Seabed oil pipeline leakage positioning method based on machine learning

Abstract

The invention discloses a seabed oil pipeline leakage positioning method based on machine learning. The method comprises the steps that a seabed oil pipeline leakage fluid mechanics simulation model is established; pressure time-varying and flow time-varying original data of an inlet and an outlet of the seabed oil pipeline under the conditions of different leakage diameters and leakage positions are collected and sorted; data without pressure and flow changes when the seabed oil pipeline does not leak is removed, and an effective data set is obtained; dividing the effective data set into a training set, a verification set and a test set; constructing a random forest regression prediction model of the leakage position of the seabed oil pipeline; and carrying out combinatorial optimization on the key parameters by utilizing grid search, determining a prediction model for predicting the most accurate leakage position of the seabed oil pipeline, and outputting a prediction result. The leakage position of the oil pipeline is predicted through a machine learning method by using the inlet pressure and outlet flow time-varying data of the seabed oil pipeline, the oil pipeline monitoring efficiency is improved, and the oil pipeline monitoring cost is saved.

Description

Technical Field

[0001] The present invention relates to the technical field of offshore engineering safety monitoring, and particularly relates to a method for locating leaks in submarine oil pipelines based on machine learning. Background Art

[0002] During the use of submarine oil pipelines, leaks are likely to occur as the service life increases. This not only reduces the oil transportation efficiency of the pipelines, but also pollutes the marine ecological environment. Currently, the methods for determining the leak location of submarine oil pipelines mainly include manual line inspection and installing pressure sensors in multiple sections of the pipeline:

[0003] Manual line inspection: Manually explore the sea surface along the oil pipeline to determine the leak location of the pipeline. Due to the long transportation distance and deep burial depth of submarine oil pipelines, this method has low efficiency and high labor costs.

[0004] Installing pressure sensors in multiple sections of the pipeline: Pressure sensors are laid in sections along the submarine oil pipeline, and the real-time data of each section of the sensors are analyzed to determine the leak location of the pipeline. This method has a cumbersome process and high costs. Summary of the Invention

[0005] In view of the defects of the existing technical methods, the present invention provides a method for locating leaks in submarine oil pipelines based on machine learning, which can accurately determine the leak location of submarine oil pipelines through the pressure and flow data at the inlet and outlet of the submarine oil pipeline, and can also save costs and computational workload.

[0006] In order to achieve the above invention objectives, the technical solutions adopted by the present invention are as follows:

[0007] A method for locating leaks in submarine oil pipelines based on machine learning, comprising the following steps:

[0008] (1) Using pipeline studio software, input parameters such as pipeline length, pipeline diameter, crude oil density, viscosity, inlet flow rate, outlet pressure, leak location, and leak diameter to establish a hydrodynamic simulation model for leaks in submarine oil pipelines.

[0009] (2) Using pipeline studio software to run the hydrodynamic simulation model for leaks in submarine oil pipelines established in step (1), output, collect, and organize the time-varying inlet pressure and time-varying outlet flow rate data of the submarine oil pipeline under different leak diameters and leak location conditions to form an original data table.

[0010] (3) Delete the invalid pressure and invalid flow rate data when there is no leak in the pipeline in the original data table in step (2) to obtain an effective data set.

[0011] (4) Divide the effective data set in step (3) into a training set, a validation set, and a test set according to the ratios of 75%, 15%, and 15%.

[0012] (5) Construct a random forest regression prediction model for the leakage location of submarine oil pipelines based on machine learning. Train multiple base learners, use the prediction results of each learner for mean regression, and combine their prediction results to reduce the risk of overfitting and improve the stability of the model. The mean regression formula is:

[0013]

[0014] Where: is the final predicted value; h i (x) is the predicted value output by the i-th base learner; m is the number of base learners.

[0015] (6) Use the grid search method to optimize the combination of key parameters (including the number of decision trees, the maximum tree depth, and the minimum number of samples for internal node splitting). Traverse the predefined hyperparameter combinations, evaluate the optimization results through the mean squared error (MSE) and the mean absolute error (MAE), find the hyperparameter values that make the model performance optimal, determine the prediction model with the most accurate prediction, and output the prediction results. The mean squared error (MSE) and the mean absolute error (MAE) are as follows:

[0016]

[0017] Where: MSE is the mean squared error;

[0018] MAE is the mean absolute error;

[0019] y i is the true value of the i-th sample;

[0020] is the predicted value of the i-th sample;

[0021] n is the number of samples.

[0022] Compared with the prior art, the advantages of the present invention are:

[0023] It can predict the leakage location of oil pipelines by using the time-varying data of the inlet pressure and outlet flow rate of existing submarine oil pipelines through machine learning methods, improve the monitoring efficiency of oil pipelines, and save the monitoring cost of oil pipelines. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 is a flowchart of a method for locating submarine oil pipeline leakage based on machine learning in an embodiment;

[0025] Figure 2It is a hydrodynamic simulation model diagram of a submarine oil pipeline leakage in an embodiment;

[0026] Figure 3 It is a time-varying diagram of the inlet pressure of a submarine oil pipeline in an embodiment;

[0027] Figure 4 It is a time-varying diagram of the outlet flow rate of a submarine oil pipeline in an embodiment;

[0028] Figure 5 It is a scatter plot of the predicted value and the true value of the leakage position of a submarine oil pipeline in an embodiment. Specific implementation manner

[0029] To make the purpose, technical solutions and advantages of the present invention clearer, the following further elaborates on the present invention in detail according to the attached drawings, tables and by listing embodiments.

[0030] As Figure 1 shown, a method for locating a leakage of a submarine oil pipeline based on machine learning includes the following steps:

[0031] 1. Using pipeline studio software, input parameters such as pipeline length, pipeline diameter, crude oil density, viscosity, inlet flow rate, outlet pressure, leakage position and leakage diameter, and establish a hydrodynamic simulation model of a submarine oil pipeline leakage as Figure 2 shown.

[0032] 2. Using pipeline studio software to run the hydrodynamic simulation model of the submarine oil pipeline leakage established in step (1), output, collect and organize the time-varying data of the inlet pressure ( Figure 3 ) and the time-varying data of the outlet flow rate ( Figure 4 ) of the submarine oil pipeline under different leakage diameters and leakage positions, and form an original data table as shown in Table 1.

[0033] Table 1 Original data table of time-varying inlet pressure and time-varying outlet flow rate

[0034]

[0035]

[0036] 3. Delete the invalid pressure and invalid flow rate data when there is no leakage in the pipeline in the original data table in step (2) to obtain a valid data set as shown in Table 2.

[0037] Table 2 Valid data table of time-varying inlet pressure and time-varying outlet flow rate

[0038]

[0039] 4. Divide the effective data set obtained in step (3) into a training set (Table 3), a validation set (Table 4), and a test set (Table 5) according to the ratios of 75%, 15%, and 15%.

[0040] Table 3 Training Set Data Table

[0041]

[0042]

[0043] Table 4 Validation Set Data Table

[0044]

[0045]

[0046] Table 5 Test Set Data Table

[0047]

[0048] 5. Construct a random forest (Random Forest) regression prediction model for the leakage location of submarine oil pipelines based on machine learning, train multiple base learners, and perform mean regression using the prediction results of each learner. The formula is as follows:

[0049]

[0050] Where: is the final predicted value; h i (x) is the predicted value output by the i-th base learner; m is the number of base learners.

[0051] 6. Use the grid search method to optimize the combination of key parameters (including the number of decision trees, the maximum tree depth, and the minimum number of samples for internal node splitting), traverse the predefined hyperparameter combinations, evaluate the optimization results through the mean squared error (MSE) and the mean absolute error (MAE), find the hyperparameter values that make the model performance optimal, determine the prediction model with the most accurate prediction, and output the prediction results (such as Figure 5 the scatter plot of predicted values and true values shown), and the formulas for the mean squared error (MSE) and the mean absolute error (MAE) are as follows:

[0052]

[0053] Where: MSE is the mean squared error;

[0054] MAE is the mean absolute error;

[0055] y i is the true value of the i-th sample;

[0056] is the predicted value for the i-th sample;

[0057] n is the number of samples.

[0058] Those of ordinary skill in the art will recognize that the embodiments described herein are provided to assist the reader in understanding the implementation methods of the present invention, and should be understood that the scope of protection of the present invention is not limited to such specific statements and embodiments. Those of ordinary skill in the art can make various other specific deformations and combinations without departing from the essence of the present invention based on these technical revelations disclosed in the present invention, and these deformations and combinations are still within the scope of protection of the present invention.

Claims

1. A method for locating leaks in submarine oil pipelines based on machine learning, characterized in that, It includes the following steps: (1) Based on the pipeline studio software, establish a hydrodynamic simulation model for submarine oil pipeline leakage; (2) Collect and organize the original data of the time-varying pressure and time-varying flow rate at the inlet and outlet under different leakage positions and leakage diameters of the submarine oil pipeline; (3) Based on the original data in step (2), organize the data to obtain an effective data set; (4) Based on the effective data set in step (3), divide it into a training set, a validation set, and a test set; (5) Construct a random forest regression prediction model for the leakage position of the submarine oil pipeline based on machine learning; (6) Use grid search to optimize the combination of key parameters, determine the prediction model for the leakage position of the submarine oil pipeline with the most accurate prediction, and output the prediction result.

2. The method for locating the leakage of a submarine oil pipeline based on machine learning according to claim 1, wherein: In step (1), use the pipeline studio software to input parameters such as pipeline length, pipeline diameter, crude oil density, viscosity, inlet flow rate, outlet pressure, leakage position, and leakage diameter, and establish a hydrodynamic simulation model for submarine oil pipeline leakage.

3. A method for locating leaks in submarine oil pipelines based on machine learning according to claim 1, characterized in that: In step (2), use the pipeline studio software to run the hydrodynamic simulation model for submarine oil pipeline leakage established in step (1), output, collect, and organize the time-varying pressure and time-varying flow rate data at the inlet and outlet under different leakage positions and leakage diameters of the submarine oil pipeline, and form an original data table.

4. A method for locating leaks in submarine oil pipelines based on machine learning according to claim 1, characterized in that: In step (3), based on the original data table in step (2), delete the pressure and flow rate data when there is no leakage in the pipeline to obtain an effective data set.

5. A method for locating leaks in subsea oil pipelines based on machine learning according to claim 1, characterized in that: In step (4), perform data segmentation on the effective data set obtained in step (3), and divide it into a training set, a validation set, and a test set according to the ratio of 75%, 15%, and 15%.

6. A method for locating leaks in a submarine oil pipeline based on machine learning according to claim 1, characterized in that: In step (5), construct a random forest regression prediction model for the leakage position of the submarine oil pipeline based on machine learning, train multiple base learners, and comprehensively combine their prediction results through mean regression to reduce the risk of overfitting and improve the stability of the model. The mean regression formula is: Wherein: is the final predicted value; h i (x) is the predicted value output by the i-th base learner; m is the number of base learners.

7. A method for locating leaks in submarine oil pipelines based on machine learning according to claim 1, characterized in that: In step (6), use the grid search method to optimize the combination of key parameters, traverse the predefined hyperparameter combinations, evaluate the optimization results through the mean squared error MSE and the mean absolute error MAE, find the hyperparameter values that make the model performance optimal, determine the prediction model with the most accurate prediction, output the prediction result, and the mean squared error MSE and the mean absolute error MAE are as follows: Where: MSE is the mean squared error; MAE is the mean absolute error; y i is the true value of the i-th sample; is the predicted value for the i-th sample; n is the number of samples.

Images