A method, device and equipment for identifying inrush current of excitation and a storage medium

Abstract

The application provides a method and device for identifying magnetizing inrush current, equipment and storage medium, and relates to the technical field of transformers. The method first acquires original transient current samples and performs empirical mode decomposition to obtain a plurality of characteristic components; then, based on a harmonic component proportion index, effective characteristic components are selected from the plurality of characteristic components; then, the effective characteristic components are input into a feature extraction network model composed of a first network unit and a second network unit connected in series, and the ability of the first network unit to extract global dependent features and the ability of the second network unit to extract time evolution features are trained. Finally, whether the current to be identified is magnetizing inrush current is identified through the target feature extraction network model obtained through training. The purity of the input features is improved through the screening mechanism, and the feature extraction capability is enhanced through the serial hybrid network, thereby realizing high-precision identification of magnetizing inrush current.

Description

Technical Field

[0001] This application relates to the field of transformer technology, and in particular to a method, apparatus, device and storage medium for identifying inrush current. Background Technology

[0002] In some cases, the inrush current generated by a transformer exhibits a high degree of similarity in transient waveform characteristics to the fault current generated when an internal fault occurs. Especially during the initial closing phase or under residual magnetism, the inrush current can generate a large non-periodic surge current, the amplitude of which may approach or even exceed the fault current.

[0003] In existing technologies, methods such as second harmonic braking, waveform symmetry criteria, transient energy criteria, and machine learning methods based on artificial feature extraction can be used to distinguish between inrush current and fault current.

[0004] However, under varying closing angles and residual magnetism conditions, the harmonic characteristics of the current may be distorted, leading to a decrease in the sensitivity of conventional methods such as second harmonic braking. Furthermore, feature extraction methods based on human experience typically employ a single network architecture, which struggles to fully characterize the complex nonlinear and temporal dynamics of the inrush current, resulting in insufficient generalization ability of the model across different fault types, closing conditions, and various scenarios. Summary of the Invention

[0005] The main objective of this application is to provide a method, apparatus, device, and storage medium for identifying inrush current, aiming to solve the technical problem of how to improve the accuracy of identifying inrush current and fault current of transformers.

[0006] To achieve the above objectives, this application provides a method for identifying inrush current, the steps of which include: Obtain the original transient current sample; Empirical mode decomposition is performed on the original transient current sample to obtain the feature components corresponding to each frequency domain feature; Based on the harmonic component proportion index, effective feature components are selected from multiple feature components. The initial feature extraction network model is trained using the effective feature components to obtain the target feature extraction network model. The initial feature extraction network model includes a first network unit for extracting the global dependency features in the effective feature components and a second network unit for extracting the temporal evolution features in the global dependency features. The target feature extraction network model is used to identify the current to be identified formed by the transformer and output the identification result. The identification result is used to characterize whether the current to be identified is an inrush current.

[0007] In one embodiment, the step of performing empirical mode decomposition on the original transient current sample to obtain the feature components corresponding to each frequency domain feature includes: The original transient current samples are standardized to obtain the original transient current sequence; The original transient current sequence is integrated along the channel dimension to form a multidimensional original feature matrix; The multidimensional original feature matrix is ​​subjected to empirical mode decomposition for each channel to obtain multiple intrinsic mode function components corresponding to each channel. The intrinsic modulus function components of each channel are classified according to their respective frequency domain characteristics to form the characteristic components.

[0008] In one embodiment, before the step of filtering out effective feature components from the plurality of feature components based on the harmonic component proportion index, the method further includes: Obtain the frequency domain spectrum of each of the aforementioned feature components; The amplitudes of the second, third, and fourth harmonics in the frequency domain spectrum are extracted respectively. The ratio of the sum of the amplitudes of the second, third, and fourth harmonics to the total amplitude of the frequency domain spectrum is used as the harmonic component proportion index.

[0009] In one embodiment, the step of filtering out effective feature scores from multiple feature components based on the harmonic component proportion index includes: The feature components whose harmonic component proportion index is not less than a preset threshold are taken as the effective feature components.

[0010] In one embodiment, the step of training the initial feature extraction network model using the effective feature components to obtain the target feature extraction network model includes: The effective feature components are divided using a sliding time window to obtain multiple feature sub-matrices with the same time window length; Using each of the aforementioned feature sub-matrices as input samples, the initial feature extraction network model is trained to obtain the target feature extraction network model.

[0011] In one embodiment, the step of using each of the feature sub-matrices as input samples to train the initial feature extraction network model to obtain the target feature extraction network model includes: The input samples are divided into a training set, a validation set, and a test set; The initial feature extraction network model is iteratively trained using the training set, and the network performance of the initial feature extraction network model during the training process is evaluated using the validation set. The initial feature extraction network model trained was tested and verified using a test set, and the verified initial feature extraction network model was used as the target feature extraction network model.

[0012] In one embodiment, the original transient current sample includes an inrush current sample and a fault current sample, and the step of obtaining the original transient current sample includes: Construct a simulation model for no-load closing and a simulation model for fault current of the transformer; The inrush current sample is generated using the no-load closing simulation model. The fault current sample is generated using the fault current simulation model.

[0013] Furthermore, to achieve the above objectives, this application also provides an inrush current identification device, the inrush current identification device comprising: The sample acquisition module is used to acquire raw transient current samples; The data processing module is used to perform empirical mode decomposition on the original transient current sample to obtain the feature components corresponding to each frequency domain feature. The data processing module is also used to filter out effective feature components from multiple feature components based on the harmonic component proportion index. The model training module is used to train the initial feature extraction network model using the effective feature components to obtain the target feature extraction network model. The initial feature extraction network model includes a first network unit for extracting the global dependency features in the effective feature components and a second network unit for extracting the temporal evolution features in the global dependency features. The discrimination module is used to identify the current to be identified formed by the transformer through the target feature extraction network model and output the identification result. The identification result is used to characterize whether the current to be identified is an inrush current.

[0014] In addition, to achieve the above objectives, this application also provides an inrush current identification device, which includes: a memory, a processor, and an inrush current identification program stored in the memory and executable on the processor. The inrush current identification program is configured to implement the steps of the inrush current identification method described above.

[0015] In addition, to achieve the above objectives, this application also provides a storage medium, which is a computer-readable storage medium, and stores an inrush current identification program thereon. When the inrush current identification program is executed by a processor, it implements the steps of the inrush current identification method described above.

[0016] This application provides a method, apparatus, device, and storage medium for identifying inrush current. The method includes the following steps: acquiring an original transient current sample; performing empirical mode decomposition on the original transient current sample to obtain feature components corresponding to each frequency domain feature; selecting effective feature components from multiple feature components based on the harmonic component ratio index; training an initial feature extraction network model using the effective feature components to obtain a target feature extraction network model, wherein the initial feature extraction network model includes a first network unit for extracting globally dependent features from the effective feature components and a second network unit for extracting time-series evolution features from the globally dependent features; and identifying the current to be identified formed by the transformer using the target feature extraction network model and outputting the identification result, wherein the identification result is used to characterize whether the current to be identified is an inrush current.

[0017] First, raw transient current samples are acquired and subjected to empirical mode decomposition to obtain multiple feature components. Then, effective feature components are selected from these components based on the harmonic component ratio. These effective feature components are then input into a feature extraction network model composed of a first network unit and a second network unit connected in series. The model is trained to improve the ability of the first network unit to extract globally dependent features and the ability of the second network unit to extract temporal evolution features. Finally, the trained target feature extraction network model is used to identify whether the current to be identified is an inrush current. By improving the purity of the input features through a selection mechanism and enhancing the feature extraction capability through a serial hybrid network, high-precision identification of inrush current is achieved. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating an embodiment of the inrush current identification method of this application. Figure 2 This is a flowchart illustrating Embodiment 2 of the excitation inrush current identification method of this application; Figure 3 This is a flowchart illustrating Embodiment 3 of the excitation inrush current identification method of this application; Figure 4 This is a schematic diagram of the confusion matrix corresponding to the test set results; Figure 5 The diagram shows the three-phase currents of a transformer under no-load switching conditions with no residual magnetism. Figure 6 The three-phase current diagram for transformer no-load closing under 70% residual magnetism conditions; Figure 7 The diagram shows the three-phase current and voltage of the transformer under internal three-phase short-circuit fault conditions. Figure 8 This is a schematic diagram of the module structure provided in Embodiment 1 of the excitation inrush current identification device of this application; Figure 9 This is a structural schematic diagram of the excitation inrush current identification device according to Embodiment 1 of this application.

[0021] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0022] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0023] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0024] This application presents a first embodiment of the inrush current identification method. Please refer to [link / reference]. Figure 1 The inrush current identification method includes steps S10-S50: Step S10: Obtain the original transient current sample; It should be understood that, in this embodiment, the executing entity can be an inrush current identification device for protecting the mechanical energy of a transformer.

[0025] It should be noted that, in this embodiment, the original transient current sample refers to the three-phase current transient waveform data sample directly collected from the transformer without preprocessing.

[0026] It is easy to understand that the original transient current sample can specifically be a current data sample collected under different operating conditions, such as a sample of inrush current generated under a certain condition, or a sample of fault current generated under a certain condition. In this embodiment, by acquiring current samples containing different operating conditions, it can be ensured that the model trained subsequently has broad applicability and can cope with various situations that may occur in actual operation, such as a certain closing angle, a certain residual magnetism condition, and a certain fault type.

[0027] Step S20: Perform empirical mode decomposition on the original transient current sample to obtain the feature components corresponding to each frequency domain feature; It should be noted that, in this embodiment, empirical mode decomposition is an adaptive nonlinear nonstationary signal decomposition algorithm that does not require preset basis functions and can complete component decoupling based on the signal's own time scale characteristics; frequency domain characteristics refer to the signal's characteristic performance in different frequency ranges, including high-frequency oscillation characteristics, harmonic characteristics, fundamental wave characteristics, etc.; characteristic components refer to the intrinsic mode function components obtained through empirical mode decomposition, and each component corresponds to a different frequency band and time scale.

[0028] It is easy to understand that, in this embodiment, the mixed original transient current sample can be decomposed into multiple intrinsic mode function components through empirical mode decomposition (EMD), with each component representing the evolution of different frequency components in the current sample. The above-mentioned EMD process does not rely on preset basis functions and can adaptively match the inherent characteristics of the transformer transient current, providing clean input data for subsequent harmonic characteristic analysis and effective component selection.

[0029] Step S30: Based on the harmonic component proportion index, select effective feature components from multiple feature components; It should be noted that, in this embodiment, the harmonic component proportion index refers to an index used to quantify the significance of harmonic information related to inrush current identification in the characteristic components. The harmonic information can stably reflect the difference between inrush current and fault current. The effective characteristic component refers to the characteristic component whose corresponding harmonic component proportion index reaches a certain level.

[0030] It is easy to understand that not all feature components obtained through empirical mode decomposition have the same value for inrush current identification. Among high-frequency noise components, low-frequency fundamental components, and trend term components, harmonic information relevant to inrush current identification accounts for a relatively low proportion. Including these invalid components in subsequent models would introduce interference and affect identification accuracy. In this embodiment, a quantitative screening based on the proportion of harmonic components can be used to eliminate invalid components at the source, retaining only the most valuable valid components and improving the signal-to-noise ratio of the input features.

[0031] Step S40: Train the initial feature extraction network model using the effective feature components to obtain the target feature extraction network model. The initial feature extraction network model includes a first network unit for extracting the global dependency features in the effective feature components and a second network unit for extracting the temporal evolution features in the global dependency features. It should be noted that, in this embodiment, the initial feature extraction network model refers to a deep learning network model that has not yet been trained or is in the process of training; the target feature extraction network model refers to a network model that has been trained and has the ability to identify inrush current; the first network unit refers to a neural network unit used to extract the global dependency features of the input parameters; and the second network unit refers to a neural network unit used to extract the temporal evolution features from the global dependency features. Here, global dependency features refer to the global correlation across time scales in the feature sequence of the current; and temporal evolution features refer to the dynamic change pattern of the feature sequence of the current over time.

[0032] It should be understood that, in this embodiment, the first network unit can specifically adopt the Transformer module, which can capture the long-distance correlation between different time points within the (current) sequence through a multi-head self-attention mechanism; the second network unit can specifically adopt the Bidirectional Long Short-Term Memory (BiLSTM) module, which can simultaneously extract the dynamic change features of the (current) sequence in both forward and reverse directions.

[0033] It is easy to understand that in this embodiment, the effective feature components are input into the initial feature extraction network model in a sequential manner to train it, thereby obtaining a target feature extraction network model that can achieve high-precision discrimination of inrush current and fault current.

[0034] It is worth noting that in this embodiment, the feature extraction network model used in this application is composed of a first network unit and a second network unit serially. Based on this serial architecture, the first network unit can first establish global dependencies across time scales, and then the second network unit can extract bidirectional temporal evolution features under global context constraints, thus forming a hierarchical representation mechanism of "global first, then temporal". This hierarchical representation mechanism can work in conjunction with the aforementioned harmonic component proportion screening mechanism. The harmonic component proportion screening mechanism is responsible for improving the feature purity of the feature components on the model input side, the first network unit is responsible for establishing global dependencies, and the second network unit is responsible for extracting temporal evolution features. The three work together to further improve the accuracy and robustness of inrush current identification under complex operating conditions.

[0035] Step S50: The target feature extraction network model is used to identify the current to be identified formed by the transformer and output the identification result. The identification result is used to characterize whether the current to be identified is an inrush current.

[0036] It should be noted that, in this embodiment, the current to be identified refers to the three-phase current data or current signal collected during the actual operation of the transformer. The type of the current data or current signal is unknown, and it is impossible to directly determine whether it is actually an inrush current or a fault current.

[0037] It is easy to understand that, in this embodiment, after obtaining the trained target feature extraction network model, it can be applied. Specifically, when the current to be identified is collected, the current to be identified can first undergo the same or similar preprocessing process as the training samples, and then the features of the processed current to be identified are input into the trained target feature extraction network model. After passing through global dependency feature extraction and temporal evolution feature extraction in sequence, the probability of it belonging to each category is calculated through the fully connected layer and output layer of the target feature extraction network model, and the corresponding identification result is output to characterize whether the current to be identified is an inrush current or a fault current.

[0038] This application provides a method for identifying inrush current. The method involves first acquiring original transient current samples and performing empirical mode decomposition to obtain multiple feature components. Then, based on the harmonic component ratio index, effective feature components are selected from these components. Next, the effective feature components are input into a feature extraction network model composed of a first network unit and a second network unit connected in series. The ability of the first network unit to extract globally dependent features and the ability of the second network unit to extract temporal evolution features are trained. Finally, the trained target feature extraction network model is used to identify whether the current to be identified is an inrush current. By improving the purity of the input features through a selection mechanism and enhancing the feature extraction capability through a serial hybrid network, high-precision identification of inrush current is achieved.

[0039] Based on the first embodiment of the inrush current identification method of this application, in the second embodiment of the inrush current identification method of this application, the content that is the same as or similar to the first embodiment of the inrush current identification method described above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 2 The step of performing empirical mode decomposition on the original transient current sample to obtain the feature components corresponding to each frequency domain feature includes: Step S21: Standardize the original transient current sample to obtain the original transient current sequence; It should be noted that, in this embodiment, standardization processing refers to a data processing method that uniformly truncates the duration and normalizes the amplitude of the original transient current samples, which can ensure that all samples have the same sequence length and consistent numerical range; the original transient current sequence refers to the time series data with a fixed number of sampling points formed after standardization processing, such as truncation of 0.2 seconds at a sampling frequency of 4kHz, corresponding to 800 sampling points.

[0040] It is easy to understand that current signals acquired under different simulation scenarios or actual measurement conditions may differ in duration and amplitude scale. In this embodiment, by standardizing the original transient current samples, the impact of these differences on subsequent analysis can be eliminated, ensuring the consistency and comparability of the feature extraction process and providing a standardized input basis for subsequent empirical mode decomposition.

[0041] Step S22: Integrate the original transient current sequence by channel dimension to form a multidimensional original feature matrix; It should be noted that, in this embodiment, channel dimension integration refers to the process of splicing the original transient current sequence according to the channel dimension to form a multidimensional matrix structure; the multidimensional original feature matrix refers to the matrix formed by the original transient current sequence under the condition of a specific number of sampling points and a specific number of channels.

[0042] It is easy to understand that the original transient current sequence of a transformer can contain multiple channels (such as phase A, phase B, and phase C), and these channels are correlated. In this embodiment, the original transient current sequence can be integrated along the channel dimension, combining the multi-channel data into a unified multi-dimensional matrix, which facilitates subsequent empirical mode decomposition by channel.

[0043] Step S23: Perform empirical mode decomposition on the multidimensional original feature matrix according to each channel to obtain multiple intrinsic mode function components corresponding to each channel; It should be noted that in this embodiment, empirical mode decomposition refers to decomposing the time sequence of each channel separately; intrinsic mode function components refer to signal components that satisfy the intrinsic mode function conditions obtained through the above decomposition process. Each component corresponds to a different frequency band and time scale, wherein the high-frequency component represents high-frequency oscillation, the mid-frequency component represents harmonic components, and the low-frequency component represents the fundamental wave and aperiodic attenuation components.

[0044] It is easy to understand that, in this embodiment, by performing empirical mode decomposition on the multidimensional original feature matrix, the mixed transient current sequence of each channel can be sequentially decomposed into several intrinsic mode function components and a residual component. The residual component represents the trend term of the signal and has no effective feature information, so it can be discarded.

[0045] Step S24: The intrinsic mode function components of each channel are classified according to their respective frequency domain characteristics to form the feature components.

[0046] It is easy to understand that classifying according to frequency domain characteristics means dividing and merging the components obtained from the decomposition of each channel according to their frequency characteristics. In this embodiment, the intrinsic modulus function components with the same index in each channel can be merged to form the aforementioned characteristic components.

[0047] Furthermore, in this embodiment, before the step of filtering out effective feature components from multiple feature components based on the harmonic component proportion index, the method further includes: Step S301: Obtain the frequency domain spectrum of each of the aforementioned feature components; It should be noted that, in this embodiment, the frequency domain spectrum refers to the amplitude-frequency relationship graph obtained after converting the time domain signal into a frequency domain representation through Fast Fourier Transform, which reflects the energy distribution of the signal at different frequency components.

[0048] It is easy to understand that, in this embodiment, the difference in harmonic characteristics between the inrush current and the internal fault current is mainly reflected in the frequency domain. By converting the time-domain signal to the frequency domain, the distribution of harmonic components in each characteristic component can be quantitatively analyzed, providing basic data for calculating the proportion of harmonic components.

[0049] Step S302: Extract the amplitudes of the second, third, and fourth harmonics from the frequency domain spectrum, respectively. It should be noted that in this embodiment, the second harmonic refers to a harmonic component with a frequency twice that of the fundamental frequency; the third harmonic refers to a harmonic component with a frequency three times that of the fundamental frequency; and the fourth harmonic refers to a harmonic component with a frequency four times that of the fundamental frequency.

[0050] It is easy to understand that inrush current typically contains abundant second, third, and fourth harmonic components, while these harmonic components are relatively few in internal fault current. In this embodiment, by extracting the amplitude of the aforementioned specific harmonics, the harmonic information content related to inrush current identification in each characteristic component can be quantified.

[0051] Step S303: The ratio of the sum of the amplitudes of the second, third, and fourth harmonics to the total amplitude of the frequency domain spectrum is used as the harmonic component proportion index.

[0052] It should be noted that, in this embodiment, the total amplitude of the frequency domain spectrum refers to the sum of the amplitudes of all frequency points in the frequency domain spectrum of the characteristic component. The expression for the proportion of harmonic components corresponding to the channel of any characteristic component is as follows: ; in, The amplitude of the second harmonic. The amplitude of the third harmonic. The amplitude of the fourth harmonic. This represents the total amplitude of the frequency domain spectrum. This is an indicator of the proportion of harmonic components.

[0053] It is easy to understand that the harmonic component proportion index does not simply represent the energy level of a certain component, but rather measures the proportion of characteristic harmonic components related to inrush current identification in the characteristic component. In this embodiment, when the harmonic component proportion of the characteristic component is high, it indicates that the second, third, and fourth harmonic components account for a larger proportion of the characteristic component, which can more fully reflect the difference between the inrush current and the internal fault current in the transient harmonic structure.

[0054] It is worth noting that in this embodiment, the arithmetic mean of the harmonic component proportion index calculated for each dimension channel can be used as the final actual harmonic component proportion index for that feature component, which facilitates subsequent calculations.

[0055] Furthermore, in this embodiment, the step of filtering out effective feature scores from multiple feature components based on the harmonic component proportion index includes: Step S31: The feature component whose harmonic component proportion index is not less than a preset threshold is taken as the effective feature component.

[0056] It should be noted that, in this embodiment, the preset threshold refers to a pre-set critical value used to determine whether a feature component is a valid component, which can be determined by a large number of actual current samples or simulated current samples. As a specific approach, the preset threshold can be set to 15%.

[0057] It is easy to understand that components with a low proportion of harmonic components (such as high-frequency noise components and fundamental frequency components) mainly reflect the fundamental frequency trend, background disturbances, or components that are weakly related to the classification task, offering limited help to the classification task and potentially even introducing interference. In this embodiment, by setting a reasonable preset threshold, feature components with a harmonic component proportion index lower than the preset threshold are discarded, while effective components with a harmonic component proportion index not lower than the preset threshold are retained. This reduces the interference of invalid features on model training from the source and improves the quality of feature input.

[0058] Based on the first and / or second embodiments of the excitation inrush current identification method of this application, in the third embodiment of the excitation inrush current identification method of this application, the content that is the same as or similar to the first and second embodiments of the excitation inrush current identification method described above can be referred to the above description and will not be repeated hereafter. Based on this, please refer to... Figure 3 The step of training the initial feature extraction network model using the effective feature components to obtain the target feature extraction network model includes: Step S41: Divide the effective feature components using a sliding time window to obtain multiple feature sub-matrices with the same time window length; It should be noted that, in this embodiment, the sliding time window refers to a method of sliding and truncating sequence segments along the time axis with a fixed window length (e.g., 80 sampling points) and a fixed step size (e.g., 2 sampling points); the time window length refers to the number of sampling points contained in each truncated segment; and the feature submatrix refers to the local feature matrix obtained by truncating.

[0059] It is easy to understand that sliding time window partitioning can extract local temporal features from effective feature components, while maximizing the number of training samples while ensuring the integrity of sample features by setting the sliding step size. In this embodiment, a time sliding window is used to partition the effective feature components to obtain a feature submatrix. Based on the above method, the model's sensitivity to local changes in transient current features can be enhanced, and the sample utilization rate can be improved.

[0060] Step S42: Using each of the feature sub-matrices as input samples, train the initial feature extraction network model to obtain the target feature extraction network model.

[0061] It is easy to understand that in this embodiment, using the feature submatrix obtained by dividing the time window as the input sample of the initial feature extraction network model enables the feature extraction network model to learn local features under different time windows during the training process, thereby enhancing the feature extraction network model's ability to identify subtle differences in transient signals.

[0062] Further, in this embodiment, the step of using each of the feature sub-matrices as input samples to train the initial feature extraction network model to obtain the target feature extraction network model includes: Step S421: Divide the input samples into a training set, a validation set, and a test set; Step S422: Iteratively train the initial feature extraction network model using the training set, and simultaneously evaluate the network performance of the initial feature extraction network model during the training process using the validation set. Step S423: Test and verify the trained initial feature extraction network model using a test set, and use the verified initial feature extraction network model as the target feature extraction network model.

[0063] It should be noted that in this embodiment, the training set refers to the sample set used for iterative updating of model parameters, the validation set refers to the sample set used for real-time evaluation of model performance and hyperparameter tuning during training, and the test set refers to the sample set used for final verification of the generalization ability after the model training is completed.

[0064] It is easy to understand that in this embodiment, the input samples (feature submatrix) can be divided into a training set, a validation set, and a test set according to a specific ratio. These three sets are independent of each other, ensuring the objectivity of the evaluation results, effectively avoiding model overfitting, and ensuring that the model maintains good generalization performance on unseen data. As a preferred approach, the ratio between the training set, validation set, and test set can be 7:2:1.

[0065] It should be noted that, in this embodiment, iterative training refers to the process of updating network parameters multiple times through the backpropagation algorithm; network performance evaluation refers to the process of measuring the model's recognition ability under the current parameters by calculating indicators such as accuracy, precision, and recall; and testing and validation refers to the process of using independent samples that were not involved in training and validation to perform a final performance evaluation of the model in order to verify the model's generalization ability.

[0066] In this embodiment, the training set can be input into the initial feature extraction network model layer by layer. Forward propagation is used to calculate the prediction results, and backpropagation is used to calculate the error (loss function) between the prediction results and the true labels. The connection weights and biases between neurons are updated layer by layer, thus iteratively training the feature extraction network model. After each iteration, a validation set is input into the current feature extraction network model to calculate performance metrics such as recognition accuracy and loss value. Based on the validation results, hyperparameters learned directly from the training data, such as the learning rate, the number of encoder layers, and the number of BiLSTM hidden units, are adaptively adjusted to evaluate the network performance. Once the feature extraction network model is trained or reaches a certain set condition, a test set is input into the model to calculate various performance metrics, such as overall accuracy, precision for each category, and recall. If the recognition accuracy on the test set reaches a very high value, such as 97.78%, the current feature extraction network model is considered to have passed validation and can be used as the target feature extraction network model.

[0067] As a practical example, the 80×6 dimensional feature submatrix obtained after data preprocessing can be used as the input sample for the initial feature extraction network model. To ensure the effectiveness of training, validation, and testing, all samples are randomly divided into training, validation, and test sets in a 7:2:1 ratio. During the partitioning process, the distribution of samples of each category in different datasets is ensured to avoid model training bias caused by sample bias. The sample distribution of each dataset is shown in Table 1 below.

[0068]

[0069] Table 1: Dataset Sample Distribution Table Subsequently, the 180 samples from the test set were input into the trained target feature extraction network model for offline testing to obtain the overall recognition performance index and the detailed performance index of each category of samples.

[0070] Table 2 shows the model's segmentation recognition performance metrics:

[0071] Table 2: Performance Indicators for Subdivision Recognition It should be noted that recall refers to the proportion of all samples that truly belong to a specific category that are correctly identified by the model. The higher the recall, the lower the probability of the model missing a class. The F1 score is the harmonic mean of precision and recall, used to comprehensively evaluate the performance of the model and prevent the model from being biased towards either precision or recall.

[0072] Based on the data shown in Table 2, the precision, recall, and F1 score of the inrush current samples (inrush current samples without residual magnetism) all reached 98.89%, demonstrating the best recognition performance. This indicates that the model can accurately capture the harmonic characteristics and temporal evolution characteristics of the inrush current and effectively distinguish it from the fault current. The indicators of the internal fault samples (single-phase grounding, two-phase short circuit, and two-phase grounding short circuit samples) are all above 97%. Only the recognition index of the three-phase short circuit sample is 95.56%, which is slightly lower than all categories, but it still maintains a high level of accuracy. The main reason is that the transient characteristics of the three-phase short circuit fault current are more complex, and the amplitude changes more violently, which has a relatively high similarity to the characteristics of the inrush current. However, the model can still effectively distinguish them.

[0073] Overall, the target feature extraction network model achieved the following recognition results on the test set: accuracy of 97.78%, precision of 97.65%, recall of 97.52%, and F1 score of 97.58%. All these indicators remained above 97%, demonstrating that the trained target feature extraction network model possesses excellent overall recognition capabilities, effectively distinguishing between inrush current and internal transformer fault current, thus meeting the accuracy requirements for practical engineering applications.

[0074] Table 3 shows the overall recognition performance metrics of the model:

[0075] Table 3: Overall Recognition Performance Indicators It should be noted that three conventional mainstream models in the field of transformer inrush current recognition (BP neural network model, single CNN model, and single LSTM model) were selected for comparison. Comparative experiments were conducted under the same dataset, training parameters, and test metrics, and the overall recognition accuracy of each model is shown in Table 3.

[0076] As shown in Table 3, the traditional BP neural network has the lowest recognition accuracy at only 82.22%, failing to capture the temporal features and global correlations of current sequences. A single CNN model can extract local features but lacks the ability to capture long-range dependencies. A single LSTM model can extract temporal features but cannot effectively capture globally relevant features. The target feature extraction network model trained in this application implements a "global first, temporal second" representation mechanism, resulting in a recognition accuracy improvement of 3.34% to 15.56% compared to conventional models in the field, fully demonstrating the superiority of the target feature extraction network model.

[0077] In addition, the corresponding model confusion matrix can be obtained after offline testing, such as Figure 4 As shown, analysis of the confusion matrix of the test set results revealed that misclassifications mainly occurred between two-phase-to-ground short circuits and three-phase short circuits, and between three-phase short circuits and single-phase-to-ground faults. Specifically, two groups of two-phase-to-ground short circuit samples were misclassified as three-phase short circuits, and one group of three-phase short circuit samples was misclassified as single-phase-to-ground faults.

[0078] It is worth noting that there were zero instances where inrush current samples without excitation were misidentified as internal fault currents; that is, all inrush current samples were correctly identified and not misidentified as fault currents. This characteristic is crucial for transformer differential protection, effectively preventing maloperation of the protection device and demonstrating the practical engineering value of this method.

[0079] Furthermore, in this embodiment, the original transient current sample includes an inrush current sample and a fault current sample, and the step of obtaining the original transient current sample includes: Step S11: Construct a simulation model of the transformer's no-load closing and a simulation model of the fault current. It should be noted that, in this embodiment, the no-load closing simulation model refers to a simulation model used to simulate the no-load closing process of a transformer, and the fault current simulation model refers to a simulation model used to simulate the internal fault process of a transformer. As a specific approach, the Simulink simulation platform can be used to set transformer parameters, simulation conditions, etc., thereby constructing the aforementioned no-load closing simulation model and fault current simulation model.

[0080] As a specific approach, transformer parameters may include a rated capacity of 200MVA, a short-circuit voltage of 10.5kV, a short-circuit loss of 135kW, an no-load loss of 22kW, a voltage ratio of 110 / 11, and a no-load current of 0.8%. Simulation conditions may include closing angle, residual magnetism conditions, and fault types (single-phase grounding, two-phase short circuit, two-phase grounding short circuit, three-phase short circuit), etc.

[0081] Step S12: Generate the inrush current sample using the no-load closing simulation model; It should be noted that, in this embodiment, the inrush current sample refers to the transient current data generated by simulating the transformer no-load closing model under different closing angles and different residual magnetism conditions.

[0082] It is easy to understand that the waveform characteristics of inrush current are closely related to the closing angle and the magnitude of residual magnetism. In this embodiment, by setting the closing angle to a range of 1° to 360°, and considering the presence or absence of residual magnetism, or other transformer parameters, simulation conditions, etc., a sample set covering various inrush current waveforms can be generated, ensuring that the model has good recognition capabilities for inrush current under various operating conditions.

[0083] Step S13: Generate the fault current sample using the fault current simulation model.

[0084] It should be noted that, in this embodiment, the fault current sample refers to transient current data generated by simulating different fault types and different closing angles using the transformer's internal fault model. Specifically, the fault types can include at least four types: single-phase grounding, two-phase short circuit, two-phase ground short circuit, and three-phase short circuit.

[0085] It is easy to understand that in this embodiment, by generating fault current samples under various fault types and closing angles through the fault current simulation model, it is possible to ensure that the model has a good ability to identify various faults and avoid missed detections.

[0086] As an example, such as Figure 5 , Figure 6 as well as Figure 7 As shown, Figure 5 The diagram shows the change of three-phase current over time when a transformer is switched on under no-load conditions with no residual magnetism. Figure 6 The diagram shows the change of three-phase currents over time when the transformer is closed under no-load conditions with 70% residual magnetism. Ia, Ib, and Ic represent the currents of phases A, B, and C, respectively, and t represents time. Figure 7 The diagram shows the changes in three-phase currents (A-phase current, B-phase current, and C-phase current) and three-phase voltages (A-phase voltage, B-phase voltage, and C-phase voltage) over time during a three-phase short-circuit fault inside a transformer.

[0087] This application also provides an inrush current identification device; please refer to... Figure 8 The inrush current identification device includes: Sample acquisition module 10 is used to acquire raw transient current samples; Data processing module 20 is used to perform empirical mode decomposition on the original transient current sample to obtain the feature components corresponding to each frequency domain feature; The data processing module 20 is also used to filter out effective feature components from multiple feature components based on the harmonic component ratio index. The model training module 30 is used to train the initial feature extraction network model using the effective feature components to obtain the target feature extraction network model. The initial feature extraction network model includes a first network unit for extracting the global dependency features in the effective feature components and a second network unit for extracting the temporal evolution features in the global dependency features. The discrimination module 40 is used to identify the current to be identified formed by the transformer through the target feature extraction network model and output the identification result. The identification result is used to characterize whether the current to be identified is an inrush current.

[0088] The inrush current identification device provided in this application, employing the inrush current identification method described in the above embodiments, can solve the technical problem of how to improve the identification accuracy of transformer inrush current and fault current. Compared with the prior art, the beneficial effects of the inrush current identification device provided in this application are the same as those of the inrush current identification method described in the above embodiments, and other technical features in the inrush current identification device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0089] This application provides an inrush current identification device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the inrush current identification method in Embodiment 1 above.

[0090] The following is for reference. Figure 9 The diagram illustrates a structural schematic suitable for implementing the inrush current identification device in the embodiments of this application. The inrush current identification device in the embodiments of this application may include, but is not limited to, fixed terminals such as vehicle-mounted terminals. Figure 9 The inrush current identification device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0091] like Figure 9As shown, the inrush current identification device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 1002 or a program loaded from storage device 1003 into random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the inrush current identification device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following devices can be connected to I / O interface 1006: input devices 1007 including, for example, touch screens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the inrush current identification device to communicate wirelessly or wiredly with other devices to exchange data. Although an inrush current identification device with various devices is shown in the figure, it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed alternatively.

[0092] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0093] The inrush current identification device provided in this application, employing the inrush current identification method in the above embodiments, can solve the technical problem of how to improve the identification accuracy of transformer inrush current and fault current. Compared with the prior art, the beneficial effects of the inrush current identification device provided in this application are the same as those of the inrush current identification method provided in the above embodiments, and other technical features in this inrush current identification device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0094] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0095] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0096] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the inrush current identification method in the above embodiments.

[0097] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor devices, apparatuses, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution apparatus, device, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, radio frequency (RF), etc., or any suitable combination thereof.

[0098] The aforementioned computer-readable storage medium may be included in the inrush current identification device; or it may exist independently and not assembled into the inrush current identification device.

[0099] The aforementioned computer-readable storage medium carries one or more programs. When these programs are executed by the inrush current identification device, the inrush current identification device performs the following actions: acquires an original transient current sample; performs empirical mode decomposition on the original transient current sample to obtain feature components corresponding to each frequency domain feature; filters out effective feature components from the multiple feature components based on the harmonic component ratio index; trains an initial feature extraction network model using the effective feature components to obtain a target feature extraction network model, wherein the initial feature extraction network model includes a first network unit for extracting globally dependent features from the effective feature components and a second network unit for extracting temporal evolution features from the globally dependent features; and identifies the current to be identified formed by the transformer using the target feature extraction network model, outputting an identification result, wherein the identification result characterizes whether the current to be identified is an inrush current.

[0100] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0101] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using dedicated hardware-based apparatus to perform the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0102] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0103] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described inrush current identification method, thereby solving the technical problem of how to improve the identification accuracy of transformer inrush current and fault current. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the inrush current identification method provided in the above embodiments, and will not be repeated here.

[0104] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent scope of this application.

Claims

1. A method for identifying inrush current, characterized in that, The steps of the excitation inrush current identification method include: Obtain the original transient current sample; Empirical mode decomposition is performed on the original transient current sample to obtain the feature components corresponding to each frequency domain feature; Based on the harmonic component proportion index, effective feature components are selected from multiple feature components. The initial feature extraction network model is trained using the effective feature components to obtain the target feature extraction network model. The initial feature extraction network model includes a first network unit for extracting the global dependency features in the effective feature components and a second network unit for extracting the temporal evolution features in the global dependency features. The target feature extraction network model is used to identify the current to be identified formed by the transformer and output the identification result. The identification result is used to characterize whether the current to be identified is an inrush current.

2. The inrush current identification method as described in claim 1, characterized in that, The step of performing empirical mode decomposition on the original transient current sample to obtain the feature components corresponding to each frequency domain feature includes: The original transient current samples are standardized to obtain the original transient current sequence; The original transient current sequence is integrated along the channel dimension to form a multidimensional original feature matrix; The multidimensional original feature matrix is ​​subjected to empirical mode decomposition for each channel to obtain multiple intrinsic mode function components corresponding to each channel. The intrinsic modulus function components of each channel are classified according to their respective frequency domain characteristics to form the characteristic components.

3. The inrush current identification method as described in claim 1, characterized in that, Before the step of filtering out effective feature components from multiple feature components based on the harmonic component proportion index, the method further includes: Obtain the frequency domain spectrum of each of the aforementioned feature components; The amplitudes of the second, third, and fourth harmonics in the frequency domain spectrum are extracted respectively. The ratio of the sum of the amplitudes of the second, third, and fourth harmonics to the total amplitude of the frequency domain spectrum is used as the harmonic component proportion index.

4. The inrush current identification method as described in claim 1, characterized in that, The step of selecting effective feature scores from multiple feature components based on the harmonic component proportion index includes: The feature components whose harmonic component proportion index is not less than a preset threshold are taken as the effective feature components.

5. The inrush current identification method as described in claim 1, characterized in that, The step of training the initial feature extraction network model using the effective feature components to obtain the target feature extraction network model includes: The effective feature components are divided using a sliding time window to obtain multiple feature sub-matrices with the same time window length; Using each of the aforementioned feature sub-matrices as input samples, the initial feature extraction network model is trained to obtain the target feature extraction network model.

6. The inrush current identification method as described in claim 5, characterized in that, The step of using each of the feature sub-matrices as input samples to train the initial feature extraction network model to obtain the target feature extraction network model includes: The input samples are divided into a training set, a validation set, and a test set; The initial feature extraction network model is iteratively trained using the training set, and the network performance of the initial feature extraction network model during the training process is evaluated using the validation set. The initial feature extraction network model trained was tested and validated using a test set, and the validated initial feature extraction network model was used as the target feature extraction network model.

7. The inrush current identification method as described in claim 1, characterized in that, The original transient current sample includes an inrush current sample and a fault current sample. The step of obtaining the original transient current sample includes: Construct a simulation model for no-load closing and a simulation model for fault current of the transformer; The inrush current sample is generated using the no-load closing simulation model. The fault current sample is generated using the fault current simulation model.

8. An inrush current identification device, characterized in that, The inrush current identification device includes: The sample acquisition module is used to acquire raw transient current samples; The data processing module is used to perform empirical mode decomposition on the original transient current sample to obtain the feature components corresponding to each frequency domain feature. The data processing module is also used to filter out effective feature components from multiple feature components based on the harmonic component proportion index. The model training module is used to train the initial feature extraction network model using the effective feature components to obtain the target feature extraction network model. The initial feature extraction network model includes a first network unit for extracting the global dependency features in the effective feature components and a second network unit for extracting the temporal evolution features in the global dependency features. The discrimination module is used to identify the current to be identified formed by the transformer through the target feature extraction network model and output the identification result. The identification result is used to characterize whether the current to be identified is an inrush current.

9. An inrush current identification device, characterized in that, The inrush current identification device includes: a memory, a processor, and an inrush current identification program stored in the memory and executable on the processor, the inrush current identification program being configured to implement the steps of the inrush current identification method as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and the computer-readable storage medium stores an inrush current identification program. When the inrush current identification program is executed by a processor, it implements the steps of the inrush current identification method as described in any one of claims 1 to 7.

Images