a CFD method is used to further simulate the triggering conditions for leak spread, reducing the system's data processing workload.

[0071] S103: Calculate the oil and gas diffusion range under different pressure and flow combinations using a leakage prediction model, and classify and predict the diffusion rate using a support vector machine (SVM) algorithm. If the predicted value exceeds a safety threshold, dynamically optimize the control strategy based on the temperature distribution and material degradation parameters, and determine the control parameters.

[0072] Furthermore, in S103, using the support vector machine algorithm to classify and predict the diffusion speed includes: setting three safety levels of low speed, medium speed, and high speed; inputting features including pressure gradient, flow dynamic value, and ambient humidity; and outputting the diffusion speed level classification result.

[0073] The essence of the support vector machine algorithm is hyperplane classification based on maximum margin, which is used in this application to classify and predict diffusion velocity. The input includes the characteristic vector of pressure fluctuation standard deviation, flow frequency domain energy concentration, and temperature spatial gradient: .

[0074] In this application, MPa / m, m / s, .

[0075] Optimal classification hyperplane: ,in is the weight vector, is the bias term.

[0076] The objective function is: .

[0077] The constraints are: , . is the penalty coefficient, balancing classification accuracy and fault tolerance; This is a slack variable that allows for a small amount of sample misclassification. The output of the support vector machine algorithm for classification prediction of diffusion velocity data is the diffusion velocity level. In this application, the diffusion velocity levels are set as: low speed <0.5 m / s, medium speed 0.5-1 m / s, and high speed >1 m / s.

[0078] Furthermore, when the diffusion rate is nonlinearly related to the characteristics, the radial basis kernel function is used: .in is the kernel parameter, which controls the influence range of the sample; and There are two independent data vectors (or feature vectors) in the input space, each of which represents the coordinates of a sample point in the feature space.

[0079] In this application, the penalty parameter Set to 1.0, the kernel parameter The value is 0.1, and the training data set is 1000 groups, thereby improving the accuracy of the test set.

[0080] This application uses vector machine algorithm and radial basis kernel function to predict the diffusion velocity of pipeline oil and gas, further improving the accuracy of pipeline dynamic leakage trend prediction, and uses the diffusion velocity level as the trigger condition of the control strategy.

[0081] In this application, if the predicted diffusion velocity exceeds a safety threshold of 1 m / s, the control intervention response mechanism is automatically activated. A fuzzy logic algorithm adjusts the intervention strategy intensity based on real-time sensor data. The temperature distribution range is set to 300-350 K, and the material degradation parameter is a strength reduction rate of 0.01-0.05. For example, if the temperature rises to 330 K, the material strength reduction rate reaches 0.03, and the intervention intensity is increased by 20%. The leakage flow is reduced by adjusting the valve opening (from 50% to 30%).

[0082] Furthermore, in S103, calculating the oil and gas diffusion range under different pressure and flow combination conditions using the leakage prediction model includes: establishing a viscous fluid motion partial differential equation and a mass conservation equation for the oil and gas flow in the pipeline.

[0083] Pipeline oil and gas flow Turbulence model, setting turbulence effect compensation mechanism and no-slip wall boundary conditions. The turbulence model takes into account the effect of corrosion roughness on turbulence intensity, customizes the wall function to deal with the irregular surface of the aging pipeline, and dynamically links the turbulence parameters with the leakage probability model.

[0084] In this application, the above-mentioned Navier-Stokes equations are used to calculate the partial differential equations of viscous fluid motion, and the mass conservation equation uses the following continuity equation:

[0085] Continuity equation: ;

[0086] Momentum equation: ;

[0087] Energy equation: ;

[0088] in, is the fluid density, is the flow velocity vector, For pressure, is the viscous stress tensor, is the acceleration due to gravity, is the thermal conductivity coefficient, For temperature.

[0089] Furthermore, in S103, the control strategy is dynamically optimized based on the temperature distribution and material degradation parameters. The control parameters are determined by adjusting the control strategy strength using a fuzzy logic algorithm. The fuzzy logic algorithm takes the temperature distribution and material degradation parameters as input and outputs the control strategy strength. The control parameters are then optimized using a genetic algorithm. The objective function is to minimize the oil and gas diffusion range and energy consumption. The population size and number of iterations are set to obtain the optimal control parameter combination.

[0090] Fuzzy logic is a mathematical model for processing continuous hierarchical logic, breaking through the limitations of traditional Boolean logic (0 / 1 binary judgments). It uses membership functions to describe partial truth values ​​of variables in the interval [0, 1], thereby simulating the decision-making process of human experts in complex systems. Specifically, the temperature distribution state parameters are mapped into the fuzzy subsets {low temperature, normal temperature, high temperature} and {flat, medium, steep}; the temperature distribution state parameters are mapped into the fuzzy subsets {low temperature, normal temperature, high temperature} and {flat, medium, steep}; the center of gravity method is used to calculate the precise control intensity value to drive valve opening adjustment. Through this scheme, this application overcomes the stability defects of traditional control algorithms in nonlinear regions and achieves dynamic matching of corrosion status and control intensity.

[0091] In this application, a genetic algorithm was used to optimize process control parameters. The objective function was to minimize the oil and gas diffusion range and energy consumption. A population size of 50 was set and 100 iterations were performed to obtain the optimal parameter combination: pressure of 8 MPa, flow rate of 0.3 m³ / s, and valve opening of 35%. This kept the oil and gas diffusion range within 3 meters, meeting safety requirements. The genetic algorithm simulates the mechanism of biological evolution, iterating through a "selection-crossover-mutation" process to search for the optimal solution, achieving the optimal balance between minimizing the oil and gas diffusion range and minimizing energy consumption.

[0092] The entire process forms a closed-loop control through real-time data collection, model calculation and algorithm optimization to ensure rigorous logic and high efficiency.

[0093] S104, adjust the pipeline control parameters in real time, and verify the stability of the operating data and the downward trend of the corrosion index through the feedback algorithm; if the conditions are met, update the leakage prediction model weight and optimize it online.

[0094] For real-time adjustment of pipeline operating parameters, the pressure, flow and temperature data of the pipeline are first collected in real time through sensors. In this application, the pressure value is 3.5MPa, the flow rate is 120m³ / h, and the temperature is 45°C.

[0095] According to the preset control intervention strategy, the system calculates the opening deviation of the regulating valve using a feedback algorithm based on the PID control algorithm. In this application, it is assumed that the current opening is 60% and the target opening is adjusted to 65%, so that the pressure is stabilized within the range of 3.0MPa.

[0096] Next, the intervention effect is monitored through a feedback control algorithm. The system collects process parameters every 5 minutes and calculates the pressure fluctuation rate. It is assumed that the fluctuation rate was ±0.3MPa before the intervention and dropped to ±0.1MPa after the intervention. At the same time, the corrosion sensor data is used to evaluate the corrosion degree index. It is assumed that the index was 0.8 before the intervention and dropped to 0.5 after the intervention, indicating that the intervention was effective.

[0097] Furthermore, in S104, by comparing whether the parameter fluctuation standard deviation after the intervention is smaller than the fluctuation standard deviation before the intervention and whether the corrosion degree assessment index decreases, it is used as a condition for updating the weight parameter of the leakage prediction model.

[0098] If the fluctuations tend to stabilize and the corrosion indicators decrease, the system automatically triggers the weight update mechanism of the leakage prediction model, adjusts the model parameters through the gradient descent algorithm, optimizes the pressure-related weight from 0.4 to 0.45, and adjusts the flow weight from 0.3 to 0.35, to ensure the model's adaptability to new data.

[0099] Furthermore, in S104, the Bayesian optimization algorithm is used to perform online optimization of the leakage prediction model, with the weighted sum of the prediction error and the calculation delay as the optimization target, the weights of the prediction error and the calculation delay are set, and the time of each optimization calculation is limited.

[0100] In this application, combined with the operating time cycle data (such as the pipeline has been running for 5000 hours), the system adopts an online learning method and uses the Bayesian optimization algorithm to optimize the model parameters and calculate the new leakage probability distribution. After optimization, the leakage probability is reduced from 0.25 to 0.18.

[0101] Finally, the system generates a revised leakage risk assessment matrix, dividing the risk level into 5 levels. The updated matrix shows that the proportion of high-risk areas has dropped from 15% to 10%, and the results are stored in the database for subsequent analysis and decision support.

[0102] Through the above process, the system realizes fully automated closed-loop management from data collection to model optimization, ensuring pipeline operation safety and prediction accuracy.

[0103] S105, using experimental data to calibrate the accuracy of the leakage prediction model, deploying it to the monitoring system after verification, and generating risk warning information based on real-time operation data.

[0104] Furthermore, in S105, the accuracy of the leakage prediction model is calibrated using experimental data, including: inputting the experimental data set into the revised prediction model, generating simulation results through the random forest algorithm, obtaining a risk prediction value, and calculating the prediction deviation between the simulation results and the experimental data based on the mean square error formula; if the prediction deviation is lower than a preset threshold, the logistic regression algorithm is used to quantify the model accuracy evaluation value.

[0105] Furthermore, in S105, risk warning information is generated based on real-time operating data, including: real-time collection of pressure change, flow dynamics and temperature distribution data from the deployment system; performing denoising and normalization operations through a data preprocessing module to generate standardized monitoring data; using a support vector machine algorithm to analyze the standardized data and output the pipeline leakage risk level; and generating a risk warning signal based on the risk level value.

[0106] In this application, to validate and calibrate the revised pipeline leakage prediction model, simulated experimental data including pressure, flow, and temperature were first generated. 100 data sets were collected, each including pipeline pressure (range 0.5-2.0 MPa), flow (range 10-50 m³ / h), and temperature (range -10-60°C). Random perturbations were used to simulate actual operating fluctuations. The model was trained using the support vector machine (SVM) algorithm, using the radial basis function (RBF) kernel with parameters C set to 1.0 and γ set to 0.1. The training and test sets were split 8:2 to verify the model's prediction accuracy. During the calibration phase, the SVM parameters were optimized using a grid search, adjusting C within the range [0.1, 10] and γ within the range [0.01, 1] to minimize the mean squared error (MSE). The simulation results were compared with the experimental data, and the deviation, such as the absolute percentage error (MAPE) of the difference between the predicted and actual pressures, was calculated. If the MAPE was < 5%, the accuracy was considered to meet engineering requirements.

[0107] For example, a data set with a measured pressure of 1.2 MPa and a predicted value of 1.15 MPa achieved a MAPE of 4.17%, meeting the requirements. After verification, the model was deployed to a pipeline monitoring system, collecting real-time sensor data with a pressure sampling frequency of 1 Hz, a flow rate accuracy of 0.1 m³ / h, and a temperature accuracy of 0.5°C. Features were extracted using a sliding window (60-second window size), including the pressure change rate (ΔP / Δt), flow rate fluctuation variance, and temperature gradient. Based on these features, the model predicted the probability of leakage and used logistic regression to map the probability to a risk level (low risk: 0-30%, medium risk: 30-70%, and high risk: 70-100%).

[0108] For example, if at a certain moment the pressure suddenly drops by 0.1 MPa / s, the flow rate variance is 0.2 m³ / h, and the temperature gradient is 2°C / m, the predicted leak probability is 75%. A high-risk warning message is generated and pushed to the monitoring platform via the MQTT protocol, triggering an automatic alarm. Prediction results are continuously updated, with features recalculated and risk levels refreshed every 5 minutes to ensure real-time performance. If the predicted probability exceeds 70% three times in a row, it is recorded as a potential leak event and stored in the database for subsequent analysis, logically ensuring seamless integration of warnings and data collection.

[0109] The present application also provides a dynamic leakage prediction system for old oil and gas pipelines, which is used to implement the above-mentioned dynamic leakage prediction method for old oil and gas pipelines.

[0110] The system includes: a multi-source sensor array, a data processing platform, a control execution unit, an early warning module, a model verification unit and a model deployment unit.

[0111] The multi-source sensor array, comprised of distributed fiber-optic pressure sensors, ultrasonic flowmeters, and infrared thermal imaging temperature monitoring units, is used to collect real-time pipeline pressure characteristics, flow dynamics, and temperature distribution data. The data processing platform is used to execute time series analysis algorithms, assess pipeline aging and corrosion levels, establish a set of leakage evolution dynamics equations, train leakage trend prediction models, perform numerical simulations, implement support vector machine classification predictions, activate control intervention response mechanisms, and perform feedback control and online model learning optimization. The control execution unit adjusts pipeline operating parameters in real time based on the control parameter combinations generated by the control intervention response mechanism. The early warning module generates early warning information based on the leakage risk level. The model validation unit verifies and calibrates the revised prediction model using similar experimental data. The model deployment unit deploys the validated model into an actual pipeline monitoring system.

[0112] This application establishes a complete technology chain of "sensor array → physical model training → AI prediction → control execution → feedback verification", breaking through the limitations of traditional static evaluation:

[0113] 1. The sensor array acquires multi-source real-time data: distributed fiber optic pressure sensors capture microscopic deformations, infrared thermal imaging temperature monitoring units identify temperature anomalies, and ultrasonic flow meters monitor turbulent disturbances, building a high-temporal and spatial resolution dataset.

[0114] 2. Physical model training: quantify the corrosion process through the wall thickness degradation dynamics equation and train the leakage prediction model using the long short-term memory neural network (LSTM);

[0115] 3. AI prediction: The leakage prediction model is used to calculate the oil and gas diffusion range under different pressure and flow combinations, and the support vector machine algorithm is used to classify and predict the diffusion rate;

[0116] 4. Control execution: Dynamically optimize the control strategy based on temperature distribution and material degradation parameters, and determine the control parameters;

[0117] 5. Feedback Verification: Adjust pipeline control parameters in real time, verify the stability of operating data and the downward trend of corrosion indicators through feedback algorithms, and update the leakage prediction model weights and optimize them online if the conditions are met. Use experimental data to calibrate the accuracy of the leakage prediction model, deploy it to the monitoring system after verification, and generate risk warning information based on real-time operating data.

[0118] It should be noted that the words "first", "second" and similar terms used in the specification and claims of this application do not indicate any order, quantity or importance, but are only used to distinguish different components. Similarly, words such as "one" or "an" do not indicate a quantity limitation, but rather indicate the presence of at least one. "Multiple" or "several" means at least two. Unless otherwise specified, words such as "front", "back", "left", "right", "bottom" and / or "top" are used for ease of description only and are not limited to one position or one spatial orientation. Words such as "include" or "comprising" and similar terms mean that the elements or objects appearing before "include" or "comprising" include the elements or objects listed after "include" or "comprising" and their equivalents, and do not exclude other elements or objects. Words such as "connected" or "connected" and similar terms are not limited to physical or mechanical connections, and may include electrical connections, whether direct or indirect.

[0119] As used in this specification and the appended claims, the singular forms "a," "an," "said," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0120] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the claims appended to this application.

Claims

1. A dynamic leakage prediction method for old oil and gas pipelines, characterized by: The dynamic leakage prediction method for old oil and gas pipelines includes: S101: Real-time collection of pipeline operating pressure, flow, and temperature distribution parameters. Pressure fluctuation characteristics are extracted using a time series analysis algorithm. When the characteristic value exceeds a preset threshold, the pipeline condition assessment result is calculated by combining the pipeline material degradation model and operating time. S102, establishing a leakage prediction model based on the pipeline status assessment results. When the pipeline aging degree exceeds a critical value, the oil and gas leakage probability is calculated by integrating historical data, and the diffusion velocity and concentration gradient are obtained through numerical simulation; S103: Calculate the oil and gas diffusion range under different pressure and flow combinations using a leakage prediction model, and classify and predict the diffusion rate using a support vector machine algorithm. If the predicted value exceeds a safety threshold, dynamically optimize the control strategy based on the temperature distribution and material degradation parameters, and determine the pipeline control parameters. S104: Adjust pipeline control parameters in real time and verify the stability of operating data and the downward trend of corrosion indicators through feedback algorithms. If the conditions are met, update the leakage prediction model weights and perform online optimization. S105: Calibrate the accuracy of the leakage prediction model using experimental data. After verification, deploy it to the monitoring system and generate risk warning information based on real-time operation data. In S101, the time series analysis algorithm uses the ARIMA model to analyze the pressure data to predict the pressure fluctuation characteristics of the future period and calculates the fluctuation standard deviation. When the fluctuation standard deviation exceeds a preset threshold, it is determined to be a fluctuation anomaly; at the same time, the flow and temperature distribution parameters are Fourier transformed to extract frequency characteristics to identify periodic anomaly patterns; the pipeline status assessment result calculated by combining the pipeline material degradation model with the operation time includes: calculating the tensile strength value at the current moment based on the initial tensile strength value of the pipeline material, the preset tensile strength annual decline rate value and the pipeline operation time, wherein the current tensile strength value shows an exponential decay trend with the operation time; based on the preset annual corrosion depth reference value, the cumulative total corrosion amount of the pipeline is accumulated according to the operation year; and the current tensile strength value and the cumulative total corrosion amount are combined to generate a status assessment result including an aging grade grading index and a corrosion risk grade grading index.

2. The method for predicting dynamic leakage of old oil and gas pipelines according to claim 1, characterized in that: In S102, after the leakage prediction model is established, a long short-term memory neural network is used to train the leakage prediction model, wherein the input features include the pipeline wall thickness measurement value, the pipeline pressure value, the flow value and the temperature value; based on several sets of historical operation data, the training set and the test set are proportionally constructed into a data set; and the optimizer is used to iteratively update the weights at a set learning rate.

3. The method for predicting dynamic leakage of old oil and gas pipelines according to claim 1, characterized in that: In S103, the calculation of the oil and gas diffusion range under different pressure and flow combination conditions by the leakage prediction model includes: establishing the viscous fluid motion partial differential equation and mass conservation equation of pipeline oil and gas flow; Turbulence model, setting turbulence effect compensation mechanism and no-slip wall boundary conditions; calculating the oil and gas diffusion range under different pressure and flow combination inputs.

4. A dynamic leakage prediction method for old oil and gas pipelines according to claim 1 or 3, characterized in that: In S103, using the support vector machine algorithm to classify and predict the diffusion velocity includes: setting three safety levels of low speed, medium speed, and high speed; inputting features including pressure gradient, flow dynamic value, and ambient humidity; and outputting the diffusion velocity classification result; In S103, the dynamic optimization of the control strategy based on the temperature distribution and material degradation parameters and the determination of the control parameters include: adjusting the control strategy strength through a fuzzy logic algorithm, the input of the fuzzy logic algorithm is the temperature distribution state and the material performance degradation parameters, and the output is the control strategy strength; and optimizing the control parameters through a genetic algorithm, the objective function is to minimize the oil and gas diffusion range and energy consumption, setting the population size and the number of iterations, and obtaining the optimal control parameter combination.

5. The method for dynamic leakage prediction of old oil and gas pipelines according to claim 1 is characterized in that: In S104, the conditions for updating the weight parameters of the leakage prediction model are determined by comparing whether the parameter fluctuation standard deviation after the intervention is smaller than the fluctuation standard deviation before the intervention and whether the corrosion degree assessment index decreases.

6. The method for dynamic leakage prediction of old oil and gas pipelines according to claim 1 is characterized in that: In S104, the Bayesian optimization algorithm is used to perform online optimization of the leakage prediction model, with the weighted sum of the prediction error and the calculation delay as the optimization target, the weights of the prediction error and the calculation delay are set, and the time of each optimization calculation is limited.

7. The method for dynamic leakage prediction of old oil and gas pipelines according to claim 1, characterized in that: In S105, calibrating the accuracy of the leakage prediction model using experimental data includes: inputting the experimental data set into the revised prediction model, generating simulation results using a random forest algorithm to obtain a risk prediction value, and calculating the prediction deviation between the simulation results and the experimental data based on a mean square error formula; if the prediction deviation is lower than a preset threshold, quantifying the model accuracy evaluation value using a logistic regression algorithm; In S105, the generation of risk warning information based on real-time operation data includes: collecting pressure change, flow dynamics and temperature distribution data from the deployment system in real time; performing denoising and normalization operations through a data preprocessing module to generate standardized monitoring data; using a support vector machine algorithm to analyze the standardized data and output the pipeline leakage risk level; and generating a risk warning signal according to the risk level value.

8. A dynamic leakage prediction system for old oil and gas pipelines, characterized by: The system is used to implement the dynamic leakage prediction method for old oil and gas pipelines according to any one of claims 1 to 7, and the system includes: A multi-source sensor array, comprising distributed fiber optic pressure sensors, ultrasonic flow meters, and infrared thermal imaging temperature monitoring units, is used to collect real-time pressure change characteristics, flow dynamics data, and temperature distribution status of pipeline operation; A data processing platform for executing time series analysis algorithms, assessing pipeline aging and corrosion levels, establishing a set of leakage evolution dynamics equations, training leakage development trend prediction models, performing numerical simulations, executing support vector machine algorithm classification predictions, activating control intervention response mechanisms, and performing feedback control and model online learning optimization; A control execution unit, configured to adjust pipeline operation parameters in real time according to a control parameter combination generated by a control intervention response mechanism; An early warning module, configured to generate early warning information according to a leakage risk level; A model verification unit, configured to verify and calibrate the revised prediction model using similar experimental data; A model deployment unit is used to deploy the verified model into an actual pipeline monitoring system.

Images