Offshore process pipeline remaining life prediction method and system based on multi-source data fusion
By employing multi-source data fusion and model fusion methods, the problem of insufficient prediction accuracy for offshore process pipelines has been solved, achieving more accurate and reliable remaining life prediction. The model is dynamically adjusted to adapt to complex environments, improving the stability and reliability of predictions, reducing maintenance costs, and ensuring the safety and efficiency of offshore oil and gas production.
Patent Information
- Application Number
- CN202510981051.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-16
- Publication Date
- 2025-11-04
AI Technical Summary
Existing methods for predicting the remaining life of offshore process pipelines suffer from insufficient prediction accuracy. In particular, empirical statistical methods, physical model-based prediction methods, and data-driven prediction methods are all limited by data quality and feature selection, resulting in inaccurate and unreliable prediction results.
A multi-source data fusion approach is adopted, which collects multi-source data of offshore process pipelines, including historical operation data, environmental data, monitoring data and maintenance data. Key features are selected using a feature selection algorithm, multiple base learners are trained, and the prediction results are integrated through a model fusion algorithm. Finally, the accuracy and reliability of the prediction results are evaluated through cross-validation.
It improves the accuracy and reliability of remaining service life prediction for offshore process pipelines, can better simulate the damage evolution process in complex environments, dynamically adjust the prediction model, reduce errors, ensure the generalization ability and robustness of the model, help to formulate scientific maintenance plans, avoid excessive or insufficient maintenance, extend pipeline service life, and reduce economic losses.
Smart Images
Figure CN120893155A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the technical field of offshore oil safety, and particularly relates to a method and system for predicting the residual life of offshore process pipelines based on multi-source data fusion. BACKGROUND
[0002] Offshore process pipelines are important facilities for offshore oil and gas production and transportation, and their safety and reliability are directly related to the safety and efficiency of offshore oil and gas production. Due to the complex and changeable offshore environment, pipelines are easily damaged and have defects due to corrosion, fatigue, stress and other factors, leading to pipeline failure. Therefore, accurately predicting the residual life of offshore process pipelines is of great significance to ensuring the safety and efficiency of offshore oil and gas production.
[0003] At present, the commonly used pipeline residual life prediction methods mainly include empirical statistical methods, physical model-based prediction methods and data-driven prediction methods. The empirical statistical method mainly predicts through historical data and statistical models, but its prediction accuracy is limited by the quality and quantity of historical data. The physical model-based prediction method simulates the damage and failure process of the pipeline by establishing a physical model of the pipeline, but its prediction accuracy is limited by the accuracy of the model and the reliability of the parameters. The data-driven prediction method uses machine learning algorithms to train and predict a large amount of data, but its prediction accuracy is limited by the quality of the data and the selection of features. SUMMARY
[0004] The problem to be solved by the present application is to provide a method and system for predicting the residual life of offshore process pipelines based on multi-source data fusion, which effectively improves the accuracy and reliability of the prediction results of the residual life of offshore process pipelines.
[0005] To solve the above technical problems, the technical scheme adopted by the present application is: a method for predicting the residual life of offshore process pipelines based on multi-source data fusion, comprising the following steps,
[0006] S1: collecting multi-source data of offshore process pipelines;
[0007] S2: preprocessing the multi-source data;
[0008] S3: using a feature selection algorithm to select features that have a significant impact on the prediction of the residual life of offshore process pipelines from the preprocessed data;
[0009] S4: training multiple base learners using the selected features;
[0010] S5: using a model fusion algorithm to fuse the prediction results of the multiple base learners to obtain the final prediction result of the residual life of offshore process pipelines;
[0011] S6: Evaluate the accuracy and reliability of the prediction results using a cross-validation method.
[0012] Further, the S1 includes the following steps,
[0013] S11: Collect historical operation data in real time through pressure sensors and temperature sensors built-in the pipeline system, or export pressure historical records and temperature historical records from a data acquisition and monitoring system;
[0014] S12: Collect environmental data in real time through temperature and humidity sensors and salinity sensors;
[0015] S13: Collect monitoring data in real time through corrosion monitoring sensors, wall thickness monitoring sensors, vibration monitoring sensors, stress and strain monitoring sensors, and displacement monitoring sensors;
[0016] S14: Extract pipeline wall thickness, corrosion defect length, corrosion defect width, corrosion defect depth, crack defect length, and weld defect distribution data from the detection database;
[0017] S15: Extract repair time, repair location, repair method, and post-repair detection result data from the repair record database.
[0018] Further, the S2 includes the following steps,
[0019] S21: Use the Isolation Forest algorithm to identify abnormal data, remove or interpolate repair abnormal values caused by sensor failure, and use the Nearest Neighbor algorithm to fill in data with a missing rate of less than 10%, and mark as invalid data when the missing rate is too high;
[0020] S22: Use Min-Max to normalize continuous pressure and temperature data, with the expression as follows:
[0021]
[0022] Where: x norm is the normalized data; x max is the maximum value of the sample data; x min is the minimum value of the sample data; x is the sample data;
[0023] S23: Use the Z-Score method to standardize time series data, with the expression as follows:
[0024]
[0025] S24: Extract the mean, variance, kurtosis, skewness time domain feature statistics, extract the main frequency and harmonic frequency domain feature components of the pressure fluctuation through fast Fourier transform, and extract the spatial feature parameters through spatial interpolation of the detection data at different positions of the pipeline.
[0026] Further, the S3 comprises the following steps,
[0027] S31: Using LASSO regression method to screen features with high correlation with remaining life;
[0028] S32: According to the pipeline failure mechanism, the key features are manually labeled.
[0029] Further, in the S4, the base learner training comprises using support vector machine, random forest, neural network algorithm model to train and predict the pipeline remaining life, using early stopping method to prevent overfitting, using 5-fold cross validation to optimize model hyperparameters, using mean square error as loss function, and the expression is as follows:
[0030]
[0031] Wherein: N is the sample number; yi is the true value of the i-th sample; is the predicted value of the i-th sample.
[0032] Further, the S5 comprises the following steps,
[0033] S51: Using weighted average method, according to the performance of the base learning period on the validation set, the weight is distributed, and the formula is as follows:
[0034]
[0035] Wherein: is the final predicted pipeline remaining life; are the pipeline remaining life predicted by SVM, RF and NN methods respectively; f SVM , f RF , f NN are the weight coefficients of the prediction results of SVM, RF and NN methods respectively, f SVM +f RF +f NN =1.
[0036] S52: The weight coefficient distribution adopts the normalization weighting method, and the performance score formula of each model is as follows:
[0037]
[0038] The weight coefficient formula is as follows:
[0039]
[0040] S53: adaptively adjusting model weights according to real-time data distribution changes.
[0041] Further, in the S6, the result evaluation includes evaluating the mean absolute error (MAE), the determination coefficient (R 2 The index is evaluated, and the expression is as follows:
[0042]
[0043] Further, the present application provides a marine process pipeline residual life prediction system based on multi-source data fusion, which utilizes the above-mentioned marine process pipeline residual life prediction method based on multi-source data fusion, comprising,
[0044] The data acquisition module is used for acquiring multi-source data of the marine process pipeline, including historical operation data, environmental data, monitoring data, detection data and maintenance data of the pipeline.
[0045] The data preprocessing module is used for preprocessing the collected multi-source data, including data cleaning, data normalization and feature extraction.
[0046] The feature selection module is used for selecting features that have a significant impact on pipeline residual life prediction from the preprocessed data by using a feature selection algorithm.
[0047] The model training module is used for training multiple base learners, including support vector machines, random forests and neural networks, by using the selected features.
[0048] The model fusion module is used for fusing the prediction results of the multiple base learners by using a model fusion algorithm to obtain the final pipeline residual life prediction result.
[0049] The result evaluation module is used for evaluating the accuracy and reliability of the prediction result by using a cross-validation method.
[0050] The report generation module is used for generating a pipeline residual life evaluation report.
[0051] Further, the present application provides an electronic device comprising a memory, a processor and an algorithm stored in the memory and executable on the processor, wherein the processor executes the above-mentioned data processing method.
[0052] Further, the present application provides a computer readable storage medium, which stores a computer algorithm, wherein the computer algorithm is executed by a processor to realize the above-mentioned data processing method.
[0053] The present application has the advantages and positive effects that:
[0054] 1、The present application can more comprehensively reflect the actual operation state and damage of offshore process pipelines by fusing multi-source data, thereby improving the accuracy of residual life prediction. Multiple base learners are used for model training, and the prediction results of multiple models are integrated through model fusion algorithm, further improving the accuracy and stability of prediction. Multi-source data reduces the error and uncertainty that may be caused by a single data source, improving the reliability of the prediction results. The prediction results are evaluated by cross-validation method to ensure the generalization ability and robustness of the model, avoiding overfitting problem.
[0055] 2、Offshore process pipelines are subjected to corrosion, fatigue, stress and other factors for a long time in complex marine environment. The present application can better simulate and predict the damage evolution process of the pipeline under complex environment by introducing environmental data. The method can dynamically adjust the prediction model to adapt to the operating state of the pipeline under different sea areas and environmental conditions. By accurately predicting the residual life of the pipeline, it can help the maintenance personnel to develop more scientific maintenance plan, avoid over-maintenance or insufficient maintenance, thereby reducing the maintenance cost. Timely discovery of potential high-risk pipeline sections can avoid sudden failure accidents and reduce production downtime and economic losses caused by pipeline failure. Through accurate residual life prediction and scientific maintenance strategy, the service life of the pipeline can be effectively prolonged, and the resource utilization rate can be improved. For the pipeline sections with shorter life in the prediction results, targeted repair or replacement measures can be taken to avoid the overall pipeline from being scrapped due to local failure. The present application can effectively cope with the influence of complex marine environment on the pipeline, and provide strong guarantee for the safety and efficiency of offshore oil and gas production. BRIEF DESCRIPTION OF DRAWINGS
[0056] Figure 1 is the overall flowchart of the method embodiment of the present application.
[0057] Figure 2 is the overall framework diagram of the system embodiment of the present application. DETAILED DESCRIPTION
[0058] The technical solutions of the present application will be described clearly and completely below in combination with the drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.
[0059] The embodiments of the present application will be further described below in combination with the drawings:
[0060] As shown in Figure 1 , the residual life prediction method for offshore process pipelines based on multi-source data fusion includes the following steps.
[0061] S1: Collecting multi-source data of offshore process pipeline. Specifically, the multi-source data includes historical operation data, environmental data, monitoring data, detection data, and maintenance data of the pipeline. Specifically, S1 includes the following steps,
[0062] S11: Collecting historical operation data in real time through pressure sensors and temperature sensors built in the pipeline system, or exporting pressure history records and temperature history records from the data acquisition and monitoring system to collect pressure fluctuation data and temperature change data.
[0063] S12: Collecting environmental data in real time through temperature and humidity sensors and salinity sensors, such as air temperature and humidity and seawater salinity.
[0064] S13: Collecting monitoring data in real time through corrosion monitoring sensors, wall thickness monitoring sensors, vibration monitoring sensors, stress and strain monitoring sensors, and displacement monitoring sensors, such as corrosion rate, wall thickness change, vibration frequency, stress and strain, and displacement change.
[0065] S14: Extracting pipeline wall thickness, corrosion defect length, corrosion defect width, corrosion defect depth, crack defect length, and weld defect distribution data from the detection database.
[0066] S15: Extracting maintenance time, maintenance location, repair method, and post-maintenance detection result data from the maintenance record database.
[0067] S2: Preprocessing multi-source data, including data cleaning, data normalization, and feature extraction. Specifically, S2 includes the following steps,
[0068] S21: Using the Isolation Forest algorithm to identify abnormal data, removing or interpolating abnormal values caused by sensor failure, filling data with a missing rate of less than 10% using the nearest neighbor algorithm, and marking invalid data when the missing rate is too high.
[0069] S22: Using Min-Max to normalize continuous data such as pressure and temperature, with the specific expression as follows:
[0070]
[0071] where: x norm is the normalized data; x max is the maximum value of the sample data; x min is the minimum value of the sample data; x is the sample data.
[0072] S23: Using the Z-Score method to standardize time series data such as pressure fluctuation, with the specific expression as follows:
[0073]
[0074] S24: Extract the mean, variance, kurtosis, skewness time domain feature statistics, extract the main frequency, harmonic frequency domain feature components of pressure fluctuation through FFT, and perform spatial interpolation on the detection data of different positions of the pipeline to extract spatial feature parameters.
[0075] S3: Using a feature selection algorithm, select features that have a significant impact on the remaining life prediction of the offshore process pipeline from the preprocessed data. Specifically, S3 includes the following steps,
[0076] S31: Use LASSO regression method to screen features with high correlation with the remaining life;
[0077] S32: According to the failure mechanism of the pipeline, such as the positive correlation between corrosion rate and salinity, manually label key features such as salinity and wall thickness reduction rate.
[0078] S4: Using the selected features, train multiple base learners. Specifically, the base learner training includes using support vector machine (SVM), random forest (RF), and neural network (NN) algorithm models to train and predict the remaining life of the pipeline, using early stopping method to prevent overfitting, using 5-fold cross-validation to optimize model hyperparameters, and using mean square error (MSE) as the loss function, the specific expression is as follows:
[0079]
[0080] Where: N is the number of samples; yi is the true value of the ith sample; is the predicted value of the ith sample.
[0081] S5: Using a model fusion algorithm, fuse the prediction results of multiple base learners to obtain the final offshore process pipeline remaining life prediction result. Specifically, S5 includes the following steps,
[0082] S51: Using a weighted average method, assign weights according to the performance of the base learning period on the validation set, the formula is as follows:
[0083] Where: is the final predicted pipeline remaining life; are the pipeline remaining life predicted by SVM, RF, and NN methods respectively; f SVM , f RF , f NN are the weight coefficients of the SVM, RF, and NN method prediction results, f SVM +f RF +f NN = 1.
[0084] S52: The weight coefficient distribution adopts a normalized weighting method, and the specific steps are as follows:
[0085] The performance scores of each model are calculated: (k=SVM, RF, NN);
[0086] The weight coefficients are calculated:
[0087] S53: According to the real-time data distribution changes (such as extreme environmental events), the model weight is adaptively adjusted. For example, when the salinity and humidity suddenly increase, the weight of the corrosion-sensitive model is increased. After the weight coefficient of a certain model is adjusted, the weight coefficients of other models are adjusted in proportion. The specific adjustment method is as follows: if the basic weight coefficient of SVM is f SVM adjusted to f' SVM , the basic weight coefficients of RF and NN models are adjusted to:
[0088]
[0089] S6: The accuracy and reliability of the prediction results are evaluated by using the cross-validation method. The result evaluation includes evaluating the mean absolute error MAE, the determination coefficient R 2 and other indicators. The specific expressions are as follows:
[0090]
[0091] As shown in Figure 2 , the application also provides a marine process pipeline residual life prediction system based on multi-source data fusion, which comprises,
[0092] A data acquisition module is configured to acquire multi-source data of the marine process pipeline, including historical operation data, environmental data, monitoring data, detection data and maintenance data of the pipeline.
[0093] A data preprocessing module is configured to preprocess the acquired multi-source data, including data cleaning, data normalization and feature extraction.
[0094] A feature selection module is configured to select features that have a significant impact on the pipeline residual life prediction from the preprocessed data by using a feature selection algorithm.
[0095] A model training module is configured to train multiple base learners, including support vector machines, random forests and neural networks, by using the selected features.
[0096] A model fusion module is configured to fuse the prediction results of the multiple base learners by using a model fusion algorithm to obtain the final pipeline residual life prediction result.
[0097] Result evaluation module: for evaluating the accuracy and reliability of the prediction results using cross-validation method.
[0098] Report generation module: for generating pipeline remaining life assessment report.
[0099] The application will be specifically described below in combination with specific embodiments:
[0100] The evaluated offshore platform process pipeline is a 12-year-old crude oil pipeline with a design pressure of 10.2 MPa, a design temperature of 121℃, a pipeline material of A106-B, a nominal diameter of DN150mm, and a nominal wall thickness of 10.97mm.
[0101] S1: data acquisition, specifically, including the following steps.
[0102] Obtain pressure data and temperature data through pressure, temperature and other sensors built-in the pipeline system.
[0103] Obtain air temperature and humidity and seawater salinity data through temperature and humidity sensors and salinity sensors.
[0104] Obtain corrosion rate, residual wall thickness, vibration frequency, stress and strain, displacement and other data through corrosion, wall thickness, vibration, stress and strain, displacement and other monitoring sensors.
[0105] Extract pipeline state data such as wall thickness, corrosion defect length, width, depth, crack defect length, and weld defect distribution from the detection database.
[0106] Extract data such as repair time, repair location, repair method, and post-repair detection results from the repair record database.
[0107] Monitoring data: real-time collection through external monitoring equipment of the pipeline system (online corrosion monitoring system, online residual wall thickness monitoring system, vibration monitor, stress and strain measuring instrument, displacement monitoring equipment, etc.). Data types: including pipeline wall thickness, corrosion rate, vibration frequency, stress, displacement, etc.
[0108] Detection data: periodically obtain pipeline state data through non-destructive testing techniques (such as ultrasonic thickness measurement, magnetic powder detection, ultrasonic detection, magnetic flux leakage detection, etc.). Data types: pipeline wall thickness, corrosion defect length, width, depth, crack defect length, and weld defect distribution.
[0109] Repair data: extracted from the repair record database, including repair time, repair location, repair method (such as repair welding, coating repair), post-repair detection results, etc. The specific collected multi-source data is shown in Table 1.
[0110] Table 1: Collected multi-source data
[0111]
[0112] S2: Data preprocessing, specifically, comprising the following steps.
[0113] Isolated forest algorithm is used to identify and eliminate pressure transient peak value (> 10 MPa, a total of 5 abnormal points).
[0114] The missing humidity data (missing rate 2%) is filled by adjacent algorithm (k = 3).
[0115] Min-Max normalization is performed on pressure and temperature: pressure normalization: Temperature normalization:
[0116] Temporal features are extracted: pressure fluctuation variance: 0.25 MPa 2 , temperature change rate: 0.1℃ / min.
[0117] Spatial features are extracted: corrosion defect aspect ratio: 320mm / 86mm≈3.72.
[0118] S3: Feature selection, specifically, comprising the following steps.
[0119] LASSO regression method is used for feature selection to retain key features, specifically, the key feature coefficients are shown in Table 2.
[0120] Table 2 Key features
[0121] Features Coefficients Corrosion rate 0.85 Current minimum wall thickness -0.78 Corrosion defect depth 0.62 Seawater salinity 0.45
[0122] S4: Model training, specifically, comprising the following steps.
[0123] Random forest feature importance ranking, specifically, as shown in Table 3.
[0124] Table 3 Top 5 feature weights
[0125] Features Importance Current minimum wall thickness 0.32 Corrosion rate 0.28 Corrosion defect depth 0.18 Pressure fluctuation variance 0.12 Air humidity 0.10
[0126] Support vector machine (SVM), random forest (RF), and neural network (NN) algorithm models are used to train the samples;
[0127] (1) Support vector machine (SVM): kernel function: RBF, C = 1.2, γ = 0.008;
[0128] (2) Random forest (RF): number of trees: 300, maximum depth: 12, feature subset ratio: 0.6;
[0129] (3) Neural Network (NN): Structure: input layer (6 nodes) - hidden layer (64 nodes, ReLU) - output layer (1 node);
[0130] Optimizer: Adam, initial learning rate 0.002, Dropout rate 0.3.
[0131] Cross-validation results, specifically, as shown in Table 4.
[0132] Table 4 5-fold cross-validation index (unit: year)
[0133] Model MAE R2 SVM 0.75 0.84 RF 0.55 0.89 NN 0.60 0.87
[0134] S5: Model fusion, specifically, comprising the following steps.
[0135] Basic weight distribution.
[0136]
[0137] The pipeline remaining life predicted by the method is 8.5 years, and the confidence interval is 7.8 to 9.2 years (95% confidence).
[0138] S6: Result evaluation, specifically, comprising the following steps.
[0139] The pipeline remaining life calculated by the method in the ASME B31G standard is 9.7 years, and the calculation process is as follows:
[0140] Allowable residual wall thickness t 允 = 0.8 x 10.97 = 8.78 mm.
[0141] Critical residual wall thickness t 临 = P x D / (2 x S x E) = 8.2 x 150 / (2 x 138 x 1.0) = 4.45 mm.
[0142] Remaining life: n = (t 允 -t 临 ) / corrosion rate = (7.2-4.45) / 0.3 = 9.17 years.
[0143] After the process pipeline continues to be used for 8 years, it is found through detection that the remaining wall thickness is 4.6 mm, which is close to the critical residual wall thickness of 4.45 mm.
[0144] The pipeline remaining life predicted by the method is 8.5 years, and the error with the actual remaining life of 8 years is only 6.25%, while the pipeline remaining life calculated by the method of ASME B31G standard is 9.17 years, and the error with the actual remaining life is 14.625%. By comparison, the accuracy and reliability of the present aspect are fully demonstrated.
[0145] The present application has the advantages and positive effects that:
[0146] 1、The present application can more comprehensively reflect the actual operation state and damage condition of offshore process pipelines by fusing multi-source data, thereby improving the accuracy of residual life prediction. Multiple base learners are used for model training, and the prediction results of multiple models are integrated through a model fusion algorithm, further improving the accuracy and stability of prediction. Multi-source data reduces the errors and uncertainties that may be caused by a single data source, improving the reliability of the prediction results. The prediction results are evaluated through cross-validation methods to ensure the generalization ability and robustness of the model, avoiding overfitting problems.
[0147] 2、Offshore process pipelines are subjected to corrosion, fatigue, stress and other factors for a long time in complex marine environments. The present application can better simulate and predict the damage evolution process of the pipeline under complex environmental conditions by introducing environmental data. This method can dynamically adjust the prediction model to adapt to the operating state of the pipeline under different sea areas and environmental conditions. By accurately predicting the residual life of the pipeline, it can help maintenance personnel develop more scientific maintenance plans to avoid over-maintenance or insufficient maintenance, thereby reducing maintenance costs. Potential high-risk pipeline sections are discovered in a timely manner to avoid sudden failure accidents and reduce production downtime and economic losses caused by pipeline failure. Through accurate residual life prediction and scientific maintenance strategies, the service life of the pipeline can be effectively extended, and resource utilization can be improved. For pipeline sections with shorter life in the prediction results, targeted repair or replacement measures can be taken to avoid the scrapping of the entire pipeline due to local failure. The present application can effectively cope with the impact of complex marine environments on pipelines, providing strong protection for the safety and efficiency of offshore oil and gas production.
[0148] The above describes one embodiment of the present application in detail, but the content described is only the preferred embodiment of the present application and cannot be considered as limiting the scope of the present application. Any equivalent changes and improvements made in accordance with the scope of the present application should still be within the scope of the present application.
Claims
1. A method for predicting the remaining life of offshore process pipelines based on multi-source data fusion, characterized in that: Includes the following steps, S1: Collect multi-source data on offshore process pipelines; S2: Preprocess the multi-source data; S3: Using a feature selection algorithm, select features that have a significant impact on the prediction of the remaining life of offshore process pipelines from the preprocessed data; S4: Using the selected features, train multiple base learners; S5: Using a model fusion algorithm, the prediction results of multiple base learners are fused to obtain the final prediction result of the remaining life of the offshore process pipeline; S6: Use cross-validation to evaluate the accuracy and reliability of the prediction results.
2. The method for predicting the remaining life of offshore process pipelines based on multi-source data fusion according to claim 1, characterized in that: S1 includes the following steps: S11: Historical operating data is collected in real time through the pressure and temperature sensors built into the pipeline system, or historical pressure and temperature data are exported from the data acquisition and monitoring system. S12: Real-time environmental data is collected via temperature and humidity sensors and salinity sensors; S13: Real-time monitoring data is collected using corrosion monitoring sensors, wall thickness monitoring sensors, vibration monitoring sensors, stress-strain monitoring sensors, and displacement monitoring sensors; S14: Extract pipe wall thickness, corrosion defect length, corrosion defect width, corrosion defect depth, crack defect length, and weld defect distribution data from the inspection database; S15: Extract maintenance time, maintenance location, repair method, and post-maintenance test results data from the maintenance record database.
3. The method for predicting the remaining life of offshore process pipelines based on multi-source data fusion according to claim 1 or 2, characterized in that: S2 includes the following steps: S21: The isolated forest algorithm is used to identify abnormal data, and outliers caused by sensor failures are removed or interpolated for repair. Data with a missing rate of less than 10% is filled with the nearest neighbor algorithm, and data with a missing rate of too high is marked as invalid data. S22: Min-Max normalization is performed on continuous pressure and temperature data, as shown in the following expression: Where: x norm The data is normalized; x max x represents the maximum value of the sample data. min The minimum value of the sample data is x; x represents the sample data. S23: The Z-Score method is used to standardize the time series data. The expression is as follows: S24: Extract time-domain characteristic statistics of mean, variance, kurtosis and skewness, extract the main frequency and harmonic frequency domain characteristic components of pressure fluctuation through fast Fourier transform, and extract spatial characteristic parameters by spatial interpolation of detection data at different locations of the pipeline.
4. The method for predicting the remaining life of offshore process pipelines based on multi-source data fusion according to claim 1 or 2, characterized in that: S3 includes the following steps: S31: Use LASSO regression to screen for features that are highly correlated with remaining lifespan; S32: Based on the pipeline failure mechanism, key features are manually marked.
5. The method for predicting the remaining life of offshore process pipelines based on multi-source data fusion according to claim 1 or 2, characterized in that: In step S4, the base learner training includes training the samples using support vector machines, random forests, and neural network algorithms to predict the remaining pipeline lifespan. Early stopping is used to prevent overfitting, 5-fold cross-validation is used to optimize the model hyperparameters, and mean squared error is used as the loss function, expressed as follows: Where: N is the number of samples; yi is the true value of the i-th sample; Let be the predicted value for the i-th sample.
6. The method for predicting the remaining life of offshore process pipelines based on multi-source data fusion according to claim 1 or 2, characterized in that: S5 includes the following steps: S51: A weighted average method is used, with weights assigned based on the performance on the validation set during the base learning period, as shown in the following formula: in: This represents the final predicted remaining lifespan of the pipeline; The remaining pipeline life predicted using SVM, RF, and NN methods are respectively; f SVM f RF f NN The weight coefficients for the prediction results of SVM, RF, and NN methods are respectively, f SVM +f RF +f NN =1; S52: The weight coefficients are allocated using a normalized weighting method. The formula for calculating the performance score of each model is as follows: The formula for calculating the weighting coefficient is as follows: S53: Adaptively adjust model weights based on real-time data distribution changes.
7. The method for predicting the remaining life of offshore process pipelines based on multi-source data fusion according to claim 1 or 2, characterized in that: In S6, the result evaluation includes the mean absolute error (MAE) and the coefficient of determination (R²). 2 The indicator is evaluated using the following expression:
8. A marine process pipeline remaining life prediction system based on multi-source data fusion, utilizing the marine process pipeline remaining life prediction method based on multi-source data fusion as described in any one of claims 1 to 7, characterized in that: include, Data acquisition module: used to collect multi-source data of offshore process pipelines, including historical pipeline operation data, environmental data, monitoring data, inspection data and maintenance data; Data preprocessing module: used to preprocess the collected multi-source data, including data cleaning, data normalization and feature extraction; Feature selection module: Used to select features that have a significant impact on the prediction of pipeline remaining life from the preprocessed data using feature selection algorithms; Model training module: used to train multiple base learners, including support vector machines, random forests, and neural networks, using selected features; Model fusion module: Used to fuse the prediction results of multiple base learners using a model fusion algorithm to obtain the final pipeline remaining life prediction result; Results evaluation module: used to evaluate the accuracy and reliability of the prediction results using cross-validation methods; Report generation module: Used to generate pipeline remaining life assessment reports.
9. An electronic device comprising a memory, a processor, and an algorithm stored in the memory and executable on the processor, characterized in that: The processor performs the data processing method as described in any one of claims 1 to 7.
10. A computer-readable storage medium storing a computer algorithm, characterized in that, When the computer algorithm is executed by the processor, it implements the data processing method as described in any one of claims 1 to 7.