A method for obtaining quantitative analysis of a magnetic signal submarine cable wire break

Abstract

The application discloses a kind of based on magnetic signal submarine cable wire break quantitative analysis acquisition method, belong to the data processing technical field of submarine cable wire break magnetic signal.For submarine cable wire break defect magnetic signal, test area is established, with the number of submarine cable wire break and the initial variable of improving separation distance, obtain X direction magnetic signal and Y direction magnetic signal, calculate the original eigenvalue, and the eigenvalue data is enhanced, finally obtain multiple factors influencing magnetic signal submarine cable wire break quantitative, the method comprehensively considers multiple influencing factors, obtains derivative characteristics from the eigenvalue with large correlation coefficient, improves the problem that model prediction accuracy is low due to low original data quality;Compared with optimizing model structure, the effect of improving model prediction accuracy by improving input data quality is more significant, and greatly saves calculation amount and calculation time, realizes based on magnetic signal submarine cable wire break quantitative analysis, promotes the development of submarine cable detection.

Description

Technical Field

[0001] The present invention belongs to the field of data processing of magnetic signals of broken submarine cables, and in particular relates to a method for quantitatively analyzing and acquiring broken submarine cables based on magnetic signals. Background Art

[0002] To achieve the goal of clean, low-carbon development, vigorous efforts are being made to develop clean energy. Offshore wind farms, known for their environmental friendliness and lack of land resources, have seen a significant increase in their presence over the past few decades, but they also face increasingly prominent challenges. Submarine cables are insulating conductors laid beneath the seabed for power transmission. Electricity generated by power generation equipment is transported via these cables to power stations for voltage step-up and step-down processing. Damage to these cables can impact power transmission. Existing data indicates that damage to submarine cables often occurs due to breakage of the steel wires in the armor layer, exposing the internal copper wires. This leads to current leakage and prevents power transmission. Therefore, detecting damage to the armor wires in in-service submarine cables is crucial for ensuring the safe operation of wind farms. Various submarine cable inspection technologies are available, including optical, electrical, acoustic, magnetic, and information fusion. Each technology has its own principles, advantages, and disadvantages, and the appropriate technology must be selected based on factors such as the inspection objective, accuracy, and cost. The challenges of submarine cable inspection lie in the complexity, uncertainty, and dynamics of the submarine environment, as well as the long distances, high voltages, and low signal-to-noise ratios of submarine cables. These factors present significant difficulties and challenges in locating, identifying, and repairing submarine cable faults. The current trend in submarine cable inspection is toward intelligent, automated, integrated, and low-cost approaches. Technologies such as artificial intelligence, big data, and the Internet of Things are being leveraged to improve the efficiency and accuracy of submarine cable inspection. Platforms such as drones, unmanned vessels, and unmanned submersibles are being used to enhance the flexibility and safety of submarine cable inspection. Technologies such as multi-source information fusion and multimodal sensors are being used to improve the reliability and robustness of submarine cable fault detection. New materials, processes, and designs are being utilized to reduce the cost and risk of submarine cable inspection.

[0003] Existing magnetic signal monitoring equipment has a sensitivity of 1nT, meeting the requirements for underwater magnetic signal monitoring. However, research on quantifying broken wires in submarine cable armor based on magnetic signals is still immature. Detection methods based on artificial intelligence and big data models offer the advantages of high efficiency and precision, but their advantages depend on high-quality model input data. Currently, underwater magnetic signal acquisition primarily involves two signals: horizontal and vertical magnetic signals. The raw feature modeling extracted from these two signals is ineffective, necessitating an effective solution to this problem. Summary of the Invention

[0004] The purpose of the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a method for quantitative analysis and acquisition of broken submarine cables based on magnetic signals, so as to solve the problems of poor feature modeling effect and low simulation accuracy in the prior art.

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

[0006] A method for quantitatively analyzing and obtaining broken submarine cables based on magnetic signals comprises the following steps:

[0007] Step 1: placing the submarine cable in a static magnetic field, taking the number of cable wire breaks and the increase in separation as initial variables, collecting magnetic signals at the cable wire break locations, wherein the magnetic signals include X-direction magnetic signals and Y-direction magnetic signals;

[0008] Step 2: Extracting original eigenvalues ​​from the X-direction magnetic signal and the Y-direction magnetic signal, wherein the original eigenvalues ​​include X amplitude, X area, X maximum gradient, X average gradient, Y amplitude, Y maximum gradient, Y average gradient, and lift-off height, setting the number of fractures as the target value, and establishing an initial model;

[0009] Step 3, calculate the Pearson correlation coefficient of the original eigenvalue;

[0010] Step 4: Sort by the size of the Pearson correlation coefficient, select the first four original eigenvalues ​​with the largest absolute value, calculate and obtain several derived eigenvalues ​​based on the first four original eigenvalues, select the first two derived eigenvalues ​​with the largest absolute value as the derived features, and at the same time obtain the original feature with the largest absolute value as the original eigenvalue one;

[0011] Step 5: Based on the improvement of separation, the maximum, minimum and average values ​​of the derived features and the original feature value in each group are calculated, and the separation height after one-hot encoding is collected to obtain the final feature value;

[0012] Step 6: After the new eigenvalues ​​are introduced into the regression prediction model, a quantitative analysis model of the magnetic signal submarine cable breakage is obtained.

[0013] A further improvement of the present invention is:

[0014] Preferably, in step 1, when the collected magnetic signal is inconsistent with the existing magnetic signal variation pattern, the static magnetic field or the variable is adjusted until the magnetic signal is consistent with the existing magnetic signal variation pattern.

[0015] Preferably, in step 4, the calculation formula of the derived eigenvalue is:

[0016] Derived eigenvalue 1 = original eigenvalue 1 × original eigenvalue 2

[0017] Derived eigenvalue 2 = original eigenvalue 2 × original eigenvalue 3

[0018] Derived eigenvalue three = original eigenvalue one × original eigenvalue two × original eigenvalue three.

[0019] Preferably, in step 3, the original eigenvalue is substituted into the regression prediction model to calculate the initial error; in step 4, the derived feature and the original eigenvalue are input into the regression prediction model as new features, the new error is calculated, and the initial error and the new error are compared. When the new error is smaller than the initial error, it indicates that the derived feature selection meets the requirements.

[0020] Preferably, in step 3, the Pearson correlation coefficient formula is:

[0021]

[0022] Where: n is the number of samples in the population, X i is the eigenvalue of the i-th sample, is the average value of the eigenvalue, Y i is the target value of the i-th sample, is the average of the target values.

[0023] Preferably, in step 4, the original eigenvalue 1 is the Y average gradient.

[0024] Preferably, in step 6, the regression prediction model is a GBDT regression prediction model.

[0025] Preferably, in step 1, the submarine cable is approximately regarded as an abnormal shape of an infinitely long horizontal cylinder, and the seabed is regarded as a rectangular domain, and the magnetic signal at the break of the submarine cable wire is simulated and collected.

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] The present invention discloses a method for quantitative analysis and acquisition of broken submarine cables based on magnetic signals. The method enhances the characteristic value data of magnetic signals of broken submarine cables. The method comprehensively considers multiple influencing factors and obtains derived features from characteristic values ​​with large correlation coefficients, thereby improving the problem of low model prediction accuracy caused by low quality of original data. Compared with optimizing the model structure, improving the quality of input data has a more significant effect on improving the model prediction accuracy, and greatly saves the amount of calculation and calculation time, thereby realizing quantitative analysis of broken submarine cables based on magnetic signals and promoting the development of submarine cable detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a submarine cable model diagram;

[0029] Figure 2 This is a cross-sectional diagram of the submarine cable;

[0030] Figure 3 is a flow chart of the method;

[0031] Figure 4 Schematic diagram for modeling submarine cables;

[0032] Figure 5 This is the magnetic signal diagram of sample 1;

[0033] Figure 6 This is the magnetic signal diagram of sample 2. DETAILED DESCRIPTION

[0034] The present invention will be described in further detail below with reference to the accompanying drawings and specific embodiments:

[0035] In the description of the present invention, it should be noted that the terms "center", "up", "down", "left", "right", "vertical", "horizontal", "inside", "outside" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the devices or components referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention; the terms "first", "second", and "third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance; in addition, unless otherwise expressly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection or a detachable connection; it can be a direct connection or an indirect connection through an intermediate medium, or it can be a communication between the internal parts of two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0036] The present invention provides a method for processing characteristic value data of a dual-path signal combining horizontal and vertical magnetic signals of a submarine cable. The submarine cable is magnetized in a geomagnetic field environment, and different numbers of broken wires are set. A high-precision magnetic sensor is used to collect magnetic signals at defects in the submarine cable. The direction parallel to the direction of the submarine cable is defined as the X direction and the direction of the seawater depth is defined as the Y direction. The lifting height of the magnetic sensor is changed to simulate different burial depths of the submarine cable under the seabed. Common characteristic values ​​of magnetic signals include amplitude, area, maximum gradient, average gradient, and lifting height, which are called original dual-path magnetic signal characteristic values. Based on the original characteristic values, new features are created by dividing continuous and discrete features, analyzing the correlation coefficient between the characteristic value and the target value, and using four arithmetic operations, group statistical features, screening, and feature derivation methods. The new features are substituted into the machine learning model, that is, improving the prediction accuracy by improving the quality of input data before model training.

[0037] The submarine cable structure of the present invention is as follows Figure 1 and Figure 2 As shown, the interior is a core cable 2, and a circle of steel wires 1 is arranged around the circumference of the core cable 2. The length direction of the steel wires 1 is parallel to the axial direction of the core cable 2, and all the steel wires 1 constitute an armor layer; the method of the present invention is based on the fracture of the steel wires 1 to perform quantitative analysis.

[0038] See also Figure 2To achieve the above-mentioned purpose, the present invention discloses a method for quantitatively analyzing and obtaining broken submarine cables based on magnetic signals, comprising the following steps:

[0039] S1. Use finite element software to establish a three-dimensional solid model of the steel wire armor structure, simulate models with different numbers of broken steel wires, and set a static magnetic field environment with a magnetic permeability of 1.

[0040] S2. Use the number of broken wires and the lift-off height as two variables to establish a sample; place the three-dimensional model of the sample in a static magnetic field to collect X- and Y-direction magnetic signals at the defective part of the submarine cable (the broken wire); determine whether the changing pattern of the magnetic signal is consistent with the existing changing pattern of the magnetic signal as the number of broken wires and the lift-off height change. If they are consistent, execute S3; otherwise, adjust the entire simulation model, such as adjusting the variables or adjusting the static magnetic field, until the changing pattern of the magnetic signal is consistent with the existing changing pattern of the magnetic signal.

[0041] S3. Extract the original eigenvalues ​​corresponding to each sample. The original eigenvalues ​​of the magnetic signals in the X and Y directions include: X amplitude, X area, X maximum gradient, X average gradient, Y amplitude, Y maximum gradient, Y average gradient, and lift-off height. The target value is the corresponding number of broken wires.

[0042] It should be understood that the model of the embodiment of the present invention is intended to determine the final number of broken wires, and therefore the lift-off height at the beginning of model establishment is also used as the original eigenvalue.

[0043] S4. Raw Data Processing

[0044] Calculate the correlation coefficients of the original eigenvalue, derived eigenvalue, and target value, and select eigenvalues ​​with higher correlation through feature screening method;

[0045] Substitute the original eigenvalues ​​into the regression prediction model and calculate the initial root mean square error of the test set, which is considered the initial error. Calculate the Pearson correlation coefficient of the original eigenvalues. Arrange the correlation coefficients from largest to smallest by absolute value, and take the first four original coefficients with the highest correlation coefficients to calculate the derived eigenvalues. The preferred regression prediction model is the GBDT model.

[0046] In descending order of absolute value, the four original eigenvalues ​​are native eigenvalue 1, native eigenvalue 2, native eigenvalue 3, and native eigenvalue 4; the calculation formula for the derived eigenvalue is:

[0047] Derived eigenvalue 1 = original eigenvalue 1 × original eigenvalue 2

[0048] Derived eigenvalue 2 = original eigenvalue 2 × original eigenvalue 3

[0049] Derived eigenvalue 3 = original eigenvalue 1 × original eigenvalue 2 × original eigenvalue 3

[0050] Arrange the three derived eigenvalues ​​from large to small according to the absolute value, select the two largest derived eigenvalues, and form a new feature matrix together with the original eigenvalues. Calculate the root mean square error of the test set and compare it with the initial error. If the root mean square error decreases, it is considered that the fitting accuracy is further improved.

[0051] Taking the lift-off height as the grouping basis, the derived eigenvalue one, derived eigenvalue two and original eigenvalue one are grouped and counted, and the maximum, minimum and average values ​​of the derived eigenvalue one, derived eigenvalue two and original eigenvalue one are calculated in each group; combined with the lift-off height after unique hot encoding, a new eigenvalue is obtained. After bringing the new eigenvalue into the GBDT regression model, the fitting coefficients of the calculation formula for quantitative analysis of magnetic signal submarine cable breakage are obtained, and then the quantitative analysis model of magnetic signal submarine cable breakage is obtained.

[0052] As the correlation coefficient with the target value is higher.

[0053] Furthermore, the correlation analysis, feature derivation, feature screening, and methods in this step are described as follows:

[0054] (1) Data correlation analysis

[0055] Data correlation analysis is a statistical method used to evaluate the strength and direction of the linear relationship between two or more variables. Data correlation analysis is used to measure the mutual influence between variables, discover potential causal relationships, or select appropriate predictive models. The commonly used indicator for data correlation analysis is the correlation coefficient, which is a numerical value between -1 and 1 that indicates the degree of positive or negative correlation between variables. The closer the correlation coefficient is to 1 or -1, the stronger the linear relationship between the variables; the closer the correlation coefficient is to 0, the weaker the linear relationship between the variables. The steps of data correlation analysis generally include: collecting and organizing data, selecting the appropriate type of correlation coefficient, calculating and testing the significance of the correlation coefficient, and interpreting and presenting the results. Data correlation analysis is a basic and important data analysis technique that can be applied to various fields and industries. For the continuous variable eigenvalues ​​in the present invention, the correlation coefficient selects the Pearson correlation coefficient, which can measure the degree of linear correlation between two variables. The calculation formula of the Pearson correlation coefficient is as follows:

[0056]

[0057] Where: n is the number of samples in the population, X i is the eigenvalue of the i-th sample, is the average value of the eigenvalue, Y i is the target value of the i-th sample, is the average of the target values.

[0058] (2) Feature Derivation

[0059] The concept of feature derivation refers to constructing new features from raw data, and is also a means of feature selection. The purpose of feature derivation is to improve the information value of features, express the relationship between features and target variables, and enhance the effectiveness and interpretability of the model. The two common types of feature derivation are feature transformation and feature combination. Feature transformation refers to the mathematical transformation of individual features, such as logarithmic transformation, discretization, encoding, etc., to change the scale, distribution, form, etc. of the features. Feature derivation needs to be based on business knowledge and data analysis to find meaningful and effective feature construction methods. Different models may have different effects and applicability for different feature derivation methods. Feature derivation also needs to pay attention to avoid problems such as overfitting, sparsity, and multicollinearity.

[0060] (3) Feature screening

[0061] High-dimensional data may contain irrelevant or redundant features, which can affect model performance and generalization, leading to overfitting or underfitting. Feature screening involves selecting features from the original feature set that are highly correlated with the target variable and less redundant. This improves model performance and interpretability while reducing computational resources and time consumption. Feature screening methods are generally categorized into three types: Filtering: Each feature is scored based on its divergence or correlation, setting a threshold or selecting features with the highest scores. Common scoring methods include the Pearson correlation coefficient, the chi-squared test, and mutual information. Wrapping: Based on the objective function (usually a predictive performance score), a number of features are selected or excluded at a time, iteratively finding the optimal feature subset. Common search methods include greedy algorithms and genetic algorithms. Embedding: Feature screening is performed automatically during model training, determining feature importance based on model parameters or coefficients. Common models include Lasso regression and decision trees.

[0062] S5. Use the eigenvalues ​​and target values ​​obtained in S4 to synthesize new sample data, substitute them into the GBDT regression model, use the eigenvalues ​​as model input, and the target value as model output to obtain a submarine cable breakage quantitative analysis model.

[0063] Furthermore, the principle and implementation steps of the GBDT regression algorithm in step S5 are as follows:

[0064] GBDT is an ensemble learning algorithm based on the concept of gradient boosting. It uses multiple regression decision trees (CARTs) as base learners and iteratively fits the residuals of all previous trees to improve the model's predictive ability. The objective function of GBDT is a differentiable loss function, such as mean squared error (MSE) or cross entropy. GBDT optimizes the objective function using gradient descent, updating the model along the negative gradient of the objective function at each iteration. The implementation steps of GBDT are: 1) Initialize the model to a constant value, such as the mean of the training data or the class ratio. 2) For each iteration m = 1, 2, …, M: Calculate the negative gradient of each example as the residual, which is the negative gradient of the difference between the true value and the current model prediction. Use the residual as the target variable to train a regression decision tree, obtaining the leaf node regions and corresponding output values. The model is updated by adding the trained tree from this round and multiplying it by a learning rate (shrinkage factor). 3) Output is the final model, which is the weighted sum of the M trees.

[0065] Example

[0066] A 3D simulation model of a submarine cable with varying numbers of broken wires was designed and placed in a static magnetic field. The magnetic signals in the X and Y directions were collected at varying lift-off heights. Eight raw eigenvalues ​​(X amplitude, X area, X maximum gradient, X average gradient, Y amplitude, Y maximum gradient, Y average gradient, and lift-off height) were extracted from the X and Y magnetic signals to form a sample. Data analysis, feature derivation, and feature screening were performed on the raw eigenvalues. The final data was then fed into a GBDT regression model to output the number of broken wires in the cable.

[0067] In geomagnetism, a submarine cable can be approximated as an abnormal shape of an infinitely long horizontal cylinder. A magnetic field model of an unpowered submarine cable is established. In addition, there are few ferromagnetic substances in the natural seabed environment, and the relative magnetic permeabilities of seabed sediments and seawater are close to 1. A rectangular domain is established, and the relative magnetic permeabilities within the domain are all set to 1 to simulate the underwater environment for detection. Figure 1 The established submarine cable model is shown in the figure; Figure 2 The figure in the figure represents the cross section of the submarine cable. The internal structure is equivalent to a copper conductor with a diameter of D = 210 mm. The outer layer of the armor is a stranded steel wire with a diameter of d = 12 mm. Figure 4 Represents the environmental sea area around the submarine cable, and the medium in the area is set to seawater.

[0068] The lifting height of the magnetic sensor and the number of broken cables are shown in Table 1:

[0069] Table 1 Lifting height and number of broken wires

[0070]

[0071] Figure 3-4 The magnetic signal curves of sample 1 and sample 2 in the X and Y directions are displayed. The signal collection length above the coastal cable is 4 meters. When the number of broken wires is the same, the lower the lifting height, the more obvious the signal is. When the number of broken wires is the same, the more obvious the signal is, which is in line with the general rules of magnetic signal simulation.

[0072] The maximum gradient and average gradient calculation formulas are as follows (2) (3):

[0073]

[0074]

[0075] Where: i = 2,…,a, a represents the number of points collected in the segment, ΔL represents the length of the specimen between two points, and ΔL is a constant value when the sampling frequency is constant.

[0076] The eight original eigenvalues ​​and the corresponding number of broken wires of the 12 groups of samples are shown in Table 2:

[0077] Table 2 Sample characteristic values ​​and corresponding number of broken wires

[0078]

[0079]

[0080] Substituting the eight original features into the GBDT regression prediction model, the root mean square error of the test set was 3.12, which was considered the initial error. The Pearson correlation coefficient between the original features and the target value (number of broken wires) was calculated using formula (1). The results are shown in Table 3.

[0081] Table 3 Correlation coefficients of original eigenvalues

[0082]

[0083] Sorted by absolute value from largest to smallest, the top four are Y-average gradient, X-average gradient, X-maximum gradient, and Y-maximum gradient. These four original features are multiplied to derive feature values: X-average gradient * Y-average gradient is recorded as derived feature 1 (-0.797712), X-average gradient * X-maximum gradient is recorded as derived feature 2 (0.837653), and X-average gradient * X-maximum gradient * Y-average gradient is recorded as derived feature 3 (-0.824377). Table 4 shows the correlation coefficients of the derived feature values.

[0084] Table 4 Correlation coefficients of derived features 1-3

[0085]

[0086] The absolute value of the correlation coefficient between derived feature 2 and derived feature 3 is above 0.8. Therefore, a new feature matrix is ​​formed by using derived feature 2, derived feature 3 and the Y average gradient in the original feature as the input of the GBDT model. The root mean square error of the test set is 2.44, which is 0.68 lower than the original root mean square error, achieving certain results.

[0087] Continue with feature derivation. Although the lift-off height feature performed poorly in the correlation analysis, actual testing experience shows that the probe lift-off height has a significant impact on the strength of the magnetic signal, so this feature needs to be reused. Since this feature has a small number of values, it can be treated as a discrete variable and one-hot encoded. Furthermore, this discrete variable is used to perform grouped feature statistics on derived eigenvalue 1, derived eigenvalue 2, and the Y-average gradient. Grouped feature statistics group continuous variables according to the values ​​of the discrete variable and calculate the mean, maximum, and minimum values ​​for each group of continuous variables.

[0088] The group statistics results are shown in Table 5:

[0089] Table 5 Derivative results of group statistical characteristics

[0090]

[0091] After joint one-hot encoding, the new feature matrix is ​​shown in Table 6, with a total of 17 features:

[0092] Table 6 Final eigenvalues

[0093]

[0094] Continued table:

[0095]

[0096] After substituting the final eigenvalues ​​obtained in Table 6 into the GBDT model, a quantitative analysis model for broken submarine cables with magnetic signals is obtained.

[0097] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for quantitatively analyzing and obtaining broken submarine cables based on magnetic signals, characterized in that: The following steps are involved: Step 1: placing the submarine cable in a static magnetic field, taking the number of cable wire breaks and the increase in separation as initial variables, collecting magnetic signals at the cable wire break locations, wherein the magnetic signals include X-direction magnetic signals and Y-direction magnetic signals; Step 2: Extracting original eigenvalues ​​from the X-direction magnetic signal and the Y-direction magnetic signal, wherein the original eigenvalues ​​include X amplitude, X area, X maximum gradient, X average gradient, Y amplitude, Y maximum gradient, Y average gradient, and lift-off height, setting the number of fractures as the target value, and establishing an initial model; Step 3, calculate the Pearson correlation coefficient of the original eigenvalue; Step 4: Sort by the size of the Pearson correlation coefficient, select the first four original eigenvalues ​​with the largest absolute value, calculate and obtain several derived eigenvalues ​​based on the first four original eigenvalues, select the first two derived eigenvalues ​​with the largest absolute value as the derived features, and at the same time obtain the original feature with the largest absolute value as the original eigenvalue one; Step 5: Based on the improvement of separation, the maximum, minimum and average values ​​of the derived features and the original feature value in each group are calculated, and the separation height after one-hot encoding is collected to obtain the final feature value; Step 6: After the new eigenvalues ​​are introduced into the regression prediction model, a quantitative analysis model of the magnetic signal submarine cable breakage is obtained.

2. The method for quantitatively analyzing and obtaining broken submarine cables based on magnetic signals according to claim 1, characterized in that: In step 1, when the collected magnetic signal is inconsistent with the existing magnetic signal variation pattern, the static magnetic field or variable is adjusted until the magnetic signal is consistent with the existing magnetic signal variation pattern.

3. The method for quantitatively analyzing and obtaining broken submarine cables based on magnetic signals according to claim 1, characterized in that: In step 4, the formula for calculating the derived eigenvalue is: Derived eigenvalue 1 = original eigenvalue 1 × original eigenvalue 2 Derived eigenvalue 2 = original eigenvalue 2 × original eigenvalue 3 Derived eigenvalue three = original eigenvalue one × original eigenvalue two × original eigenvalue three.

4. The method for quantitatively analyzing and obtaining broken submarine cables based on magnetic signals according to claim 1, characterized in that: In step 3, the original eigenvalue is substituted into the regression prediction model to calculate the initial error; in step 4, the derived feature and the original eigenvalue are input into the regression prediction model as new features to calculate the new error. The initial error and the new error are compared. When the new error is smaller than the initial error, it means that the derived feature selection meets the requirements.

5. The method for quantitatively analyzing and obtaining broken submarine cables based on magnetic signals according to claim 1, characterized in that: In step 3, the Pearson correlation coefficient formula is: Where: n is the number of total samples, X i is the eigenvalue of the i-th sample, is the average value of the eigenvalue, Y i is the target value of the i-th sample, is the average of the target values.

6. The method for quantitatively analyzing and obtaining broken submarine cables based on magnetic signals according to claim 1, characterized in that: In step 4, the original eigenvalue 1 is the Y average gradient.

7. The method for quantitatively analyzing and obtaining broken submarine cables based on magnetic signals according to claim 1, characterized in that: In step 6, the regression prediction model is the GBDT regression prediction model.

8. A method for quantitatively analyzing and obtaining broken submarine cables based on magnetic signals according to any one of claims 1 to 7, characterized in that: In step 1, the submarine cable is approximately regarded as an infinitely long horizontal cylinder with abnormal shape, and the seabed is regarded as a rectangular domain, and the magnetic signal at the break of the submarine cable wire is simulated and collected.

Images