Transform-based multi-mode DC transmission line fault identification and classification method and system

Abstract

The invention discloses a multi-mode direct current transmission line fault identification and classification method and system based on Transform, and belongs to the technical field of power system direct current transmission and relay protection. The method comprises the following steps: respectively inputting a collected voltage signal and a collected current signal of a direct current transmission line into a pre-established voltage channel and a pre-established current channel, and extracting voltage channel characteristics and current channel characteristics; carrying out attention calculation on the extracted voltage channel characteristics and current channel characteristics through a cross mode attention mechanism based on Transform, and carrying out fusion splicing on calculation results; and inputting a fused and spliced calculation result into a Softmax classifier, and carrying out fault identification and classification on the direct-current transmission line. According to the method, the accuracy, robustness and transparency of fault identification and classification of the direct-current transmission line are improved, and technical support is provided for intelligent operation maintenance and fault positioning diagnosis of a direct-current system.

Description

Technical Field

[0001] This invention relates to the field of DC transmission and relay protection technology in power systems, and more specifically, to a method and system for fault identification and classification of multimodal DC transmission lines based on Transformer. Background Technology

[0002] my country's energy resources and electricity load are extremely unevenly distributed, with concentrated energy resources in the Northwest and dense loads in the eastern coastal areas. This uneven distribution of energy sources and loads places higher demands on the safe and stable operation of the power system. To achieve optimized inter-regional energy allocation and efficient long-distance transmission, high-voltage direct current (HVDC) transmission technology, with its advantages of large transmission capacity, low loss, and superior control performance, has become an important component of my country's ultra-high-voltage power grid.

[0003] High-voltage direct current (HVDC) transmission systems typically traverse complex and varied geographical environments, such as mountainous areas, plateaus, deserts, and coastal regions, making them susceptible to faults caused by lightning strikes, insulation aging, and other external factors. A system fault can lead to converter lockout, current interruption, and other problems, seriously threatening power safety. Therefore, efficient, reliable, and intelligent fault identification and classification technologies are crucial for ensuring the safe, stable, and economical operation of HVDC systems.

[0004] Currently, research on fault identification and classification in high-voltage direct current transmission systems mainly includes the following technical approaches:

[0005] Analytical model methods:

[0006] Based on the physical and electromagnetic characteristics of power systems, this paper establishes a transient mathematical model of DC lines to theoretically analyze the voltage and current response characteristics under different types of faults (such as single-pole grounding, bipolar short circuit, and inter-pole short circuit), thereby identifying the fault type. This method, based on system parameters and operating status information, has clear physical meaning and good interpretability.

[0007] A method combining feature engineering and machine learning:

[0008] This type of method combines signal processing and statistical analysis techniques. It extracts feature vectors, such as energy distribution, frequency domain characteristics, and phase differences, by processing transient voltage and current signals during faults using Fourier transform, wavelet analysis, or Empirical Mode Decomposition (EMD). Then, machine learning algorithms (such as Support Vector Machine (SVM), decision trees, random forests, and K-Nearest Neighbors (KNN)) are used to construct a classification model, achieving accurate fault type identification. This method makes full use of historical data and operational samples, exhibiting high recognition accuracy and real-time performance.

[0009] Deep learning methods:

[0010] With the rapid development of artificial intelligence, deep learning technology has been widely applied in HVDC fault identification. By constructing deep structural models such as convolutional neural networks (CNN), long short-term memory networks (LSTM), and temporal convolutional networks (TCN), spatiotemporal features can be automatically extracted directly from the original waveform data, achieving end-to-end fault identification and classification. This method possesses high-dimensional feature representation capabilities and good generalization performance, making it suitable for complex and variable operating conditions and effectively improving the intelligent diagnostic level of the system.

[0011] The existing methods described above still have the following shortcomings in complex DC transmission fault identification scenarios:

[0012] (1) Analytical model method: Traditional analytical models rely on simplified line electrical parameters and idealized assumptions, which makes it difficult to fully characterize the uneven distribution of capacitance and inductance in actual systems and the nonlinear characteristics introduced by converter control. During the fault transient stage, the accuracy of analytical models drops significantly, making it difficult to effectively reflect the propagation and coupling relationship of high-frequency components, resulting in unstable fault feature extraction and insufficient identification accuracy.

[0013] (2) Feature engineering combined with machine learning: This type of method usually relies on manually extracted statistics, time-frequency indicators, or empirical features, and lacks the ability to automatically mine fault signal features. There are information redundancy and scale mismatch problems between different features, which makes it difficult for machine learning models to converge and have limited generalization performance during the training phase. In addition, in the face of complex transient disturbances and noisy backgrounds, manual features are easily affected by external factors, resulting in insufficient model stability.

[0014] (3) Deep learning technology methods: At present, the input of neural networks is mostly single-type data and feature sequences, with low information utilization and unable to support complex multi-fault scenarios; in addition, the reasoning process of deep learning models used in the field of DC power transmission lacks interpretability and is difficult to provide reliable support for engineering decisions. Summary of the Invention

[0015] To address the above problems, this invention proposes a Transformer-based method for fault identification and classification of multimode DC transmission lines, comprising:

[0016] The collected voltage and current signals of the DC transmission line are input into the pre-established voltage and current channels respectively to extract the voltage channel features and current channel features.

[0017] The extracted voltage channel features and current channel features are subjected to attention calculation through a Transformer-based cross-modal attention mechanism, and the calculation results are fused and stitched together.

[0018] The fused and spliced ​​calculation results are input into the Softmax classifier for DC transmission line fault identification and classification.

[0019] Optionally, the voltage and current signals can be discrete signal sequences.

[0020] Optional extraction of voltage path features and current path features includes:

[0021] For voltage signals in the voltage channel, convolution and self-attention transformation are performed in the time domain to extract the voltage channel features of the voltage signal.

[0022] The voltage channel characteristics include: transient characteristics and local patterns in the time domain of the voltage signal.

[0023] In the current channel, the current channel characteristics of the current signal are extracted through frequency domain transformation and spectral feature model.

[0024] The characteristics of the current channel include: the energy distribution and harmonic characteristics of the current signal in the frequency domain.

[0025] Optionally, attention calculations are performed on the extracted voltage channel features and current channel features using a Transformer-based cross-modal attention mechanism, including:

[0026] Using the extracted voltage channel features as queries and the current channel features as keys and values, we calculate the attention representation of voltage-focused current.

[0027] Using the extracted current channel features as queries and the voltage channel features as keys and values, the attention representation of current-focused voltage is calculated.

[0028] Optionally, the calculation results can be fused and stitched together, including:

[0029] We perform a weighted summation of the attention representations of voltage with respect to current and current with respect to voltage in the calculation results.

[0030] Optionally, the fused and stitched calculation results are input into a Softmax classifier for fault identification and classification of DC transmission lines, including:

[0031] After the fusion and splicing calculation results are input into the Softmax classifier, the fault categories are classified based on the Softmax classifier and the calculation results of the fusion and splicing. The fault probability of each fault category is calculated, and the fault category with the highest fault probability is output as the fault identification result.

[0032] Optionally, the method also includes:

[0033] Based on the fault identification and classification results of DC transmission lines, and using class activation map generation and multimodal heat map mapping, the region or frequency band in the input signal that contributes the most to the fault classification decision is displayed.

[0034] Furthermore, this invention also proposes a Transformer-based multimodal DC transmission line fault identification and classification system, comprising:

[0035] The feature extraction unit is used to input the collected voltage and current signals of the DC transmission line into the pre-established voltage and current channels, respectively, to extract the voltage channel features and current channel features.

[0036] The feature fusion unit is used to perform attention calculations on the extracted voltage channel features and current channel features through a Transformer-based cross-modal attention mechanism, and to fuse and stitch the calculation results.

[0037] The output unit is used to input the fused and spliced ​​calculation results into the Softmax classifier for DC transmission line fault identification and classification.

[0038] In another aspect, the present invention also provides a computing device, comprising: one or more processors;

[0039] A processor is used to execute one or more programs;

[0040] When the one or more programs are executed by the one or more processors, the method described above is implemented.

[0041] In another aspect, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements the method described above.

[0042] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0043] This invention provides a Transformer-based multimodal DC transmission line fault identification and classification method, comprising: inputting the collected voltage and current signals of the DC transmission line into pre-established voltage and current channels respectively to extract voltage and current channel features; performing attention calculation on the extracted voltage and current channel features using a Transformer-based cross-modal attention mechanism, and fusing and stitching the calculation results; and inputting the fused and stitched calculation results into a Softmax classifier for DC transmission line fault identification and classification. This invention extracts voltage and current channel features from pre-established voltage and current channels, performs attention calculation on the extracted features based on the Transformer cross-modal attention mechanism, and then uses the stitched features as input to the model for fault identification. Therefore, it achieves feature complementarity, correlates key feature regions with physical mechanisms, and improves the accuracy, robustness, and transparency of DC transmission line fault identification and classification, providing technical support for intelligent operation and maintenance and fault location diagnosis of DC systems. Attached Figure Description

[0044] Figure 1 This is a flowchart of a multimodal DC transmission line fault identification and classification method based on Transformer according to the present invention;

[0045] Figure 2 This is a schematic diagram of a multi-modal DC transmission line fault identification and classification method based on Transformer according to the present invention.

[0046] Figure 3 This is a structural diagram of a Transformer-based multimodal DC transmission line fault identification and classification system according to the present invention. Detailed Implementation

[0047] Exemplary embodiments of the invention will now be described with reference to the accompanying drawings. However, the invention may be embodied in many different forms and is not limited to the embodiments described herein. These embodiments are provided to fully and completely disclose the invention and to fully convey its scope to those skilled in the art. The terminology used in the exemplary embodiments illustrated in the drawings is not intended to limit the invention. In the drawings, the same units / elements are referred to by the same reference numerals.

[0048] Unless otherwise stated, the terms used herein (including technical terms) have their common meaning as understood by one of ordinary skill in the art. Furthermore, it is understood that terms defined in commonly used dictionaries should be understood to have a meaning consistent with the context of their relevant field, and not to be interpreted as having an idealized or overly formal meaning.

[0049] Example 1:

[0050] This invention proposes a Transformer-based multimodal DC transmission line fault identification and classification method S100, such as... Figure 1 As shown, its principle is as follows Figure 2 As shown, it specifically includes:

[0051] S101, the collected voltage and current signals of the DC transmission line are input to the pre-established voltage and current channels respectively, and the voltage channel features and current channel features are extracted.

[0052] S102 uses a Transformer-based cross-modal attention mechanism to perform attention calculations on the extracted voltage channel features and current channel features, and then fuses and splices the calculation results.

[0053] S103 inputs the fused and spliced ​​calculation results into the Softmax classifier to perform fault identification and classification of DC transmission lines.

[0054] Among them, the voltage signal and the current signal are discrete signal sequence signals.

[0055] Specifically, voltage and current signals are synchronously acquired from the DC transmission system, and the sampling frequency should be high enough to capture transient characteristics of faults.

[0056] For the acquired discrete signal sequence x(n), the z-score method is used for normalization, and the expression is:

[0057]

[0058] Where μx and σ x These represent the mean and standard deviation of the signal, respectively.

[0059] The sampled discrete signal sequence is represented as:

[0060] v(n), i(n), n = 0, 1, ..., N-1

[0061] Where x(n) is the voltage channel sequence and i(n) is the current channel sequence.

[0062] After normalization, the voltage channel sequence v(n) and the current channel sequence i(n) are obtained, which are used as inputs to the multimodal model.

[0063] The extraction of voltage channel features and current channel features includes:

[0064] For voltage signals in the voltage channel, convolution and self-attention transformation are performed in the time domain to extract the voltage channel features of the voltage signal.

[0065] The voltage channel characteristics include: transient characteristics and local patterns in the time domain of the voltage signal.

[0066] In the current channel, the current channel characteristics of the current signal are extracted through frequency domain transformation and spectral feature model.

[0067] The characteristics of the current channel include: the energy distribution and harmonic characteristics of the current signal in the frequency domain.

[0068] Specifically, separate feature extraction channels are established for the voltage and current modes. The voltage channel undergoes convolution and self-attention transformation in the time domain to extract transient features and local patterns in the time domain; the current channel employs frequency domain transformation and spectral feature modeling to capture the energy distribution and harmonic characteristics in the frequency domain.

[0069] Features are extracted independently from the two channels while maintaining modal differences, and then feature fusion is achieved through an alignment mechanism.

[0070] Specifically, voltage channel feature extraction includes:

[0071] A one-dimensional convolutional neural network (CNN) is used to extract local temporal features from the voltage signal channel. Let the voltage signal length be L, the convolution kernel weights be w, and the input signal be v(t). Then the convolution output feature map h(t) can be expressed as:

[0072]

[0073] Where b is the bias term. A pooling layer follows the convolutional layer to reduce the feature length, resulting in a convolutional feature mapping sequence:

[0074] Z v =[z v,1 ,z v,2 ,…,z v,T ]

[0075] A position code is added to each time location to incorporate timing information. The position code uses a sine / cosine form:

[0076]

[0077] Add PE to the corresponding convolutional feature vector.

[0078] Then, a multi-head self-attention mechanism is used to calculate the long-range dependence of the voltage characteristics on Z. v Linear mapping generates query, key, and value matrices:

[0079] Q v=Z v W Q ,K v =Z v W K V v =Z v W V

[0080] For each attention head, compute the scaled dot product attention:

[0081]

[0082] Where, d k The key vector dimension is used. The outputs of multiple attention heads are concatenated and linearly transformed to obtain the updated voltage feature sequence:

[0083] Z′ v =Concat(head1,head2,…,head H W O

[0084] Among them, W O It is a linear transformation matrix used to fuse the outputs of each head, Z ′ v This refers to the voltage characteristic F of the final output of this channel. v .

[0085] Specifically, current channel feature extraction (discrete wavelet transform or empirical mode decomposition can also be used instead of fast Fourier transform for current channel feature extraction) includes:

[0086] Applying the Fast Fourier Transform to the current sequence, its frequency domain representation is as follows:

[0087]

[0088] After taking the logarithm of the spectrum and performing min-max normalization, the amplitude spectrum is obtained. As a fundamental frequency domain vector.

[0089] Similar to the time-domain voltage signal, for the frequency-domain current signal, a one-dimensional convolution kernel is slid along the frequency axis, and the convolution output feature map h(f) is:

[0090]

[0091] Add a position code to each frequency location, also using a sine / cosine form:

[0092]

[0093] Add PE to the corresponding convolutional feature vector.

[0094] Then, similar to the multi-head self-attention mechanism of the voltage signal channel, the current characteristic F of the final output of this channel is obtained. i .

[0095] Specifically, an attention calculation is performed on the extracted voltage channel features and current channel features using a Transformer-based cross-modal attention mechanism, including:

[0096] Using the extracted voltage channel features as queries and the current channel features as keys and values, we calculate the attention representation of voltage-focused current.

[0097] Using the extracted current channel features as queries and the voltage channel features as keys and values, the attention representation of current-focused voltage is calculated.

[0098] Specifically, cross-modal attention mechanisms (which can be replaced by co-attention or tensor fusion networks) include:

[0099] After obtaining the characteristics of both voltage and current modes, information fusion is achieved through a cross-attention mechanism, which performs bidirectional calculations between the two modes.

[0100] First, the voltage channel characteristics are used as the query (Q) v Current channel characteristics are used as bonds and values ​​(K). i V i The attention representation of "voltage-focused current" is calculated using the following formula:

[0101]

[0102] Among them, H v←i This represents the voltage feature representation incorporating current mode information, where d is the dimension of the key vector.

[0103] Then, the current path characteristics are used as the query (Q) i Voltage channel characteristics are used as keys and values ​​(K). v V v ), calculate the attention representation of "current-of-interest voltage":

[0104]

[0105] Where H i←v This represents the current characteristics that incorporate voltage mode information.

[0106] Finally, the aligned features are obtained:

[0107] H v←i H i←v

[0108] These two sets of features serve as the inputs for the subsequent cross-modal fusion and classification modules.

[0109] The process of fusing and stitching together the calculation results includes:

[0110] We perform a weighted summation of the attention representations of voltage with respect to current and current with respect to voltage in the calculation results.

[0111] Specifically, modality fusion and feature concatenation include:

[0112] The characteristics of voltage and current channels H v←i and H i←v Weighted summation, i.e.:

[0113] F fusion =W v ⊙H v←i +W i ⊙H i←v

[0114] Where ⊙ represents the Hadama product, W v and W i This is a learnable weight vector used to balance the contributions of the two modes.

[0115] The fused and spliced ​​calculation results are input into a Softmax classifier for fault identification and classification of DC transmission lines, including:

[0116] After the fusion and splicing calculation results are input into the Softmax classifier, the fault categories are classified based on the Softmax classifier and the calculation results of the fusion and splicing. The fault probability of each fault category is calculated, and the fault category with the highest fault probability is output as the fault identification result.

[0117] Specifically, the Softmax classifier consists of multiple layers of fully connected neurons. The last layer uses the Softmax activation function to output the probability distribution of each fault type, and then fuses the features F. fusion Input a fault classifier, assuming there are C fault categories, and let the linear output of the last layer of the classifier be:

[0118] o = [o1, o2, ..., o C ]

[0119] The Softmax function transforms these unnormalized outputs into probability distributions p for each class. j :

[0120]

[0121] in:

[0122] And 0≤p j ≤1.

[0123] The cross-entropy loss function is chosen and used in conjunction with the softmax output to train the model. After training, the model outputs the class probabilities p for each class when inferring a new signal. j The final fault identification result is the category with the highest probability:

[0124]

[0125] The methods also include:

[0126] Based on the fault identification and classification results of DC transmission lines, and using class activation map generation and multimodal heat map mapping, the region or frequency band in the input signal that contributes the most to the fault classification decision is displayed.

[0127] Specifically, class activation graph generation and multimodal heatmap mapping (or attention weight visualization can be used to generate a display graph) include:

[0128] To enhance model interpretability, this invention employs a class activation mapping mechanism to generate heatmaps, which visually display the regions or frequency bands in the input signal that contribute the most to fault classification decisions.

[0129] Taking the voltage channel as an example, we extract the feature map of the model in the last convolutional layer. (Activation of the k-th channel at time t), and the weights of the Softmax fully connected layer corresponding to fault category c. The voltage-level activation mapping of c can be represented as a weighted sum of the output of that class for each convolutional channel activation:

[0130]

[0131] in, This represents the activation intensity of category c at time t. A larger value indicates a greater contribution of the voltage signal at time t to the identification of category c.

[0132] Similarly, for current modes, use the final frequency domain convolutional layer or feature map:

[0133]

[0134] and corresponding weights Calculate the CAM at frequency f:

[0135]

[0136] This result reflects the importance of frequency f for category c decision-making.

[0137] The obtained time-domain and frequency-domain heatmaps are plotted in relation to the original voltage waveform and current spectrum, respectively. The highlighted areas are the basis for the fault.

[0138] This invention constructs a multimodal neural network model based on a cross-attention mechanism by fusing voltage time-domain features and current frequency-domain features, achieving efficient alignment and feature complementarity between different signal modes. By introducing activation-based visualization technology, the interpretability of the model criteria is enhanced, linking key feature regions with physical mechanisms, thereby improving the accuracy, robustness, and transparency of fault identification and providing technical support for intelligent operation and maintenance and fault location diagnosis of DC systems.

[0139] Example 2:

[0140] This invention also proposes a Transformer-based multimodal DC transmission line fault identification and classification system 200, such as... Figure 2 As shown, it includes:

[0141] The feature extraction unit 201 is used to input the collected voltage and current signals of the DC transmission line into the pre-established voltage and current channels respectively, and extract the voltage channel features and current channel features.

[0142] The feature fusion unit 202 is used to perform attention calculation on the extracted voltage channel features and current channel features through a Transformer-based cross-modal attention mechanism, and to fuse and stitch the calculation results.

[0143] The output unit 203 is used to input the fused and spliced ​​calculation results into the Softmax classifier for DC transmission line fault identification and classification.

[0144] This invention extracts voltage and current channel features from pre-established voltage and current channels. The extracted features are then used for attention calculation based on the Transformer's cross-modal attention mechanism, and the resulting concatenation is used as input to the model for fault identification. This achieves feature complementarity, linking key feature regions with physical mechanisms, thus improving the accuracy, robustness, and transparency of fault identification and classification for DC transmission lines. This provides technical support for intelligent operation, maintenance, and fault location diagnosis of DC systems.

[0145] Example 3:

[0146] Based on the same inventive concept, this invention also provides a computer device, which includes a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the computer storage medium to implement corresponding method flows or corresponding functions, thereby implementing the steps of the methods in the above embodiments.

[0147] Example 4:

[0148] Based on the same inventive concept, this invention also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the steps of the method in the above embodiments.

[0149] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0150] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0151] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0152] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0153] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0154] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for fault identification and classification of multimodal DC transmission lines based on Transformer, characterized in that, include: The collected voltage and current signals of the DC transmission line are input into the pre-established voltage and current channels respectively to extract the voltage channel features and current channel features. The extracted voltage channel features and current channel features are subjected to attention calculation through a Transformer-based cross-modal attention mechanism, and the calculation results are fused and stitched together. The fused and spliced ​​calculation results are input into the Softmax classifier for DC transmission line fault identification and classification.

2. The multimode DC transmission line fault identification and classification method according to claim 1, characterized in that, The voltage and current signals are discrete signal sequences.

3. The multimode DC transmission line fault identification and classification method according to claim 1, characterized in that, Extraction of voltage channel features and current channel features, including: For voltage signals in the voltage channel, convolution and self-attention transformation are performed in the time domain to extract the voltage channel features of the voltage signal. The voltage channel characteristics include: transient characteristics and local patterns in the time domain of the voltage signal; In the current channel, the current channel characteristics of the current signal are extracted through frequency domain transformation and spectral feature model. The characteristics of the current channel include: the energy distribution and harmonic characteristics of the current signal in the frequency domain.

4. The multimode DC transmission line fault identification and classification method according to claim 1, characterized in that, A Transformer-based cross-modal attention mechanism is used to perform attention calculations on the extracted voltage channel features and current channel features, including: Using the extracted voltage channel features as queries and the current channel features as keys and values, we calculate the attention representation of voltage-focused current. Using the extracted current channel features as queries and the voltage channel features as keys and values, the attention representation of current-focused voltage is calculated.

5. The multimode DC transmission line fault identification and classification method according to claim 1, characterized in that, The calculation results are merged and stitched together, including: We perform a weighted summation of the attention representations of voltage with respect to current and current with respect to voltage in the calculation results.

6. The multimode DC transmission line fault identification and classification method according to claim 1, characterized in that, The fused and stitched calculation results are input into a Softmax classifier for fault identification and classification of DC transmission lines, including: After the fusion and splicing calculation results are input into the Softmax classifier, the fault categories are classified based on the Softmax classifier and the calculation results of the fusion and splicing. The fault probability of each fault category is calculated, and the fault category with the highest fault probability is output as the fault identification result.

7. The multimode DC transmission line fault identification and classification method according to claim 1, characterized in that, The method further includes: Based on the fault identification and classification results of DC transmission lines, and using class activation map generation and multimodal heat map mapping, the region or frequency band in the input signal that contributes the most to the fault classification decision is displayed.

8. A Transformer-based multimodal DC transmission line fault identification and classification system, characterized in that, include: The feature extraction unit is used to input the collected voltage and current signals of the DC transmission line into the pre-established voltage and current channels, respectively, to extract the voltage channel features and current channel features. The feature fusion unit is used to perform attention calculations on the extracted voltage channel features and current channel features through a Transformer-based cross-modal attention mechanism, and to fuse and stitch the calculation results. The output unit is used to input the fused and spliced ​​calculation results into the Softmax classifier for DC transmission line fault identification and classification.

9. A computer device, characterized in that, include: One or more processors; A processor is used to execute one or more programs; When the one or more programs are executed by the one or more processors, the method described in claims 1-7 is implemented.

10. A computer-readable storage medium, characterized in that, It contains a computer program, which, when executed, implements the method as described in claims 1-7.

Images