Submarine cable interlayer fault detection method and system for submarine cable grounding system, and medium

Abstract

The invention discloses a submarine cable interlayer fault detection method and system for a submarine cable grounding system, and a medium. The method comprises the following steps: S1, pre-installing a power frequency current sensor, a power frequency voltage sensor and a traveling wave sensor on a sheath layer and an armor layer of landing sections at two sides of the high-voltage alternating-current submarine cable grounding system; the method comprises the following steps: extracting current and voltage signal data of a fault state through a power frequency current sensor, a power frequency voltage sensor and a traveling wave sensor, and carrying out data preprocessing including detection cleaning, data interpolation and data normalization on the current and voltage signal data; according to the method, fault codes are obtained through data collection, preprocessing, feature fusion, training learning and coding reconstruction and classification, and accurate fault detection is carried out on submarine cable interlayer faults through the fault codes.

Description

Technical Field

[0001] The present invention relates to the field of power systems, and in particular to a method, system and medium for detecting faults between submarine cables in a submarine cable grounding system. Background Art

[0002] Long-distance, high-capacity ocean power transmission technology is a key development direction for my country's power system. As the core transmission equipment in ocean power transmission projects, the operational reliability of high-voltage AC submarine cables (submarine cables) is crucial. Submarine cable laying environments are unique, and the cables themselves are susceptible to damage from factors such as seabed topography and external stress, causing short circuits between cable layers. To ensure reliable power supply, fault detection and timely repair are essential.

[0003] However, due to the special laying environment of submarine cables, it is difficult to conduct daily inspections for faults, and it is difficult to repair hidden faults that occur between submarine cable layers. Once a fault occurs, it will cause the offshore power transmission system to collapse. Therefore, it is necessary not only to prevent submarine cable grounding faults or line break faults, but also to detect interlayer faults caused by stress damage between submarine cable layers, so as to achieve rapid diagnosis of submarine cable faults and reduce the outage rate of the ocean power transmission network. Summary of the Invention

[0004] The present invention aims to provide a method, system, and medium for detecting inter-layer faults in a submarine cable grounding system. The present invention obtains fault codes through data collection, preprocessing, feature fusion, training and learning, and reconstruction coding and classification. The fault codes are then used to accurately detect inter-layer faults in submarine cables.

[0005] The technical solution of the present invention is a method for detecting faults between submarine cables in a submarine cable grounding system, which is carried out according to the following steps:

[0006] Step S1: pre-installing power frequency current sensors, power frequency voltage sensors, and traveling wave sensors on the sheath layer and armor layer of the landing section on both sides of the high-voltage AC submarine cable grounding system; extracting current and voltage signal data of the fault state through the power frequency current sensors, power frequency voltage sensors, and traveling wave sensors, and then performing data preprocessing on the current and voltage signal data, including detection cleaning, data interpolation, and data normalization;

[0007] Step S2: dividing the pre-processed current and voltage signal data into a learning data group and a test data group, performing feature fusion on the learning data group to obtain the fault signal feature quantity of the high-voltage AC submarine cable grounding system;

[0008] Step S3: Input the fault feature into the deep belief network for repeated training, combine the test data set to complete feature learning until convergence, reconstruct the trained fault feature, and then classify the reconstructed code according to the fault state to obtain the fault code under various fault states;

[0009] Step S4: Determine the fault between the submarine cables of the high-voltage AC submarine cable grounding system through the fault code.

[0010] In the above-mentioned method for detecting inter-layer faults in a submarine cable grounding system, the fault states include a short circuit fault between the submarine cable core and the sheath, a short circuit fault between the submarine cable sheath and the armor, and a short circuit fault between the submarine cable core and the armor.

[0011] In the above-mentioned method for detecting faults between submarine cables in a submarine cable grounding system, the detection and cleaning method utilizes a K-means clustering algorithm based on Euclidean distance, which divides the data points of the current and voltage signal data into K categories to form a set C = {c i ,i=1,2,3,...,K}, set μi to be the value of each class c i The center of the class, using Euclidean distance as the criterion for judging data similarity, the distance from each class to the class center μ i The calculation formula of the square sum of distance J(C) is as follows:

[0012]

[0013] Where x i represents a d-dimensional vector, k represents a natural number between 1 and K in the set C, Select c k The average value of each data point in the cluster is the cluster center μ i ;dis ki Represents vector x i To the cluster center μ i The Euclidean distance between .

[0014] The sum of squared distances J(C) is iterated until no change occurs. During the calculation process, data with a Euclidean distance greater than the current average Euclidean distance is determined to be abnormal data. After the calculation is completed, the abnormal data is cleared and the detection and cleaning is completed.

[0015] In the aforementioned method for detecting inter-layer faults in a submarine cable grounding system, the data interpolation method interpolates missing values in the current and voltage signal data after detection and cleaning using the least squares regression interpolation method. First, a regression model is used to establish a mathematical relationship between the current and voltage signal data after detection and cleaning, as shown in the following formula:

[0016]

[0017] Y=Xβ+ε1;

[0018] In the formula, Y is the known data; Z is the missing value; X and W are the independent variables of Y and Z respectively; β is the estimated value; ε1 and ε2 are the error terms respectively;

[0019] The least squares method is used to calculate β, and the missing data value is calculated by β, as shown in the following formula:

[0020] Z=Wβ+ε2。

[0021] In the aforementioned method for detecting inter-layer faults in a submarine cable grounding system, data normalization is to convert the current and voltage signal data after data interpolation into the range of [0, 1] to obtain a normalized signal. The calculation formula is as follows:

[0022]

[0023] Where, X norm is the normalized result; X is the original data; X max 、X min are the maximum and minimum values of the original data set, respectively.

[0024] In the above-mentioned method for detecting inter-layer faults in a submarine cable grounding system, the feature fusion is realized by an empirical mode algorithm, which uses the preprocessed learning data set as the initial signal x(t), fits the extreme points of the initial signal x(t), and obtains the maximum envelope e max (t) and the minimum envelope e min (t), and then calculate the maximum envelope e max (t) and the minimum envelope e min (t), and then remove the average value m1(t) as the low-frequency signal:

[0025]

[0026] Each subsequent calculation of the above step For depreciation, the extreme value average value m of the new envelope i (t) is the value to be subtracted and the above operation is repeated until the kth time the first-order eigenmode condition is obtained. Based on the above steps, the intrinsic mode component c1(t) of the initial signal x(t) is obtained, which is calculated as follows:

[0027]

[0028] Subtract the high-frequency signal c1(t) from the original signal x(t) to obtain the new signal r1(t) without the high-frequency signal. The calculation is shown in the following formula:

[0029] r1(t)=x(t)-c1(t);

[0030] Use the above steps to solve the extreme envelope average value q1(t) of the residual component r1(t), and remove the average value q1(t) as a low-frequency signal:

[0031]

[0032] Each subsequent calculation of the above step For the deduction, the extreme value average value q of the new envelope i (t) is the value to be subtracted and the above operation is repeated until the kth time the first-order eigenmode condition is obtained. Based on the above steps, the eigenmode component c2(t) of the residual component r1(t) is obtained, which is calculated as shown in the following formula:

[0033]

[0034] The second eigenmode component c2(t) is obtained from the above, and the calculation is repeated until the nth order eigenmode component c n (t) or its residual component r n (t) is less than the preset value, or when the residual component r n When (t) is a monotonic function or a constant, the empirical mode decomposition process is determined to be finished;

[0035] Finally, the initial signal x(t) is decomposed by empirical mode to obtain the fault signal characteristic, which is expressed as follows:

[0036]

[0037] In the aforementioned method for detecting inter-layer faults in a submarine cable grounding system, the deep belief network includes a visible layer, a hidden layer, and an output layer connected in sequence. The training process is to input the fault feature quantity into the visible layer and train a restricted Boltzmann machine in the visible layer. The calculation of the restricted Boltzmann machine energy function E(v,h) is shown in the following formula:

[0038]

[0039] Where, v i is the neuron state of the visible layer, h j is the neuron state of the hidden layer, a i and b i is the offset, w ij is the connection weight;

[0040] Fix the weights and offsets of the previous restricted Boltzmann machine and use the neuron state of its hidden layer as the input of the next restricted Boltzmann machine for training, where the weight Δw ij The calculation is as follows:

[0041] Δw ij =η( <v i h j > data - <v i h j > recon );

[0042] Where η is the learning rate, <·> data Indicates the expected value of the data, <·> recon Indicates the expected value during the reconstruction process;

[0043] Repeat the training multiple times to complete the signal feature learning in the hidden layer until convergence.

[0044] In the aforementioned method for detecting inter-layer faults in a submarine cable grounding system, when the hidden layer data is smaller than the visible layer data during the deep belief network learning process, the hidden layer reconstructs and encodes the fault feature quantity that has completed training.

[0045] The system of the aforementioned method for detecting inter-layer faults in a submarine cable grounding system comprises a data acquisition and processing module, a deep fusion module, a fault classification module, and a fault judgment module connected in sequence;

[0046] The data acquisition and processing module is used to collect the current and voltage signal data of the fault state extracted by the power frequency current sensor, power frequency voltage sensor and traveling wave sensor, and perform data preprocessing;

[0047] The deep fusion module is used to perform feature fusion on the learning data group to obtain the fault signal feature value;

[0048] The fault classification module is used to integrate the deep belief network, perform feature learning on the fault signal feature quantity and classify it to obtain the fault code;

[0049] The fault judgment module is used to judge the fault between the submarine cables of the high-voltage AC submarine cable grounding system through the fault code.

[0050] A computer-readable storage medium stores computer-executable instructions, which, when executed, implement the method for detecting inter-layer faults in a submarine cable grounding system.

[0051] Compared with the existing technology, the present invention performs data preprocessing including detection and cleaning, data interpolation and data normalization on the current and voltage signal data of the fault state, eliminating the misdiagnosis that may be caused by the differences in the data itself; the processed data is subjected to feature fusion to obtain the fault signal feature quantity, thereby reflecting more fault characteristics and improving accuracy; based on the deep belief network, the fault signal feature quantity of the inter-layer fault of the submarine cable is comprehensively learned and reconstructed into coding classification, accurately reflecting the characteristics of various types of faults, and improving the efficiency and accuracy of submarine cable fault detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 Schematic diagram of the flow of a method for detecting inter-layer faults in a submarine cable grounding system of the present invention;

[0053] Figure 2 This is a schematic diagram of a power transmission circuit of a high-voltage AC submarine cable grounding system in Example 1 of the present invention;

[0054] Figure 3 1 is a schematic diagram of a load circuit of a high-voltage AC submarine cable grounding system in Example 1 of the present invention;

[0055] Figure 4 The specific arrangement position of the sensor in the submarine cable in Example 1 of the present invention;

[0056] Figure 5 Schematic diagram of inter-layer fault types of submarine cables in Example 1 of the present invention;

[0057] Figure 6 The interlayer fault voltage signal and current signal measured on the submarine cable sheath at the landing section at one end of the line in Example 1 of the present invention;

[0058] Figure 7 The interlayer fault voltage signal and current signal measured on the submarine cable armor at the landing section at one end of the line in Example 1 of the present invention;

[0059] Figure 8 This is a schematic diagram of the framework of the submarine cable inter-layer fault detection system of the submarine cable grounding system of the present invention;

[0060] Figure 9 This is a topological diagram of the DBN network training process in Example 1 of the present invention. DETAILED DESCRIPTION

[0061] The present invention will be further described below with reference to the accompanying drawings and examples, but they are not intended to limit the present invention.

[0062] Example 1: A method for detecting faults between submarine cables in a submarine cable grounding system, as shown in the attached Figure 1 To the attached Figure 4 As shown, follow the steps below:

[0063] Step S1: The submarine cable grounding system is a double-end grounded 110KV high-voltage AC submarine cable grounding system. Power frequency current sensors (CT1, CT2, CT3, CT4), power frequency voltage sensors (PT1, PT2, PT3, PT4) and traveling wave sensors (P1, P2, P3, P4) are pre-installed on the submarine cable sheath and armor, and the sensors are integrated into an onshore data acquisition box; the power frequency current sensor is used to extract the current of the submarine cable sheath and armor under the fault state, the power frequency voltage sensor is used to extract the voltage of the submarine cable sheath and armor under the fault state, and the traveling wave sensor is used to capture the traveling wave signal to detect the fault and location on the submarine cable sheath and armor, as shown in the attached figure. Figure 5 As shown, the fault status includes the following Figure 5 (a) The short circuit fault between the submarine cable core and the sheath, Figure 5 (b) The short circuit fault between the submarine cable sheath and armor, Figure 5 (c) The short circuit fault between the submarine cable core and the armor is as shown in the attached Figure 6 and attached Figure 7 As shown, attached Figure 6 (a) The current in the sheath layer suddenly changes after the fault occurs. Figure 6 (b) The voltage of the sheath layer suddenly changes after the fault occurs. Figure 7 (a) The current in the armor layer suddenly changes after the fault occurs. Figure 7 (b) The voltage of the armor layer suddenly changes after the fault occurs; the current and voltage signal data are preprocessed including detection cleaning, data interpolation and data normalization;

[0064] The detection and cleaning is performed using the K-means clustering algorithm based on Euclidean distance. First, the Euclidean distance between each data point and the cluster center is calculated. In multidimensional space, the Euclidean distance can be used to represent the sample x using the real distance between two points. i and x j The similarity between them is calculated using the Euclidean distance formula as follows:

[0065]

[0066] Where x i and x j Represents d-dimensional vectors, dis(x i ,x j ) is smaller, indicating x i and x j The higher the similarity, the smaller the difference; Based on the above Euclidean distance judgment standard and calculation principle, the data points of the current and voltage signal data are divided into K categories to form a set C = {c i ,i=1,2,3,...,K}, set μi to be the value of each class c iThe center of the class, using Euclidean distance as the criterion for judging data similarity, the distance from each class to the class center μ i The calculation formula of the square sum of distance J(C) is as follows:

[0067]

[0068] Then, the total distance square sum J(C) of each category is calculated as follows:

[0069]

[0070] Where x i represents a d-dimensional vector, k represents a natural number between 1 and K in the set C, Select c k The average value of each data point in the cluster is the cluster center μ i ;dis ki Represents vector x i To the cluster center μ i The Euclidean distance between .

[0071] The sum of squared distances J(C) is iterated until no significant changes occur. During the calculation process, data with a Euclidean distance greater than the current average Euclidean distance is considered abnormal data. After the calculation is completed, the abnormal data is cleared and the detection and cleaning is completed.

[0072] The data interpolation is to interpolate the missing data values in the current and voltage signal data after detection and cleaning using the least squares regression interpolation method. First, a regression model is used to establish a mathematical relationship between the current and voltage signal data after detection and cleaning, as shown in the following formula:

[0073]

[0074] The mathematical relationship established for the data based on the above model is as follows:

[0075] Y=Xβ+ε1;

[0076] In the formula, Y is the known data; Z is the missing value; X and W are the independent variables of Y and Z respectively; β is the estimated value; ε1 and ε2 are the error terms respectively;

[0077] The least squares method is used to calculate β, and the missing data value is calculated by β, as shown in the following formula:

[0078] Z=Wβ+ε2。

[0079] The data normalization is to convert the current and voltage signal data after data interpolation into the range of [0,1]. The calculation formula is as follows:

[0080]

[0081] Where, X norm is the normalized result; X is the original data; X max 、X min are the maximum and minimum values of the original data set, respectively.

[0082] Step S2: Divide the preprocessed current and voltage signal data into a learning data group and a test data group (for testing the learning effect of the subsequent network algorithm), perform feature fusion on the learning data group to obtain the fault signal feature quantity (such as time-frequency domain signal) of the high-voltage AC submarine cable grounding system;

[0083] The feature fusion is achieved by the empirical mode algorithm, which takes the preprocessed learning data set as the initial signal x(t), fits the extreme points of the initial signal x(t), and obtains the maximum envelope e max (t) and the minimum envelope e min (t), and then calculate the maximum envelope e max (t) and the minimum envelope e min (t), and then remove the average value m1(t) as the low-frequency signal:

[0084]

[0085] Then remove the low-frequency signal and calculate as follows:

[0086]

[0087] Each subsequent calculation of the above step For depreciation, the extreme value average value m of the new envelope i (t)(i=1,2,3…) is the value to be subtracted and the above operation is repeated until the kth time the first-order eigenmode condition is obtained. Based on the above steps, the intrinsic mode component c1(t) of the initial signal x(t) is obtained, which is calculated as follows:

[0088]

[0089] Subtract the high-frequency signal c1(t) from the original signal x(t) to obtain the new signal r1(t) without the high-frequency signal. The calculation is shown in the following formula:

[0090] r1(t)=x(t)-c1(t);

[0091] Use the above steps to solve the extreme envelope average value q1(t) of the residual component r1(t), and remove the average value q1(t) as a low-frequency signal:

[0092]

[0093] Each subsequent calculation of the above step For the deduction, the extreme value average value q of the new envelope i (t)(i=1,2,3…) is the value to be subtracted and the above operation is repeated until the kth time the first-order eigenmode condition is obtained. Based on the above steps, the eigenmode component c2(t) of the residual component r1(t) is obtained, which is calculated as shown in the following formula:

[0094]

[0095] The second eigenmode component c2(t) is obtained from the above, and the calculation is repeated until the nth order eigenmode component c n (t) or its residual component r n (t) is less than the preset value, or when the residual component r n When (t) is a monotonic function or a constant, the empirical mode decomposition process is determined to be finished;

[0096] Finally, the initial signal x(t) is decomposed by empirical mode to obtain the fault signal characteristic, which is expressed as follows:

[0097]

[0098] Step S3: As shown in the attached Figure 9 As shown, the fault feature quantity is input into the deep belief network for repeated training, and the feature learning is completed in combination with the test data set until convergence;

[0099] The deep belief network (DBN) includes a visible layer, a hidden layer, and an output layer connected in sequence. The component element of the deep belief network is a restricted Boltzmann machine (RBM), wherein each RBM acts as a separate classifier. The input training data (i.e., the fault feature quantity) enters the visible layer of the RBM, and the feature extraction of the training data occurs in the hidden layer of the RBM. The visible layer has a total of m nodes, and the hidden layer has n nodes. The connection weight matrix, the entire deep belief network is composed of multiple restricted Boltzmann machines stacked together. After inputting the data, it is trained in the first RBM and then input into the second RBM. The initial weight is set to According to the initialization weights and the input training data, the feature matrix H = (h1 h2 h3…h l-1 h l ), the data is pre-trained and back-propagated layer by layer in the deep belief network, and finally feature learning is completed in the hidden layer until the model converges; the training process specifically inputs the fault feature quantity into the visible layer and trains the restricted Boltzmann machine in the visible layer. The calculation of the restricted Boltzmann machine energy function E(v,h) is shown in the following formula:

[0100]

[0101] Where, v i is the neuron state of the visible layer, h j is the neuron state of the hidden layer, a i and b i is the offset, w ij is the connection weight;

[0102] Fix the weights and offsets of the previous restricted Boltzmann machine and use the neuron state of its hidden layer as the input of the next restricted Boltzmann machine for training, where the weight Δw ij The calculation is as follows:

[0103] Δw ij =η( <v i h j > data - <v i h j > recon );

[0104] Where η is the learning rate, <·> data Indicates the expected value of the data, <·> recon Indicates the expected value during the reconstruction process;

[0105] Repeat the training multiple times to complete the signal feature learning in the hidden layer until convergence.

[0106] During the deep belief network learning process, when the hidden layer data is smaller than the visible layer data, the hidden layer automatically reconstructs and encodes the fault feature quantity that has completed training (after several training sessions, the hidden layer will more accurately display the features of the visible layer and restore the visible layer at the same time). The reconstructed code is classified according to the fault state to obtain the fault code under various fault states. The following algorithm example explains the reconstruction code:

[0107] For each record t in the learning data set, there is:

[0108] Assign t to the visible layer v (0) , calculate the probability P of activating the hidden layer neurons:

[0109]

[0110] Where, subscript j represents dimension, z represents hidden layer neurons, and W represents weight;

[0111] Then, draw a sample from the calculated probability distribution:

[0112] z (0) ~P(z(0) |v (0) )

[0113] Then, use z (0) Reconstruct the visible layer:

[0114] P(v i (1) =1|z (0) )=σ(W i T z (0) )

[0115] Similarly, take a sample of the visible layer:

[0116] v (1) ~P(v (1) |z (0) )

[0117] The reconstructed visible layer neurons are used again to calculate the hidden layer activation probability:

[0118]

[0119] Repeat the above operation until the amount of hidden layer data is less than the amount of visible layer data, and the training is completed. After the training, the RBM can obtain the accurate features of the visible layer in the reconstruction process, and the encoding is obtained after the above process. The data passing through the hidden layer removes redundant information and restores all the features of the initial sample to the greatest extent.

[0120] Step S4: Determine the fault between the submarine cables of the high-voltage AC submarine cable grounding system through the fault code.

[0121] In summary, the present invention performs data preprocessing including detection cleaning, data interpolation and data normalization on the fault signals detected by the high-voltage AC submarine cable grounding system; when the fault signal is collected, the data judged as abnormal is first cleaned to obtain net data, and the net data may have individual data missing values, so the missing values need to be interpolated for the net data. Since each fault signal is different, the output supplemented data needs to be normalized, and the fluctuation of the fault data finally obtained occurs within the specified range. This data preprocessing prevents the conditional restrictions or poor data quality in the fault data collection process from affecting the detection accuracy; the preprocessed current and voltage signal data are deeply fused to obtain multiple feature quantities of the signal, which can reflect more fault characteristics and improve accuracy; the entire analysis and learning process is based on a deep belief network, which makes full use of the characteristics of machine learning to comprehensively learn and classify the fault feature quantities of submarine cable interlayer faults, more accurately reflecting the characteristics of various types of faults, and the entire method is more detailed in reflecting the fault types for submarine cable interlayer faults compared with traditional fault detection methods, thereby improving the efficiency and accuracy of submarine cable fault detection.

[0122] Example 2: A submarine cable grounding system submarine cable interlayer fault detection system, used for a submarine cable grounding system submarine cable interlayer fault detection method, as shown in the attached Figure 8 As shown, it includes a data acquisition and processing module, a deep fusion module, a fault classification module and a fault judgment module connected in sequence;

[0123] The data acquisition and processing module is used to collect current and voltage signal data of fault states extracted by the power frequency current sensor, the power frequency voltage sensor and the traveling wave sensor, and perform data preprocessing; the data acquisition and processing module is an acquisition device, a signal conditioning circuit and a controller. The acquisition device includes but is not limited to a PLC and a single-chip microcomputer, which converts the acquired signal into a digital signal that can be processed by the microcontroller. The signal conditioning circuit amplifies (gain adjustment), filters (noise removal), isolates (anti-interference), linearizes (compensates for nonlinear errors), and performs analog-to-digital conversion (ADC) on the acquired signal. The controller can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), off-the-shelf programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc.

[0124] The deep fusion module is an embedded processing system based on digital signal processing, or an embedded processing system based on a field programmable gate array, which integrates an empirical mode algorithm for performing feature fusion on the learning data group to obtain the fault signal feature quantity of the high-voltage AC submarine cable grounding system;

[0125] The fault classification module is an embedded processing system based on digital signal processing, or an embedded processing system based on a field programmable gate array, which integrates a deep belief network to perform feature learning on fault signal features, reconstruct and encode the trained fault features, and then classify the reconstructed codes according to the fault state to obtain fault codes under various fault states;

[0126] The fault judgment module is an embedded processing system based on digital signal processing, and can also be an embedded processing system based on a field programmable gate array. It matches the fault code with the corresponding fault current and voltage signals, and judges the inter-cable fault of the high-voltage AC submarine cable grounding system through the fault code.

[0127] Embodiment 3: A computer-readable storage medium stores computer-executable instructions, which, when executed, implement a method for detecting inter-layer faults in a submarine cable grounding system.

[0128] In this embodiment, it should be understood that the disclosed system can be implemented in other ways. For example, the system embodiment described above is merely illustrative. For example, the module division is merely a logical functional division. In actual implementation, other division methods may be used. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not implemented. In addition, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interface, device or module, and may be electrical or other forms.

[0129] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules may be selected to achieve the purpose of the present embodiment according to actual needs.

[0130] In addition, the functional modules in various embodiments of the present invention may be integrated into a single processing module, or each module may exist physically separately, or two or more modules may be integrated into a single module. The aforementioned integrated modules may be implemented in the form of hardware or software functional modules.

[0131] If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk, etc. Various media that can store program codes.

Claims

1. A method for detecting faults between submarine cables in a submarine cable grounding system, characterized by: Follow these steps: Step S1: pre-installing power frequency current sensors, power frequency voltage sensors, and traveling wave sensors on the sheath layer and armor layer of the landing section on both sides of the high-voltage AC submarine cable grounding system; extracting current and voltage signal data of the fault state through the power frequency current sensors, power frequency voltage sensors, and traveling wave sensors, and then performing data preprocessing on the current and voltage signal data, including detection cleaning, data interpolation, and data normalization; Step S2: dividing the pre-processed current and voltage signal data into a learning data group and a test data group, performing feature fusion on the learning data group to obtain the fault signal feature quantity of the high-voltage AC submarine cable grounding system; Step S3: Input the fault feature into the deep belief network for repeated training, combine the test data set to complete feature learning until convergence, reconstruct the trained fault feature, and then classify the reconstructed code according to the fault state to obtain the fault code under various fault states; Step S4: Determine the fault between the submarine cables of the high-voltage AC submarine cable grounding system through the fault code.

2. The method for detecting inter-layer faults in a submarine cable grounding system according to claim 1, wherein: The fault states include a short circuit fault between the submarine cable core and the sheath, a short circuit fault between the submarine cable sheath and the armor, and a short circuit fault between the submarine cable core and the armor.

3. The method for detecting inter-layer faults in a submarine cable grounding system according to claim 1, wherein: The detection and cleaning method utilizes the K-means clustering algorithm based on Euclidean distance, which divides the data points of the current and voltage signal data into K categories, forming a set C = {c i ,i=1,2,3,...,K}, set μi to be the value of each class c i The center of the class, using Euclidean distance as the criterion for judging data similarity, the distance from the data point of each class to the class center μ i The calculation formula of the square sum of distance J(C) is as follows: Where x i represents a d-dimensional vector, k represents a natural number between 1 and K in the set C, Select c k The average value of each data point in the cluster is the cluster center μ i ;dis ki Represents vector x i To the cluster center μ i The Euclidean distance between . The sum of squared distances J(C) is iterated until no change occurs. During the calculation process, data with a Euclidean distance greater than the current average Euclidean distance is determined to be abnormal data. After the calculation is completed, the abnormal data is cleared and the detection and cleaning is completed.

4. The method for detecting inter-layer faults in a submarine cable grounding system according to claim 3, characterized in that: The data interpolation is to interpolate the missing data values in the current and voltage signal data after detection and cleaning using the least squares regression interpolation method. First, a regression model is used to establish a mathematical relationship between the current and voltage signal data after detection and cleaning, as shown in the following formula: Y=Xβ+ε1; In the formula, Y is the known data; Z is the missing value; X and W are the independent variables of Y and Z respectively; β is the estimated value; ε1 and ε2 are the error terms respectively; The least squares method is used to calculate β, and the missing data value is calculated by β, as shown in the following formula: Z=Wβ+ε2。 5. The method for detecting inter-layer faults in a submarine cable grounding system according to claim 4, characterized in that: The data normalization is to convert the current and voltage signal data after data interpolation into the range of [0, 1] to obtain a normalized signal. The calculation formula is as follows: Where, X norm is the normalized result; X is the original data; X max 、X min are the maximum and minimum values of the original data set, respectively.

6. The method for detecting inter-layer faults in a submarine cable grounding system according to claim 1, characterized in that: The feature fusion is achieved through the empirical mode algorithm, which takes the preprocessed learning data set as the initial signal x(t), fits the extreme points of the initial signal x(t), and obtains the maximum envelope e max (t) and the minimum envelope e min (t), and then calculate the maximum envelope e max (t) and the minimum envelope e min (t), and then remove the average value m1(t) as the low-frequency signal: Each subsequent calculation of the above step For depreciation, the extreme value average value m of the new envelope i (t) is the value to be subtracted and the above operation is repeated until the kth time the first-order eigenmode condition is obtained. Based on the above steps, the intrinsic mode component c1(t) of the initial signal x(t) is obtained, which is calculated as follows: Subtract the high-frequency signal c1(t) from the original signal x(t) to obtain the new signal r1(t) without the high-frequency signal. The calculation is shown in the following formula: r1(t)=x(t)-c1(t); Use the above steps to solve the extreme envelope average value q1(t) of the residual component r1(t), and remove the average value q1(t) as a low-frequency signal: Each subsequent calculation of the above step For the deduction, the extreme value average value q of the new envelope i (t) is the value to be subtracted and the above operation is repeated until the kth time the first-order eigenmode condition is obtained. Based on the above steps, the eigenmode component c2(t) of the residual component r1(t) is obtained, which is calculated as follows: The second eigenmode component c2(t) is obtained from the above, and the calculation is repeated until the nth order eigenmode component c n (t) or its residual component r n (t) is less than the preset value, or when the residual component r n When (t) is a monotonic function or a constant, the empirical mode decomposition process is determined to be complete; Finally, the initial signal x(t) is decomposed by empirical mode to obtain the fault signal characteristic, which is expressed as follows:

7. The method for detecting inter-layer faults in a submarine cable grounding system according to claim 1, wherein: The deep belief network includes a visible layer, a hidden layer, and an output layer connected in sequence. The training process is to input the fault feature into the visible layer and train the restricted Boltzmann machine in the visible layer. The calculation of the restricted Boltzmann machine energy function E(v,h) is shown in the following formula: Where, v i is the neuron state of the visible layer, h j is the neuron state of the hidden layer, a i and b i is the offset, w ij is the connection weight; Fix the weights and offsets of the previous restricted Boltzmann machine and use the neuron state of its hidden layer as the input of the next restricted Boltzmann machine for training, where the weight Δw ij The calculation is as follows: Δw ij =η( <v i h j > data - <v i h j > recon ); Where η is the learning rate, <·> data Indicates the expected value of the data, <·> recon Indicates the expected value during the reconstruction process; Repeat the training multiple times to complete the signal feature learning in the hidden layer until convergence.

8. The method for detecting inter-layer faults in a submarine cable grounding system according to claim 7, characterized in that: During the deep belief network learning process, when the hidden layer data is smaller than the visible layer data, the hidden layer reconstructs and encodes the fault feature quantity that has completed training.

9. The method for detecting inter-layer faults in a submarine cable grounding system according to any one of claims 1 to 8, characterized in that: The system includes a data acquisition and processing module, a deep fusion module, a fault classification module and a fault judgment module connected in sequence; The data acquisition and processing module is used to collect the current and voltage signal data of the fault state extracted by the power frequency current sensor, power frequency voltage sensor and traveling wave sensor, and perform data preprocessing; The deep fusion module is used to perform feature fusion on the learning data group to obtain the fault signal feature value; The fault classification module is used to integrate the deep belief network, perform feature learning on the fault signal feature quantity and classify it to obtain the fault code; The fault judgment module is used to judge the fault between the submarine cables of the high-voltage AC submarine cable grounding system through the fault code.

10. A computer-readable storage medium storing computer-executable instructions, wherein when the computer-executable instructions are executed, the method for detecting inter-layer faults in a submarine cable grounding system according to any one of claims 1 to 8 is implemented.

Images