A TMR-based method and system for cable defect detection and identification

Abstract

A method and system for cable defect detection and identification based on TMR (Transmission Measuring and Reconstructing) is disclosed. The method includes: receiving high-frequency pulse current signals from a power cable using an array rectangular probe integrating an excitation coil and a TMR sensor array; comparing the signal differences between defective and normal locations to obtain differential signals; and performing optimal spatial reconstruction of the differential signals using phase space reconstruction, symbol entropy, and recursive hierarchical search algorithms to obtain a differential reconstruction space matrix. The HIS (Hyperspective of Influence) distribution curve is then calculated and used as input features for a power cable defect identification model with an adaptive activation function, enabling automatic defect identification. This invention employs phase space reconstruction technology to reconstruct the spatial matrix of the differential signals. By calculating the HIS distribution curve of the differential reconstruction space matrix, the signal distribution differences reflected in the reconstruction matrix caused by fluctuations are eliminated, resulting in an overall 12% improvement in recognition rate compared to traditional methods combining statistical features such as peak voltage with models such as SVM and 1DCNN.

Description

Technical Field

[0001] This invention relates to the field of partial discharge signal identification technology for power distribution cables, and in particular to a method and system for cable defect detection and identification based on tunnel magnetoresistance (TMR). Background Technology

[0002] With the rapid development of industrialization and urbanization, cables, as core components of power transmission and distribution systems, have expanded their application to underground, underwater, and even inside various buildings. However, cables are susceptible to various factors under complex and variable operating conditions, such as physical damage, environmental corrosion, insulation aging, overload, and external mechanical stress. These factors can lead to a series of problems in cables, including insulation failure, breakage, short circuits, high-resistance faults, and flashover faults, seriously affecting the safe and stable operation of the power system.

[0003] Traditional cable fault detection methods, such as megohmmeter measurement, bridge methods (resistance bridge and capacitance bridge), pulse reflection methods (such as low-voltage pulse reflection method), and standing wave methods, while effective in locating certain types of faults, face limitations in handling high-impedance, flashover faults, and cable identification and location in complex environments. These techniques often require manual intervention, are time-consuming and inefficient, and their diagnostic accuracy and efficiency are affected when dealing with long-distance, deeply buried, or complexly laid-out cables.

[0004] In recent years, with the development of artificial intelligence, big data analytics, and sensor technology, intelligent cable monitoring and fault diagnosis systems have become a research hotspot. For example, the 8-channel symmetrical TMR sensor array probe designed by BABBAR and UNDERHILL has achieved initial success in crack defect detection through frequency domain analysis and PCA algorithms, but its applicability is mainly limited to crack detection around rivets, and it is still insufficient for identifying a wider range of cable defects. Similarly, while the research by SERGEEVA et al. demonstrated the application potential of TMR magnetoresistive array sensors in array pulse current technology and differentiated the advantages of dual-coil versus single-coil detection of different types of defects, its generalization ability and recognition accuracy under complex working conditions still need to be improved.

[0005] Therefore, although current cable defect detection technology has made some progress, it still generally faces problems such as low identification accuracy, slow response speed, and poor adaptability to complex operating conditions, especially in processing complex signal patterns and extracting deep features. There is an urgent need for a new cable defect detection technology that can deeply mine the spatial reconstruction features of current pulses, efficiently utilize adaptive intra- and inter-class correlations, and significantly improve the accuracy and response speed of defect signal identification under complex operating conditions. Summary of the Invention

[0006] To overcome the shortcomings of the existing technologies, this invention proposes a cable defect detection and identification method and system based on TMR. It uses phase space to reconstruct the differential signal space matrix, combines symbol entropy and iterative hierarchical optimization algorithm to reconstruct parameters, uses the HIS distribution curve as the input form of the identification model, and introduces the latest adaptive activation function to realize nonlinear data mapping, effectively eliminating the influence of signal fluctuations and improving the identification accuracy and response speed of defect signals under complex working conditions.

[0007] In accordance with the aforementioned objective, in a first aspect, the present invention proposes a cable defect detection and identification method based on TMR, characterized in that the method includes the following steps:

[0008] Step 1. Signal reception: A rectangular array probe integrating an excitation coil and a TMR sensor array is used to receive the reference signal of the high-frequency pulse current in the power cable;

[0009] Step 2. Differential signal acquisition: Acquire the high-frequency pulse current signal of partial discharge that changes when the power cable passes through the defect location, and obtain the differential signal by comparing it with the reference signal;

[0010] Step 3. Spatial Reconstruction: Using phase space reconstruction, combined with a symbol entropy-based and recursive hierarchical search algorithm, the differential signal is optimally reconstructed spatially to obtain the differential reconstruction space matrix. The recursive hierarchical search algorithm introduces a cosine mechanism and a random mechanism to avoid getting trapped in local optima and to determine the optimal embedding dimension and time delay parameters.

[0011] Step 4. HIS distribution curve calculation: Calculate the HIS distribution curve of the differential reconstruction space matrix. The HIS distribution curve is obtained by calculating the pixel histogram of the differential reconstruction space matrix and connecting the pixel distribution peaks. It is used to reflect the intensity and spatial distribution characteristics of pixel values ​​in the differential reconstruction space matrix.

[0012] Step 5. Use the HIS distribution curve as the input feature of the power cable defect identification model with an adaptive activation function to train and validate the power cable defect identification model. The adaptive activation function is applied to the last layer of the power cable defect identification model to implement adaptive nonlinear mapping of the data.

[0013] Preferably, the array rectangular probe in step 1 includes an excitation coil and a TMR sensor array. The excitation coil is used to generate a magnetic field to excite a high-frequency pulse current in the power cable, and the TMR sensor array is used to receive and convert the magnetic field changes into electrical signals.

[0014] Preferably, the excitation coil is made by winding enameled wire on a rectangular magnetic core, and the TMR sensor array is arranged in a straight line and soldered on the PCB board, with a preset distance between each pair.

[0015] Furthermore, the reference signal of the high-frequency pulse current in the power cable is acquired by placing the array rectangular probe at a position 180° to the defective end of the power cable.

[0016] By placing the array rectangular probe at a position 45° to the plane of the power cable, high-frequency pulse current signals in the power cable are collected, resulting in N high-frequency pulse current signals, where N is the total number of channels of the TMR sensor array.

[0017] Each high-frequency pulse current signal is compared with the reference signal to obtain N differential signals;

[0018] The peak value of the differential signal is extracted as the final partial discharge high-frequency pulse current differential signal by using a peak detection algorithm or a threshold setting method.

[0019] Furthermore, in step 3, based on symbolic entropy and a recursive hierarchical search algorithm, the differential signal is reconstructed in an optimal space to obtain the differential reconstruction space matrix, as shown in the following formula:

[0020]

[0021] In the formula, m is the embedding dimension and τ is the time delay;

[0022] Define each set of dimensional spaces generated by the phase space reconstruction as an m-dimensional vector v = (ν1,ν2,…,ν). m If ), then the symbolic entropy of all m vectors is h(m,τ)=-∑p A ln(p A ), PA is the mapping probability, f ε (i) is the tag mapping.

[0023] Furthermore, the recursive hierarchical search algorithm specifically includes initializing the particle matrix, defining the fitness function, iteratively updating the particle positions, and determining the optimal embedding dimension and time delay parameters based on the fitness value.

[0024] Furthermore, step 5, the power cable defect identification model, specifically includes:

[0025] - The input layer receives a 256*1 HIS distribution curve;

[0026] - It contains three convolutional layers and pooling layers, with convolutional kernel parameters of 10×1, 5×1 and 2×1, and the number of convolutional kernels set to 64, 64 and 128 respectively, with a stride of 1;

[0027] - The features after convolutional pooling are input into a fully connected layer for classification;

[0028] - Add a Dropout layer before the fully connected layer with a Dropout ratio of 0.2 to prevent overfitting;

[0029] - The batch processing sample size is 256;

[0030] - An adaptive activation function is introduced and applied to the last layer of the network. The adaptive activation function includes a correction factor and inter-class constraint factors P1 and P2, which are initialized to P1 = 1 and P2 = 0 to implement an adaptive nonlinear mapping of the data.

[0031] On the other hand, the present invention also provides a TMR-based cable defect detection and identification system, characterized in that it includes:

[0032] The array TMR rectangular probe module is used to acquire reference signals and differential signals of high-frequency pulse current in power cables and to preliminarily detect cable defects based on the differential signals.

[0033] The differential reconstruction module is used to perform optimal spatial reconstruction of the differential signal obtained by the array TMR rectangular probe module based on symbol entropy and recursive hierarchical search algorithm, obtain the differential reconstruction space matrix, and calculate the HIS distribution curve of the differential reconstruction space matrix.

[0034] The defect identification module uses the HIS distribution curve output by the differential reconstruction module as input to the AIICNN with an adaptive activation function. By training the AIICNN neural network model, it can automatically identify defects in power cables and output the defect identification results.

[0035] Furthermore, in the differential reconstruction module, symbolic entropy is used to symbolize the differential signal, and a recursive hierarchical search algorithm is used to determine the optimal embedding dimension and time delay parameters for phase space reconstruction, thereby obtaining the differential reconstruction space matrix.

[0036] Furthermore, the AIICNN neural network model in the defect recognition module includes an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer. The input layer receives the HIS distribution curve as input data, the convolutional and pooling layers are used to extract features, the fully connected layer is used for classification, and the output layer outputs the defect recognition result.

[0037] The cable defect detection and identification method and system based on TMR proposed in this invention have the following advantages and beneficial effects:

[0038] Unlike existing partial discharge pattern recognition schemes that use PRPD spectra and statistical features as input, this paper reconstructs the spatial matrix of the differential signal using phase space reconstruction. A method combining symbolic entropy and iterative hierarchical optimization is proposed for reconstructing parameters and optimizing the matrix. The HIS distribution curve of the differential reconstruction spatial matrix is ​​calculated and used as input to the subsequent recognition model. A novel adaptive activation function is introduced and applied to the last layer of the network to implement adaptive nonlinear mapping of the data. This eliminates the signal distribution differences reflected in the reconstruction matrix caused by fluctuations, resulting in an overall 12% improvement in recognition rate compared to traditional methods combining statistical features such as peak voltage with models like SVM and 1DCNN. Attached Figure Description

[0039] Figure 1 This is a flowchart of the cable defect detection and identification method based on TMR according to the present invention.

[0040] Figure 2 This is a flowchart of the defect identification process of the present invention.

[0041] Figure 3 This is a structural diagram of the rectangular probe of the present invention.

[0042] Figure 4 The defect identification performance of different classifiers.

[0043] Figure 5 The defect recognition performance of different classifiers under different signal-to-noise ratios. Detailed Implementation

[0044] The cable defect detection and identification method and system based on TMR described in this invention will be further explained and described below with reference to the accompanying drawings and specific embodiments. However, this explanation and description do not constitute an undue limitation on the technical solution of this invention.

[0045] like Figure 1 As shown, a TMR-based cable defect detection and identification method differs from traditional partial discharge pattern recognition schemes that use phase-analyzed partial discharge (PRPD) spectra and statistical features as input. This method employs phase space reconstruction technology to reconstruct the spatial matrix of the differential signal and integrates symbol entropy and an iterative hierarchical optimization algorithm to optimize the selection of reconstruction parameters. The HIS distribution curve of the differential reconstruction spatial matrix is ​​calculated and used as the input feature for the subsequent identification model. Simultaneously, a state-of-the-art adaptive activation function is introduced and applied to the last layer of the neural network model to implement adaptive nonlinear mapping of the data, thereby enhancing the model's recognition capability.

[0046] Example:

[0047] A high-frequency pulse current detection platform was built, which mainly consists of four parts: a linear array rectangular probe, a specially designed test cable (which incorporates four simulated defects such as air gaps), a 6-channel synchronous data acquisition and processing module, and a data acquisition module built based on LabVIEW virtual instrument technology.

[0048] During the testing process, the cable under test is energized to generate a high-frequency excitation pulse signal. This signal is applied to the probe's excitation coil, thereby inducing a current in the cable. A TMR chip is placed below the excitation coil to receive and acquire the signal. These acquired signals are transmitted to a computer via a data acquisition card, and the current data is displayed and saved in real time using the LabVIEW software platform. The excitation current is 0.7A, the sampling frequency is 400kHz, the number of sampling points is 20,000, the duty cycle is 50%, and the signal is triggered on the rising edge.

[0049] Step 1: Use an array rectangular probe combined with a TMR sensor chip to receive the high-frequency pulse current reference signal from the power cable;

[0050] like Figure 3 As shown, the array rectangular probe includes an excitation coil and a TMR sensor array. The excitation coil generates a magnetic field to excite a high-frequency pulsed current in the power cable. In this embodiment, the excitation coil is constructed by winding 350 turns of 0.57mm enameled wire on a rectangular magnetic core with a length of 50mm, a width of 37mm, and a thickness of 10mm. The TMR sensor array consists of six TMR sensors arranged in a straight line and soldered onto a PCB board, with a distance of 0.5mm between each pair. The TMR sensor array is embedded in a perforated transparent acrylic plate and positioned at the bottom center of the excitation coil, ensuring that the sensitive direction of the TMR chip receiving the magnetic field is aligned with the direction of the magnetic field, thereby maximizing reception efficiency.

[0051] To capture reference signals of high-frequency pulse currents in power cables, data can be acquired at a 180° angle between the array rectangular probe and defects such as cracks in the power cable. This is because when defects (such as cracks) exist in a power cable, the high-frequency pulse current generates specific magnetic field changes as it passes through these defects. These changes can be captured by the TMR sensor array and converted into electrical signals. By acquiring signals at a 180° angle, these magnetic field changes caused by defects can be captured to the maximum extent, thus providing a reliable reference for subsequent defect detection and identification.

[0052] Step 2: Collect the high-frequency pulse current signal of partial discharge that changes when the power cable passes through the defect location and obtain the differential signal to detect the cable defect.

[0053] Acquisition of high-frequency pulse current signal from partial discharge:

[0054] When defects (such as cracks or air gaps) exist in a power cable, they can cause uneven electric field distribution inside the cable, resulting in partial discharge at the defect location. This is collected using a specially designed array rectangular probe, as described in step 1.

[0055] In this embodiment, the optimal acquisition point is selected at 45° to the cable plane from the array rectangular probe. Simultaneously, a six-channel TMR sensor array is used for signal acquisition, with each channel capturing high-frequency pulse current signals at different locations within the cable. The signals acquired from these six channels are differentially processed with the signal acquired from the defective end of the cable, and the peak values ​​of the differential signals from the six channels are extracted as the final partial discharge high-frequency pulse current differential signal.

[0056] Once a differential signal is generated, it indicates that the partial discharge phenomenon can be detected.

[0057] Step 3: Based on symbolic entropy and recursive hierarchical search algorithms, the acquired differential signals are reconstructed using optimal space to obtain the differential reconstruction space matrix. The phase space reconstruction method is suitable for nonlinear time series analysis of dynamic systems. In the case of partial discharge signals, this method can extract their dynamic characteristics and reduce the impact on pattern recognition performance, making it an effective alternative to traditional signal classification methods.

[0058] The time series signal is reconstructed in phase space, where the embedding dimension is m and the time delay parameter is τ:

[0059]

[0060] Each set of dimensional spaces generated by the phase space reconstruction is denoted as an m-dimensional vector v = (ν1,ν2,…,ν). m If all m-dimensional vectors have the same symbolic entropy, then all m-dimensional vectors are labeled with the same symbol, and the corresponding indicator functions and label mappings are as follows:

[0061]

[0062] f ε (*)=(δ 1,2 ,δ 2,3 ,…,δ m-1,m )

[0063] Where |*| is the modulo operation, ε is the decision threshold, and is defined as follows: PA is the mapping probability, f ε (*) represents the set of tag maps.

[0064] Then the symbolic entropy is: h(m,τ)=-∑p A ln(p A ).

[0065] Determining the time delay parameter m and choosing the correct τ is crucial. To address this, we propose a recursive hierarchical search algorithm. By introducing a cosine mechanism and a stochastic mechanism, we make the parameter updates more random. Simultaneously, during iterative optimization, we consider the best and worst results at certain stages to prevent parameter updates from falling into local optima. The recursive hierarchical search algorithm mainly includes:

[0066] (1) First, initialize the number of particles to n, each particle as the embedding dimension m, and set a certain parameter for the time delay parameter τ;

[0067] (2) Define the particle matrix The particle position x is initialized as x = lb + rand(ub - lb), where lb and ub are the upper and lower boundaries of the dimension and the time delay parameter space, respectively.

[0068] (3) The fitness function is constructed by combining symbolic entropy with peak detection function.

[0069]

[0070]

[0071]

[0072]

[0073]

[0074]

[0075]

[0076] Where k is the adaptation factor.

[0077] (4) The particle position is iteratively updated using the following formula:

[0078]

[0079]

[0080] Where, β = 0.75 + e -i r2 is a random value between 0 and 1, x best x is the position of the particle with the best fitness value in 5 iterations. worse This represents the position of the particle with the worst fitness value in 5 iterations. Let be the particle position in the i-th iteration.

[0081] (5) When the fitness exceeds the set threshold or reaches a certain number of iterations, the optimal embedding dimension m and time delay parameter τ are obtained, and then the optimal difference reconstruction space matrix is ​​obtained.

[0082] Step 4: Calculate the HIS distribution curve of the difference reconstruction space matrix:

[0083] Step 4.1: Based on the differential reconstruction space matrix obtained in Step 3 (which contains the signal data after phase space reconstruction and differential processing), calculate the pixel histogram of the differential reconstruction space matrix;

[0084] Step 4.2 Find the peak values ​​in the pixel histogram and connect these pixel peak values ​​to form a curve, namely the HIS distribution curve.

[0085] Step 5: Use the HIS distribution curve obtained in Step 4 as input data for AIICNN (Adaptive Activation Function Improved Convolutional Neural Network) to train and validate the power cable defect identification model.

[0086] The activation function in the final feature nonlinear mapping process of traditional 1D Convolutional Neural Networks (1dCNNs) cannot adaptively handle the nonlinear mapping of data, especially under conditions of data distribution differences caused by fluctuations. Therefore, a state-of-the-art adaptive activation function is introduced and applied to the last layer of the network to implement adaptive nonlinear mapping of the data. This helps to eliminate data distribution differences caused by fluctuating operating conditions. Step 5 of the AIICNN specifically includes:

[0087] The input is a 256*1 HIS distribution curve. There are three layers in total: convolutional and pooling layers. The convolutional kernel parameters are 10×1, 5×1, and 2×1, with the number of kernels set to 64, 64, and 128, and a stride of 1. The features after convolution and pooling are input into a fully connected layer for classification, and a Dropout layer with a stride of 0.2 is added to prevent overfitting. The batch size is 256 samples. The following activation function is introduced and applied to the last layer of the network to implement an adaptive non-linear mapping of the data.

[0088]

[0089] Where, η a (x)=p1x,η b (x) = p2x (p1≠p2), β is the correction factor, P1 and P2 are inter-class constraint factors and are initialized to p1 = 1 and p2 = 0.

[0090] Training and validation:

[0091] During the training phase, the HIS distribution curve is randomly divided into training and test sets according to a certain proportion. The training set is used to train the AIICNN model, while the test set is used to verify the model's performance. By continuously adjusting the network parameters and structure, the optimal power cable defect identification model is obtained and used for power cable defect identification.

[0092] In the experiment, 10,000 samples were divided into an 80% training set and a 20% test set. The experiment was conducted with 300 iterations and 5 trials, with the average value taken. The PyTorch framework, i7-10700F CPU, NVIDIA 3050 GPU, and 32GB RAM were used. Pattern recognition results were evaluated using overall accuracy and F1 score.

[0093] The differential space reconstruction feature selection method based on symbolic entropy and HIS distribution curves can achieve advantages in defect identification. As shown in Table 1, this method achieves 96.55% and 94.76% in ACC and F1 scores, respectively.

[0094] Table 1

[0095]

[0096] To verify the advantages of the HIS distribution features extracted by differential spatial reconstruction-HIS in cable defect identification, this invention compared AIICNN with traditional PRPD maps and 14 statistical feature defect identification methods such as mean voltage, peak voltage, and crest factor, using differential spatial reconstruction-HIS features as input. The results are shown in Table 2 and Appendix. Figure 4 As shown.

[0097] The differential reconstruction-HIS+AIICNN pattern recognition method outperforms other mainstream methods, achieving an accuracy of 98.55% and an F1 score of 96.76%, with relatively good discriminative power. Compared to methods using statistical features (TA) as input, this invention's method can improve accuracy by up to 6.35%; compared to methods using PRPD maps as input, the accuracy improvement is approximately 2%. AIICNN demonstrates a significant advantage in handling differential spatial reconstruction-HIS datasets for partial discharges caused by cable defects, achieving an overall accuracy of 98.55%. In contrast, other classifiers have relatively lower F1 scores, ranging from 0.40% to 5.41%. (See attached image) Figure 5 As shown, the defect recognition rates of various models under different SNRs for the differential space reconstruction-HIS features described in this invention are also compared.

[0098] Table 2

[0099]

[0100] Secondly, we also propose a TMR-based cable defect detection and identification system. This includes:

[0101] Array TMR rectangular probe module: used to acquire reference signals and differential signals of high-frequency pulse current in power cables and to preliminarily detect cable defects based on the differential signals;

[0102] Differential reconstruction module: It is used to perform optimal spatial reconstruction of the acquired differential signals based on symbol entropy and recursive hierarchical search algorithm to obtain the differential reconstruction space matrix and calculate the HIS distribution curve of the differential reconstruction space matrix;

[0103] Defect identification module: It is used to train the AIICNN neural network by using the HIS distribution curve as input to obtain the defect identification results of power cables.

Claims

1. A method for cable defect detection and identification based on TMR, characterized in that, The method includes the following steps: Step 1. Signal reception: A rectangular array probe integrating an excitation coil and a TMR sensor array is used to receive the reference signal of the high-frequency pulse current in the power cable; Step 2. Differential signal acquisition: Acquire the high-frequency pulse current signal of partial discharge that changes when the power cable passes through the defect location, and obtain the differential signal by comparing it with the reference signal; Step 3. Spatial Reconstruction: Using phase space reconstruction, combined with a symbol entropy-based and recursive hierarchical search algorithm, the differential signal is optimally reconstructed spatially to obtain the differential reconstruction space matrix. The recursive hierarchical search algorithm introduces a cosine mechanism and a random mechanism to avoid getting trapped in local optima and to determine the optimal embedding dimension and time delay parameters. Step 4. HIS distribution curve calculation: Calculate the HIS distribution curve of the differential reconstruction space matrix. The HIS distribution curve is obtained by calculating the pixel histogram of the differential reconstruction space matrix and connecting the pixel distribution peaks. It is used to reflect the intensity and spatial distribution characteristics of pixel values ​​in the differential reconstruction space matrix. Step 5. Use the HIS distribution curve as the input feature of the power cable defect identification model with an adaptive activation function to train and validate the power cable defect identification model. The adaptive activation function is applied to the last layer of the power cable defect identification model to implement adaptive nonlinear mapping of the data.

2. The cable defect detection and identification method based on TMR according to claim 1, characterized in that, The array rectangular probe in step 1 includes an excitation coil and a TMR sensor array. The excitation coil is used to generate a magnetic field to excite a high-frequency pulse current in the power cable, and the TMR sensor array is used to receive and convert the magnetic field changes into electrical signals.

3. The cable defect detection and identification method based on TMR according to claim 2, characterized in that, The excitation coil is made by winding enameled wire on a rectangular magnetic core. The TMR sensor array is arranged in a straight line and soldered onto the PCB board, with a preset distance between each pair.

4. The cable defect detection and identification method based on TMR according to claim 1, characterized in that, The reference signal of the high-frequency pulse current in the power cable is acquired by placing the array rectangular probe at a position 180° to the defective end of the power cable. By placing the array rectangular probe at a position 45° to the plane of the power cable, high-frequency pulse current signals in the power cable are collected, resulting in N high-frequency pulse current signals, where N is the total number of channels of the TMR sensor array. Each high-frequency pulse current signal is compared with the reference signal to obtain N differential signals; The peak value of the differential signal is extracted as the final partial discharge high-frequency pulse current differential signal by using a peak detection algorithm or a threshold setting method.

5. The cable defect detection and identification method based on TMR according to claim 1, characterized in that, Step 3. Based on symbolic entropy and recursive hierarchical search algorithm, the differential signal is reconstructed in optimal space to obtain the differential reconstruction space matrix, as shown in the following formula: In the formula, m is the embedding dimension and τ is the time delay; Define each set of dimensional spaces generated by the phase space reconstruction as an m-dimensional vector v = (v1, v2, ..., ν) m If ), then the symbolic entropy of all m vectors is h(m,τ)=-∑p A ln(p A ), PA is the mapping probability, f ε (i) is the tag mapping.

6. The cable defect detection and identification method based on TMR according to claim 5, characterized in that, The recursive hierarchical search algorithm specifically includes initializing the particle matrix, defining the fitness function, iteratively updating the particle positions, and determining the optimal embedding dimension and time delay parameters based on the fitness value.

7. The cable defect detection and identification method based on TMR according to claim 1, characterized in that, Step 5, the power cable defect identification model, specifically includes: - The input layer receives a 256*1 HIS distribution curve; - It contains three convolutional layers and pooling layers, with convolutional kernel parameters of 10×1, 5×1 and 2×1, and the number of convolutional kernels set to 64, 64 and 128 respectively, with a stride of 1; - The features after convolutional pooling are input into a fully connected layer for classification; - Add a Dropout layer before the fully connected layer with a Dropout ratio of 0.2 to prevent overfitting; - The batch processing sample size is 256; - An adaptive activation function is introduced and applied to the last layer of the network. The adaptive activation function includes a correction factor and inter-class constraint factors P1 and P2, which are initialized to P1 = 1 and P2 = 0 to implement an adaptive nonlinear mapping of the data.

8. A cable defect detection and identification system based on TMR, characterized in that, include: The array TMR rectangular probe module is used to acquire reference signals and differential signals of high-frequency pulse current in power cables and to preliminarily detect cable defects based on the differential signals. The differential reconstruction module is used to perform optimal spatial reconstruction of the differential signal obtained by the array TMR rectangular probe module based on symbol entropy and recursive hierarchical search algorithm, obtain the differential reconstruction space matrix, and calculate the HIS distribution curve of the differential reconstruction space matrix. The defect identification module uses the HIS distribution curve output by the differential reconstruction module as input to the AIICNN with an adaptive activation function. By training the AIICNN neural network model, it can automatically identify defects in power cables and output the defect identification results.

9. The cable defect detection and identification system based on TMR according to claim 8, characterized in that, In the differential reconstruction module, symbolic entropy is used to symbolize the differential signal, and a recursive hierarchical search algorithm is used to determine the optimal embedding dimension and time delay parameters for phase space reconstruction, thereby obtaining the differential reconstruction space matrix.

10. The cable defect detection and identification system based on TMR according to claim 8, characterized in that, The AIICNN neural network model in the defect recognition module includes an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer. The input layer receives the HIS distribution curve as input data, the convolutional and pooling layers are used to extract features, the fully connected layer is used for classification, and the output layer outputs the defect recognition result.

Images