CN-121980383-A - Power electronic transformer working mode identification method, device, equipment, medium and product

CN121980383ACN 121980383 ACN121980383 ACN 121980383ACN-121980383-A

Abstract

The application discloses a method, a device, equipment, a medium and a product for identifying the working mode of a power electronic transformer, and relates to the field of artificial intelligence; the method comprises the steps of carrying out self-adaptive segmentation normalization processing on original inductor current data to generate a normalized inductor current sequence, calculating wavelet packet energy entropy and time-domain differential entropy of the normalized inductor current sequence, splicing the wavelet packet energy entropy and the time-domain differential entropy into two-dimensional fusion feature vectors, inputting the normalized inductor current sequence and the two-dimensional fusion feature vectors into a trained deep learning model to obtain a predictive probability vector, and selecting a working mode type with the maximum probability value as a recognition result based on the predictive probability vector.

Inventors

LI YONGJIAN
LI XIANGYU
YANG KEWEI
YANG YI
WEI YOUXI

Assignees

浙江江山变压器股份有限公司

Dates

Publication Date: 20260505
Application Date: 20260120

Claims (10)

1. The power electronic transformer working mode identification method is characterized by comprising the following steps of: Collecting original inductance current data of the power electronic transformer during operation; performing self-adaptive segmentation normalization processing on the original inductor current data to generate a normalized inductor current sequence; calculating wavelet packet energy entropy and time-domain differential entropy of a normalized inductive current sequence, and splicing the wavelet packet energy entropy and the time-domain differential entropy into a two-dimensional fusion feature vector; inputting the normalized inductor current sequence and the two-dimensional fusion feature vector into a trained deep learning model to obtain a predictive probability vector; and selecting the working mode category with the largest probability value as a recognition result based on the predictive probability vector.
2. The method for identifying a working mode of a power electronic transformer according to claim 1, wherein the performing adaptive segmentation normalization processing on the raw inductor current data to generate a normalized inductor current sequence specifically includes: Calculating a local mean value and a local standard deviation of an original inductance current value of a current sampling point in a sliding window mode; Calculating a mutation enhancement factor based on the absolute difference value of the original inductance current value of the current sampling point and the previous sampling point; and (3) subtracting the local mean value from the original inductance current value, and carrying out normalization processing by combining the local standard deviation and the mutation enhancement factor.
3. The method for identifying an operation mode of a power electronic transformer according to claim 1, wherein calculating a wavelet packet energy entropy and a time-domain differential entropy of a normalized inductor current sequence and splicing the wavelet packet energy entropy and the time-domain differential entropy into a two-dimensional fusion feature vector comprises: Carrying out wavelet packet decomposition on the normalized inductance current sequence, calculating the normalized energy duty ratio of each sub-band, and obtaining the wavelet packet energy entropy through an information entropy formula; Calculating normalized gradient probability of adjacent sampling points in the normalized inductance current sequence, and obtaining the time-domain differential entropy through an information entropy formula; And splicing the wavelet packet energy entropy and the time-domain differential entropy to form the two-dimensional fusion feature vector.
4. The method for identifying the working mode of the power electronic transformer according to claim 1, wherein the deep learning model comprises: The first feature extraction branch is used for processing the normalized inductor current sequence by adopting a one-dimensional depth separable convolution network with a dynamic convolution kernel and outputting a time domain feature matrix, wherein the dynamic convolution kernel is dynamically generated based on the two-dimensional fusion feature vector; The second feature extraction branch is used for carrying out high-dimensional mapping on the two-dimensional fusion feature vector by adopting a full-connection layer, and outputting an enhanced feature vector; The feature fusion layer is used for splicing the time domain feature matrix and the enhanced feature vector to obtain a final fusion feature; The mode feature decoupling layer is used for inhibiting the commonality information among different working mode categories from the final fusion features, enhancing the degree of distinction among the working modes and outputting the feature vectors after decoupling; And the pattern recognition classification layer is used for outputting a prediction probability vector of each working pattern class of the original inductance current data based on the decoupled characteristic vector.
5. The method for identifying the working mode of the power electronic transformer according to claim 1 or 4, wherein the training mode of the deep learning model is as follows: Randomly extracting batch samples from the training set, and sequentially executing forward propagation calculation prediction probability; calculating a loss value of the current batch through a loss function; obtaining gradients of the loss values on trainable parameters of each layer of the deep learning model by using a back propagation algorithm; adopting an Adam optimizer to update parameters to minimize loss in combination with the adaptive learning rate; performing iterative training through the steps, in the training process, evaluating the recognition accuracy rate in a verification set after each iteration, starting an early-stopping mechanism when the continuous multi-round accuracy rate is not improved, and finally reserving model parameters with optimal performance to obtain a trained deep learning model; Wherein, the training set and the verification set are both from an inductor current data set which is built in advance.
6. The method for identifying an operation mode of a power electronic transformer according to claim 5, wherein the method for calculating the adaptive learning rate is as follows: calculating a local signal to noise ratio through the absolute maximum value of the normalized inductor current sequence and the noise standard deviation; Mapping the local signal-to-noise ratio into a learning rate adjustment coefficient based on a Sigmoid function; Multiplying the learning rate adjustment coefficient by the initial learning rate to obtain the self-adaptive learning rate.
7. A power electronic transformer operating mode identification device, characterized in that the power electronic transformer operating mode identification device comprises: the data acquisition unit is used for acquiring original inductance current data when the power electronic transformer operates; the preprocessing unit is used for carrying out self-adaptive segmentation normalization processing on the original inductor current data to generate a normalized inductor current sequence; the characteristic splicing unit is used for calculating wavelet packet energy entropy and time-domain differential entropy of the normalized inductive current sequence and splicing the wavelet packet energy entropy and the time-domain differential entropy into a two-dimensional fusion characteristic vector; the probability prediction unit is used for inputting the normalized inductive current sequence and the two-dimensional fusion feature vector into a trained deep learning model to obtain a prediction probability vector; And the result output unit is used for selecting the working mode category with the maximum probability value as the identification result based on the prediction probability vector.
8. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor executes the computer program to implement the steps of the power electronic transformer operation pattern recognition method of any one of claims 1-6.
9. A computer-readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, carries out the steps of the power electronic transformer operating mode identification method of any one of claims 1-6.
10. A computer program product comprising a computer program, characterized in that the computer program, when being executed by a processor, realizes the steps of the power electronic transformer operating mode identification method of any one of claims 1-6.

Description

Power electronic transformer working mode identification method, device, equipment, medium and product Technical Field The application relates to the technical field of artificial intelligence, in particular to a method, a device, equipment, a medium and a product for identifying a working mode of a power electronic transformer. Background Along with the rapid development of the power system in the intelligent, high-efficiency and flexible directions, the power electronic transformer (Power Electronic Transformer, PET) gradually becomes an important component of novel power equipment due to the advantages of small size, flexible regulation, high functional integration level and the like. The PET realizes voltage conversion and energy transmission through high-frequency conversion, and the core structure of the PET comprises a plurality of working modes, such as Buck (Buck), boost (Boost), LLC (resonant) and the like, and the modes can be dynamically switched according to load change, power grid fluctuation or control strategies in actual operation. The current working mode is accurately and real-timely identified, and the method is a key premise for ensuring the stable operation of the system and realizing self-adaptive control and quick fault response. However, the operation environment of the PET system is generally accompanied by high-frequency switching noise, transient disturbance and multi-working condition switching, and the conventional recognition method based on the fixed threshold value or static characteristic rule has difficulty in meeting the dual requirements of recognition accuracy and response speed. Meanwhile, most of the existing data driving identification methods are based on deep learning, the collected original current data is subjected to simple normalization processing, and transient abrupt change characteristics related to mode switching in the original data are easy to weaken. Meanwhile, most methods directly utilize the normalized current sequence as the only input of the deep learning model. However, such a single current sequence can only reflect the time-series variation of the signal, failing to provide a statistical description of the signal as a whole in terms of energy distribution or variation law. Therefore, the features learned from such input by the deep learning model are difficult to clearly distinguish between the operation modes of Buck, boost, LLC and the like which are different in nature, and particularly, the recognition accuracy is easy to be reduced under the working conditions of noise interference or rapid mode switching. Disclosure of Invention The application aims to provide a method, a device, equipment, a medium and a product for identifying the working mode of a power electronic transformer, which can ensure the identification precision of the working mode of the power electronic transformer under the working condition of noisy high interference or rapid mode switching. In order to achieve the above object, the present application provides the following solutions: In a first aspect, the present application provides a method for identifying an operating mode of a power electronic transformer, including: Collecting original inductance current data of the power electronic transformer during operation; performing self-adaptive segmentation normalization processing on the original inductor current data to generate a normalized inductor current sequence; calculating wavelet packet energy entropy and time-domain differential entropy of a normalized inductive current sequence, and splicing the wavelet packet energy entropy and the time-domain differential entropy into a two-dimensional fusion feature vector; inputting the normalized inductor current sequence and the two-dimensional fusion feature vector into a trained deep learning model to obtain a predictive probability vector; and selecting the working mode category with the largest probability value as a recognition result based on the predictive probability vector. Optionally, the performing adaptive segmentation normalization processing on the original inductor current data to generate a normalized inductor current sequence specifically includes: Calculating a local mean value and a local standard deviation of an original inductance current value of a current sampling point in a sliding window mode; Calculating a mutation enhancement factor based on the absolute difference value of the original inductance current value of the current sampling point and the previous sampling point; and (3) subtracting the local mean value from the original inductance current value, and carrying out normalization processing by combining the local standard deviation and the mutation enhancement factor. Optionally, the calculating the wavelet packet energy entropy and the time-domain differential entropy of the normalized inductive current sequence, and splicing the wavelet packet energy entropy and the time-domain different