Cable fault identification method and system based on convolutional neural network

Abstract

The invention relates to the technical field of cable defect identification, in particular to a cable fault identification method based on a convolutional neural network. The method comprises the following specific steps: establishing a one-dimensional convolutional neural network model based on a transmission line model to predict a cable fault type and a fault point location; building a test platform to collect data of cables of different models and with a certain length, wherein the data is used for building a data set for model training; training the prediction model by using the data set, and iterating for multiple times until the target accuracy is met; reflected wave data of a to-be-detected cable are collected and input into the trained prediction model, and a fault type probability vector and the distance between the fault position and the initial end of the cable are output through the prediction model. The invention provides a one-dimensional convolutional neural network prediction model fusing multi-scale feature extraction and a time sequence attention mechanism, the model has good effects on cable fault identification and fault position prediction, and a cable fault identification model for different types and different defect positions is realized.

Description

Technical Field

[0001] This invention relates to the field of cable defect identification technology, and in particular to a cable fault identification method based on convolutional neural networks. Background Technology

[0002] With the acceleration of urbanization and the continuous advancement of infrastructure construction, the laying of underground cables is becoming increasingly widespread. However, due to the special nature of their installation locations and methods, daily maintenance and repair are quite challenging. Locating the cable and its defects are two major problems in these processes. For cable location, there are methods such as dual-power comparison, electromagnetic radiation, infrared scanning, and high-voltage harmonic methods. Among these, the non-contact cable location method based on the principle of electromagnetic radiation has been achieved and has yielded good results. Domestic companies such as Jingmingshu and Fuyi have also developed corresponding cable locators. However, simply being able to locate the target cable is insufficient to complete the maintenance task. To solve the problem of cable fault location, some researchers have proposed a single-end detection method, which involves applying a signal to one end of the cable and using the returned signal to determine the type and location of the cable defect. In existing technologies, Jia Shih-hao, in his research on fault location and diagnosis methods for power cables [D], Liaoning University of Engineering and Technology, 2023, proposed using a single-ended time-domain reflectometer to calculate the cable terminal distance and simple fault types (open circuit, short circuit) based on the different reflection waveforms under different cable conditions and the propagation speed and reflection time of the electrical signal in the cable. However, in actual use, there are various cable models, and the location of cable faults is often somewhere between the beginning and the end of the cable. At the same time, the types of cable faults are very complex, especially incomplete core breakage and core compression deformation. If these problems can be solved, the time and cost of cable maintenance and repair will be greatly reduced.

[0003] In recent years, artificial intelligence has developed rapidly, with various improved machine learning and deep learning algorithms emerging one after another. Convolutional neural networks (CNNs), as one of the most representative models in the field of deep learning, offer a new technical approach to solving complex cable fault detection problems due to their powerful feature extraction capabilities and spatial modeling advantages. Unlike traditional threshold criterion methods that rely on human experience, CNNs, through an end-to-end training mechanism, can automatically extract deep spatiotemporal correlation features from raw reflection waveforms. To address the aforementioned technical challenges, this application aims to build a one-dimensional CNN that integrates multi-scale feature extraction and temporal attention mechanisms to achieve a cable fault identification model for different types and locations of defects. Summary of the Invention

[0004] The purpose of this invention is to address the problems existing in the background technology by proposing a cable fault identification method based on convolutional neural networks.

[0005] The technical solution of the present invention, in its first aspect, proposes a cable fault identification method based on a convolutional neural network, comprising the following specific steps:

[0006] S1. Based on the transmission line model, a one-dimensional convolutional neural network model is established to predict cable fault types and fault locations;

[0007] S2. Build a test platform to collect data on cables of different models of a certain length to build a dataset for model training;

[0008] S3. Train the prediction model using the dataset, iterating multiple times until the target accuracy is met.

[0009] S4. Collect the reflected wave data of the cable to be tested and input it into the trained prediction model. The prediction model outputs the fault type probability vector and the distance between the fault location and the beginning of the cable.

[0010] Preferably, in step S1, the one-dimensional convolutional neural network model integrates multi-scale feature extraction and temporal attention mechanisms.

[0011] Preferably, the cable type in step S2 includes 500 meters each of RVV2, BVV4 and BV6 type cables;

[0012] The cables used are all multi-core wires, including three-color main core wires and their respective branch core wires.

[0013] Preferably, in step S2, the cables of different models are simulated in real-world scenarios of incomplete breakage, complete breakage, or short circuit.

[0014] In this case, the incomplete breakage was simulated by cutting each colored core wire in half.

[0015] Simulate a complete break by completely severing the entire cable;

[0016] A short circuit is simulated by shorting the cable core and the shielding layer.

[0017] Preferably, in step S3, the number of iterations and the model convergence constraints are set; when the predetermined number of iterations is reached, the model accuracy data is checked to see if the target convergence constraints are met. If the target is met, training is stopped; if not, the iterations are extended for several more iterations until the target convergence constraints are met.

[0018] Preferably, step S4 further includes the following steps:

[0019] S41, Input reflected wave data

[0020] S42. Regarding the original reflected waveform data x rawAfter normalization, the preprocessed input waveform data is obtained, as shown in the following formula:

[0021]

[0022] In the above formula x raw The mean and standard deviation;

[0023] S43. For the preprocessed waveform data The formula for spatiotemporal feature extraction is as follows:

[0024]

[0025] h = h + M i (x),

[0026] in, Each feature block consists of four sub-blocks, each containing a multi-scale convolution M. i (x) Temporal attention mechanism Connect h with the residual. Represents network connectivity, ReLU is the ramp activation function, k is the kernel width (corresponding to different spatial scales), and w i,k ,α t These are all network parameters, which are determined during subsequent network training. Multi-scale features are fused through channel splicing to enhance the model's ability to jointly model near-end fast pulses (small scale) and far-end decaying oscillations (large scale). The h obtained in this step will be used for two branches: cable fault type prediction and fault distance prediction.

[0027] S44. For cable fault type prediction; global feature compression is performed using a pooling layer, with the following formula:

[0028]

[0029] In the above formula, h[:,t] represents the t-th column of h; this formula can compress temporal features into a global descriptor while preserving channel dimension information; fully connected layers are used for fully connected mapping, and the formula is as follows:

[0030]

[0031] z1 and z2 are the class scores for open and short circuits, respectively, and are linear combinations of global features; finally, the Softmax function is used for probability normalization.

[0032]

[0033] Where p1 and p2 are the probabilities of open circuit and short circuit, respectively; the one with the highest probability is taken as the final cable state type;

[0034] S45. For fault distance prediction; after extracting spatiotemporal features, temporal feature enhancement is performed through convolutional layers, with the following formula:

[0035]

[0036] C reg :Conv1D(k=3,C out =128),

[0037]

[0038] Among them, C reg It is a one-dimensional convolutional layer with a kernel width k = 3 and h pool This is a global pooling layer; then it is mapped through a fully connected layer, with the following formula:

[0039]

[0040] in These are all parameters of the fully connected layer, which will be selected during subsequent training. Finally, the predicted distance d can be obtained through decoding. pred The decoding formula is:

[0041]

[0042] Where v p =2×10 8 m / s is the speed of light, and Tmax is the maximum time window, determined by the sampling rate and the number of points.

[0043] S46. Output the prediction results.

[0044] Preferably, in step S4, the reflected wave data is input along with the cable model data and the trained prediction model is input. The prediction model outputs the fault type probability vector and the distance between the fault location and the cable start end.

[0045] A second aspect of the present invention proposes a cable fault identification system based on a convolutional neural network, which uses the above-described method for fault identification and includes a data acquisition module, a data processing module, a calculation module, and an interactive output module.

[0046] The data acquisition module is connected to the beginning of the cable under test to acquire reflected wave data or model data of the cable.

[0047] The data processing module preprocesses the collected data and then outputs it to the calculation module;

[0048] The calculation module performs calculations based on the trained prediction model, outputs the prediction results, and outputs them through an interactive output module.

[0049] The interactive output module outputs the predicted fault type and specific information about the fault location in the form of text, charts, or voice broadcast.

[0050] A third aspect of the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described cable fault identification method based on a convolutional neural network.

[0051] A fourth aspect of the present invention provides a computer-readable medium carrying one or more programs, which, when executed by an electronic device, cause the electronic device to implement the steps of the above-described cable fault identification method based on a convolutional neural network.

[0052] Compared with the prior art, the present invention has the following beneficial technical effects:

[0053] This invention addresses the challenge of locating defects in underground cables by proposing a one-dimensional convolutional neural network prediction model that integrates multi-scale feature extraction and temporal attention mechanisms. This model demonstrates good performance in identifying cable faults and predicting their locations. On a self-made dataset, the accuracy rate for fault categories on the validation set reaches approximately 90%, with an average fault distance error of approximately 0.8 m and a maximum error of 1.6 m. It also exhibits robustness to cable materials. This invention realizes a cable fault identification model for different types and locations of defects. Attached Figure Description

[0054] Figure 1 This is a schematic diagram of the transmission line model in an embodiment of the present invention;

[0055] Figure 2 This is a schematic diagram of the voltage reflection waveform under open-circuit and short-circuit conditions in an embodiment of the present invention;

[0056] Figure 3 This is a flowchart of the model prediction process in an embodiment of the present invention;

[0057] Figure 4 This is a test result diagram of model training in an embodiment of the present invention;

[0058] Figure 5 This is a flowchart of the cable fault identification method in an embodiment of the present invention. Detailed Implementation

[0059] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0060] Transmission line theory is a core tool for analyzing the propagation of electrical signals in cables, and its research is based on the transmission line model. The transmission line model consists of discrete resistance, inductance, capacitance, and conductance (e.g., ...). Figure 1 (As shown). Each small segment of the transmission line, dz, can be considered as consisting of a series resistor R′, a series inductance L′, a parallel conductance G′, and a parallel capacitance C′. Among them, R′, L′, and G′ are all related to the frequency of the signal source. For example, the values ​​of R′ and L′ will change due to the skin effect and proximity effect, while the value of G′ will change due to frequency-dependent dielectric loss.

[0061] The characteristic impedance Z0 of a transmission line can be expressed by the formula

[0062]

[0063] The description is as follows: where ω is the angular frequency, R′ is the series resistance per unit length, L′ is the series inductance per unit length, G′ is the parallel inductance per unit length, and C′ is the parallel capacitance per unit length. For an ideal lossless transmission line, R′ and G′ are zero, and the characteristic impedance simplifies to:

[0064]

[0065] In a lossless transmission line, assume the characteristic impedance is Z0 and the load impedance is Z. L According to transmission line theory, the formula for calculating the voltage reflection coefficient Γ0 is:

[0066]

[0067] When the terminal is open-circuited, Z L As we approach infinity, we have:

[0068]

[0069] The reflected voltage wave is in phase with the incident voltage wave. Similarly, when the terminal is short-circuited (Z... L →0), Γ0=-1, the reflected voltage wave is out of phase with the incident voltage wave; when the terminal impedance is matched (Z L =Z0), Γ0=0, the reflected voltage wave will not be reflected.

[0070] like Figure 2 As shown, Figure 2 The following is a simple example: Point A is the cable start point, point B is the bifurcation point, point C is the open-circuit terminal, and point D is the short-circuit terminal. A narrow pulse voltage signal is applied to the cable start point, and the voltage at that point is measured. Because the signal emitted by the excitation source is an extremely narrow pulse signal, and the cable core itself has resistance, the signal attenuates with each reflection, ultimately forming... Figure 2 The waveform shown.

[0071] Based on transmission line theory, the voltage reflection wave carries information related to the cable's condition. The cable's fault state, core material, and length are all integrated into the voltage reflection wave, thus providing a theoretical basis for constructing cable fault detection using convolutional neural networks in the following paper.

[0072] Example 1

[0073] A cable fault identification method based on convolutional neural networks includes the following specific steps:

[0074] S1. A one-dimensional convolutional neural network model is established based on the transmission line model to predict cable fault types and fault locations; the one-dimensional convolutional neural network model integrates multi-scale feature extraction and temporal attention mechanism;

[0075] To address the spatiotemporal characteristics of cable reflection waveforms, this embodiment constructs a one-dimensional convolutional neural network that integrates multi-scale feature extraction and temporal attention mechanisms. The input to this model is the reflected wave data. The output is a probability vector of the fault type. (Open circuit, short circuit) and the distance l between the fault location and the cable start point pred (Distance from the fault to the starting point along the cable path). The model's prediction flowchart is as follows: Figure 3 As shown, the entire prediction is divided into two branches: fault defect type prediction and fault distance prediction. Data preprocessing and spatiotemporal feature extraction are shared layers.

[0076] To eliminate the characteristic offset caused by cable model differences, the original reflected waveform data x raw After normalization, the preprocessed input waveform data is obtained, as shown in the following formula:

[0077]

[0078] In the above formula x raw The mean and standard deviation;

[0079] For preprocessed waveform data The formula for spatiotemporal feature extraction is as follows:

[0080]

[0081] h = h + M i (x),

[0082] in, Each feature block consists of four sub-blocks, each containing a multi-scale convolution M. i (x) Temporal attention mechanism Connect h with the residual. Represents network connectivity, ReLU is the ramp activation function, k is the kernel width (corresponding to different spatial scales), and w i,k ,α t These are all network parameters, which are determined during subsequent network training. Multi-scale features are fused through channel splicing to enhance the model's ability to jointly model near-end fast pulses (small scale) and far-end decaying oscillations (large scale). The h obtained in this step will be used for two branches: cable fault type prediction and fault distance prediction.

[0083] For cable fault type prediction, global feature compression is performed using a pooling layer, with the following formula:

[0084]

[0085] In the above formula, h[:,t] represents the t-th column of h; this formula can compress temporal features into a global descriptor while preserving channel dimension information; fully connected layers are used for fully connected mapping, and the formula is as follows:

[0086]

[0087] z1 and z2 are the class scores for open and short circuits, respectively, and are linear combinations of global features; finally, the Softmax function is used for probability normalization.

[0088]

[0089] Where p1 and p2 are the probabilities of open circuit and short circuit, respectively; the one with the highest probability is taken as the final cable state type;

[0090] For fault distance prediction, after spatiotemporal feature extraction, temporal feature enhancement is performed through convolutional layers, as shown in the formula:

[0091]

[0092] C reg :Conv1D(k=3,C out =128),

[0093]

[0094] Among them, C reg It is a one-dimensional convolutional layer with a kernel width k = 3 and h pool This is a global pooling layer; then it is mapped through a fully connected layer, with the following formula:

[0095]

[0096] in These are all parameters of the fully connected layer, which will be selected during subsequent training. Finally, the predicted distance d can be obtained through decoding.pred The decoding formula is:

[0097]

[0098] Where v p =2×10 8 m / s is the speed of light, and Tmax is the maximum time window, determined by the sampling rate and the number of points.

[0099] To jointly optimize the cable condition classification and fault location prediction tasks, a hybrid loss function is defined as follows:

[0100]

[0101] in For classification loss, cross-entropy loss is used. For regression loss, Huber loss is used, where d is the ground truth fault distance, dpred is the network predicted value, λ = 0.7 is the task weight coefficient, and y c It is a unique thermal label representing the actual fault state of the cable, with δ = 0.1 being the Huber loss threshold.

[0102] S2. Set up a test platform to collect data on cables of different types and lengths for model training; the cable types include 500 meters each of RVV2, BVV4, and BV6 cables.

[0103] The cables used are all multi-core wires, including three-color main core wires and their respective branch core wires.

[0104] Simulate incomplete breakage, complete breakage, or short circuit scenarios for different cable models in real-world conditions.

[0105] In this case, the incomplete breakage was simulated by cutting each colored core wire in half.

[0106] Simulate a complete break by completely severing the entire cable;

[0107] A short circuit is simulated by shorting the cable core and the shielding layer.

[0108] The method for constructing a dataset includes the following steps:

[0109] Set up an experimental platform to collect data:

[0110] Simulate a fault point near the end of the cable under test; capture waveform data using an oscilloscope and save the raw data to a USB flash drive;

[0111] Noise disturbances are added to the raw waveforms acquired each time in real time;

[0112] Waveform transformation is performed based on the current cable data to obtain simulated waveform data at different fault locations;

[0113] The cable containing the faulty section is completely cut off to obtain a new length of cable test material without intermediate faults.

[0114] Simulation data is generated by preserving the physical parameters of the resected segment;

[0115] Repeat the aforementioned steps to collect sufficient real test data until the cable length no longer meets the test requirements;

[0116] Verify the physical consistency of all generated data and discard simulated waveform data that do not meet the constraints;

[0117] The generated simulated waveform data is mixed with the original waveform data according to a preset ratio to construct a test dataset.

[0118] In this embodiment, an experimental platform is built to construct a dataset. A narrow pulse signal of approximately 90V with a rise time of about 350ps is sent to the core of the cable under test via a transmitter. The reflected wave is then captured by an oscilloscope, and the data is stored on a USB flash drive. First, the cable is laid out to avoid tangling. It is then connected on the experimental platform, and the oscilloscope is switched to threshold-triggered capture mode. The transmitter is then turned on to send a signal to the cable under test. Finally, the waveform data captured by the oscilloscope is saved to the USB flash drive. Each experiment repeats the capture of waveform data 30 times as sample data under this type of condition.

[0119] S3. Train the prediction model using the dataset, iterating multiple times until the target accuracy is met. In step S3, set the number of iterations and the model convergence constraints. When the predetermined number of iterations is reached, check the model accuracy data to see if the target convergence constraints are met. If the target is met, stop training. If not, extend the iterations for several more times until the target convergence constraints are met.

[0120] Based on the network structure constructed above, some hyperparameters of the model also need to be configured, as shown in Table 1. For data partitioning, a configuration of training set:validation set:test set = 7:1.5:1.5 is chosen. Furthermore, this experimental model uses PyTorch as the training platform and is accelerated using a GPU.

[0121] Table 1 Model Hyperparameter Configuration Table

[0122]

[0123] After multiple training iterations, the training loss remained almost constant after approximately 800 iterations. However, a sudden drop in model performance might occur between 600 and 800 iterations. This could be because the model is adapting to defect detection for a particular cable type (mainly BVV4 cable), causing deterioration in results for the other two cable types. To be on the safe side, this embodiment selects 1000 iterations. Training results show that with this number of iterations, model adaptation issues can be largely avoided while maintaining training time. One training result is shown below. Figure 4 As shown, the training loss decreases as the number of iterations increases. For fault classification, the training set and test set eventually stabilize at around 93% and 90%, respectively. For fault distance prediction, the MAE eventually stabilizes at around 0.8m, with a maximum prediction error of 1.6m.

[0124] S4. Collect the reflected wave data of the cable to be tested and input it into the trained prediction model. The prediction model outputs the fault type probability vector and the distance between the fault location and the beginning of the cable.

[0125] Step S4 also includes the following steps:

[0126] S41, Input reflected wave data

[0127] S42. Regarding the original reflected waveform data x raw After normalization, the preprocessed input waveform data is obtained, as shown in the following formula:

[0128]

[0129] In the above formula x raw The mean and standard deviation;

[0130] S43. For the preprocessed waveform data The formula for spatiotemporal feature extraction is as follows:

[0131]

[0132] h = h + M i (x),

[0133] in, Each feature block consists of four sub-blocks, each containing a multi-scale convolution M. i (x) Temporal attention mechanism Connect h with the residual. Represents network connectivity, ReLU is the ramp activation function, k is the kernel width (corresponding to different spatial scales), and w i,k ,α tThese are all network parameters, which are determined during subsequent network training. Multi-scale features are fused through channel splicing to enhance the model's ability to jointly model near-end fast pulses (small scale) and far-end decaying oscillations (large scale). The h obtained in this step will be used for two branches: cable fault type prediction and fault distance prediction.

[0134] S44. For cable fault type prediction; global feature compression is performed using a pooling layer, with the following formula:

[0135]

[0136] In the above formula, h[:,t] represents the t-th column of h; this formula can compress temporal features into a global descriptor while preserving channel dimension information; fully connected layers are used for fully connected mapping, and the formula is as follows:

[0137]

[0138] z1 and z2 are the class scores for open and short circuits, respectively, and are linear combinations of global features; finally, the Softmax function is used for probability normalization.

[0139]

[0140] Where p1 and p2 are the probabilities of open circuit and short circuit, respectively; the one with the highest probability is taken as the final cable state type;

[0141] S45. For fault distance prediction; after extracting spatiotemporal features, temporal feature enhancement is performed through convolutional layers, with the following formula:

[0142]

[0143] C reg :Conv1D(k=3,C out =128),

[0144]

[0145] Among them, C reg It is a one-dimensional convolutional layer with a kernel width k = 3 and h pool This is a global pooling layer; then it is mapped through a fully connected layer, with the following formula:

[0146]

[0147] in These are all parameters of the fully connected layer, which will be selected during subsequent training. Finally, the predicted distance d can be obtained through decoding. pred The decoding formula is:

[0148]

[0149] Where v p =2×10 8 m / s is the speed of light, and Tmax is the maximum time window, determined by the sampling rate and the number of points.

[0150] S46. Output the prediction results.

[0151] In an optional embodiment, in step S4, the reflected wave data and cable model data are simultaneously input into the trained prediction model. The prediction model then outputs a fault type probability vector and the distance between the fault location and the cable's starting end. If the cable model information can be determined during actual use, inputting it into the prediction model can improve the model's recognition accuracy.

[0152] The following uses a specific case to verify the solution of this embodiment.

[0153] The model trained in this embodiment was used to test a 60m long VY4 cable. An open circuit was simulated at the cable terminal. The model output showed an open circuit and a fault distance of 58.48m. The actual fault distance differed from the predicted value by 2.42m. This demonstrates that the prediction model in this embodiment exhibits good robustness when applied to different cable types for which data has not been collected. Experiments were conducted on several other cable types. The results showed that all fault categories were successfully identified, with a maximum fault distance prediction error of 3.45m (for a 700m test cable, not yet laid out). Therefore, the model constructed in this paper possesses a certain degree of robustness. MAE is the mean absolute error, calculated using the following formula:

[0154]

[0155] Where n is the total number of samples, d is the true fault distance, and dpred is the network prediction value; the test results are as follows: Figure 4 As shown.

[0156] Example 2

[0157] This embodiment proposes a cable fault identification system based on a convolutional neural network, which uses the method in Embodiment 1 for fault identification, and includes a data acquisition module, a data processing module, a calculation module, and an interactive output module.

[0158] The data acquisition module is connected to the beginning of the cable under test to acquire reflected wave data or model data of the cable.

[0159] The data processing module preprocesses the collected data and then outputs it to the calculation module;

[0160] The calculation module performs calculations based on the trained prediction model, outputs the prediction results, and outputs them through an interactive output module.

[0161] The interactive output module outputs the predicted fault type and specific information about the fault location in the form of text, charts, or voice broadcast.

[0162] Example 3

[0163] This embodiment proposes a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described cable fault identification method based on a convolutional neural network.

[0164] Example 4

[0165] This embodiment proposes a computer-readable medium carrying one or more programs, which, when executed by an electronic device, enable the electronic device to implement the steps of the above-described cable fault identification method based on a convolutional neural network.

[0166] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A cable fault identification method based on convolutional neural networks, characterized in that, The specific steps include the following: S1. Based on the transmission line model, a one-dimensional convolutional neural network model is established to predict cable fault types and fault locations; S2. Build a test platform to collect data on cables of different models of a certain length to build a dataset for model training; S3. Train the prediction model using the dataset, iterating multiple times until the target accuracy is met. S4. Collect the reflected wave data of the cable to be tested and input it into the trained prediction model. The prediction model outputs the fault type probability vector and the distance between the fault location and the beginning of the cable.

2. The cable fault identification method based on convolutional neural networks according to claim 1, characterized in that, In step S1, the one-dimensional convolutional neural network model integrates multi-scale feature extraction and temporal attention mechanisms.

3. The cable fault identification method based on convolutional neural networks according to claim 1, characterized in that, The cable types in step S2 include 500 meters each of RVV2, BVV4, and BV6 type cables; The cables used are all multi-core wires, including three-color main core wires and their respective branch core wires.

4. A cable fault identification method based on a convolutional neural network according to claim 1 or 4, characterized in that, In step S2, different types of cables are simulated in real-world scenarios of incomplete breakage, complete breakage, or short circuit. In this case, the incomplete breakage was simulated by cutting each colored core wire in half. Simulate a complete break by completely severing the entire cable; A short circuit is simulated by shorting the cable core and the shielding layer.

5. The cable fault identification method based on a convolutional neural network according to claim 1, characterized in that, In step S3, the number of iterations and the model convergence constraints are set. When the predetermined number of iterations is reached, the model accuracy data is checked to see if the target convergence constraints are met. If the target is met, training is stopped. If not, the iterations are extended for several more iterations until the target convergence constraints are met.

6. The cable fault identification method based on a convolutional neural network according to claim 1, characterized in that, Step S4 also includes the following steps: S41, Input reflected wave data S42. Regarding the original reflected waveform data x raw After normalization, the preprocessed input waveform data is obtained, as shown in the following formula: In the above formula x raw The mean and standard deviation; S43. For the preprocessed waveform data The formula for spatiotemporal feature extraction is as follows: M i (x)=∪ k∈{3,5,7} ReLU(W i，k *x), h=h+M i (x), in, Each feature block consists of four sub-blocks, each containing a multi-scale convolution M. i (x) Temporal attention mechanism Connect h with the residual. Represents network connectivity, ReLU is the ramp activation function, k is the kernel width (corresponding to different spatial scales), and w i ,k,α t All of these are network parameters, and multi-scale features are integrated through channel splicing to achieve information fusion. S44. For cable fault type prediction; global feature compression is performed using a pooling layer, with the following formula: In the above formula, h[:,t] represents the t-th column of h; using a fully connected layer for fully connected mapping, the formula is as follows: z1 and z2 are the class scores for open and short circuits, respectively, and are linear combinations of global features; finally, the Softmax function is used for probability normalization. Where p1 and p2 are the probabilities of open circuit and short circuit, respectively; the one with the highest probability is taken as the final cable state type; S45. For fault distance prediction; after extracting spatiotemporal features, temporal feature enhancement is performed through convolutional layers, with the following formula: C reg :Conv1D(k=3,C out =128), Among them, C reg It is a one-dimensional convolutional layer with a kernel width k = 3 and h pool This is a global pooling layer; then it is mapped through a fully connected layer, with the following formula: in These are all parameters of the fully connected layer; the predicted distance d can be obtained by decoding. pred The decoding formula is: Where v p =2×10 8 m / s is the speed of light, and Tmax is the maximum time window, determined by the sampling rate and the number of points. S46. Output the prediction results.

7. The cable fault identification method based on a convolutional neural network according to claim 1, characterized in that, In step S4, the reflected wave data and cable model data are input into the trained prediction model. The prediction model outputs the fault type probability vector and the distance between the fault location and the cable start end.

8. A cable fault identification system based on a convolutional neural network, wherein the fault identification is performed using the method described in any one of claims 1-7, characterized in that, It includes a data acquisition module, a data processing module, a calculation module, and an interactive output module; The data acquisition module is connected to the beginning of the cable under test to acquire reflected wave data or model data of the cable. The data processing module preprocesses the collected data and then outputs it to the calculation module; The calculation module performs calculations based on the trained prediction model, outputs the prediction results, and outputs them through an interactive output module. The interactive output module outputs the predicted fault type and specific information about the fault location in the form of text, charts, or voice broadcast.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the cable fault identification method based on convolutional neural networks as described in any one of claims 1-7.

10. A computer-readable medium carrying one or more programs, characterized in that, When one or more programs are executed by an electronic device, the electronic device implements the steps of the cable fault identification method based on convolutional neural networks as described in any one of claims 1-7.

Images