A method and system for predicting the error state of a DC voltage transformer

Abstract

This invention relates to a method and system for predicting the error state of a DC voltage transformer, belonging to the field of online power metering and monitoring technology. The method includes: acquiring error data of the target DC voltage transformer; segmenting the error data using a two-order sliding window; extracting features from the time-series matrix using Fast Fourier Transform to obtain time-varying and time-invariant features; fusing the time-varying and time-invariant features using a multilayer perceptron to obtain fused features; and inputting the time-varying, time-invariant, and fused features into an error prediction model based on deep learning to obtain the predicted error state of the target DC voltage transformer. This invention improves the adaptability, accuracy, and stability of DC voltage transformer error prediction through the synergistic effect of adaptive windowing, feature extraction and fusion, and sliding value correction, providing more reliable technical support for online monitoring and anomaly early warning of power systems.

Description

Technical Field

[0001] This invention relates to the field of online power metering and monitoring technology, and mainly to a method and system for predicting the error status of a DC voltage transformer. Background Technology

[0002] DC voltage transformers, as core equipment for voltage measurement and signal conversion in power systems, play a crucial role in DC power transmission and new energy grid integration. In DC transmission systems, the measurement accuracy of DC voltage transformers directly affects power calculation, fault diagnosis, and system control effectiveness. If abnormal errors are not detected in time, they can lead to metering deviations, protection malfunctions, and even large-scale power outages. Therefore, real-time and accurate prediction of their errors is of significant practical importance.

[0003] In existing technologies, prediction methods for DC voltage transformer errors mainly focus on traditional statistical models (such as regression analysis and time series models) and early machine learning methods (such as support vector machines and backpropagation neural networks). However, these methods have obvious limitations: First, the data processing method is rigid: traditional methods often use a fixed-length sliding window to segment the error data, which cannot adapt to the dynamic changes in the error data. When error data exhibits high-frequency fluctuations or abrupt trend changes, a fixed window can easily lead to excessive redundant information or loss of key features, reducing data utilization. Secondly, feature extraction is insufficient: existing technologies for feature mining of error data often remain at the time domain level, or only extract spectral features through simple Fourier transforms, failing to effectively distinguish between stable time-invariant features (such as inherent low-frequency characteristics of equipment) and dynamic time-varying features (such as high-frequency noise and sudden interference), resulting in insufficient feature representation capabilities. Thirdly, the predictive model has poor adaptability: traditional models use simple feature fusion methods (such as direct splicing), making it difficult to fully utilize the complementarity of time-varying and time-invariant features; simultaneously, they lack a dynamic adjustment mechanism based on prediction bias, and when the prediction result deviates significantly from the actual error, they cannot correct model parameters or window strategies in real time, leading to a decrease in prediction accuracy.

[0004] Therefore, there is an urgent need for an error prediction method that can adaptively adjust the data window, accurately extract multi-dimensional features, and dynamically optimize the prediction strategy. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention proposes a method and system for predicting the error state of a DC voltage transformer.

[0006] The technical solution of the present invention is as follows:

[0007] On one hand, this invention proposes a method for predicting the error state of a DC voltage transformer, the method comprising:

[0008] Obtain the error data of the target DC voltage transformer;

[0009] The error data is segmented using a two-order sliding window to obtain the time series matrix;

[0010] The time series matrix is ​​subjected to fast Fourier transform to extract features, resulting in time-varying and time-invariant features;

[0011] A multilayer perceptron is used to fuse time-varying and time-invariant features to obtain fused features;

[0012] By inputting time-varying features, time-invariant features, and fused features into an error prediction model built on deep learning, the error prediction state of the target DC voltage transformer is obtained.

[0013] Preferably, the two-order sliding window includes a first-order segmentation and a second-order segmentation, wherein the first-order segmentation specifically involves calculating the information entropy and autocorrelation matrix of the error data to obtain the first-order sliding window value, wherein:

[0014] The information entropy of the error data is calculated using the following formula:

[0015] ;

[0016] ;

[0017] In the formula, Representing error data Information entropy; This indicates that the error data belongs to the first... The probability of each error interval; Indicates the first The index value of each error interval; Indicates the first One error data point; Indicates the first One error data point; Indicates the length of the error data; Indicates the first The index value of each error data;

[0018] The autocorrelation matrix of the error data is calculated using the following formula:

[0019] ;

[0020] ;

[0021] In the formula, Representing error data The autocorrelation matrix; Expressing expectations; Representing error data The conjugate transpose of; Representing error data The complex conjugate; Indicates the transpose operation;

[0022] The value of the first-order sliding window is obtained by the following formula:

[0023] ;

[0024] In the formula, This represents the value of the first-order sliding window; This indicates the preset first-order fixed length; This represents a pre-defined first-order constant; This represents the floor function; The sign that indicates positive or negative;

[0025] The first-order segmentation of the error data based on the first-order sliding window value is specifically as follows: ... Each error data point is divided into... A length of The first-order subsequence, where , This indicates the number of first-order subsequences.

[0026] Preferably, the second-order segmentation specifically involves adjusting the second-order sliding window value based on the accumulated error, as expressed by the formula:

[0027] ;

[0028] ;

[0029] ;

[0030] In the formula, Indicates the value of the second-order sliding window; express Error prediction deviation at any given time; express Error prediction status at any given time; express The actual error at time; express The cumulative amount of error at any given time; express The cumulative amount of error at any given time; Indicates the error weight; This indicates a preset second-order fixed length; This represents a pre-defined second-order constant;

[0031] Each length is The first-order subsequence is further divided into A length of The second-order subsequence;

[0032] Combine all second-order subsequences into a time series matrix.

[0033] Preferably, the time series matrix is ​​used to extract features using the Fast Fourier Transform, as expressed by the formula:

[0034] ;

[0035] ;

[0036] In the formula, Represents time-invariant features; Indicates time-varying characteristics; Represents a convolutional neural network; Represents the filter function; Represents the Fast Fourier Transform function; Represents a time series matrix; Indicates the dominant spectrum; Indicates other spectrum.

[0037] The preferred method for calculating the dominant spectrum and other spectra is expressed by the following formula:

[0038] ;

[0039] ;

[0040] ;

[0041] In the formula, This represents the amplitude corresponding to the frequency of each sampling point in the time series matrix; The function representing the mean; This represents the amplitude extraction function; Indicates taking the first Greater than or equal to The spectral frequency is used as the dominant spectrum; Indicates taking the first Less than The spectral frequency is used as other spectra; This represents the preset spectral constant.

[0042] Preferably, a multilayer perceptron is used to fuse time-varying features and time-invariant features, specifically as follows:

[0043] Perform tensor operations on the time-invariant features to obtain the result of the first tensor operation.

[0044] The result of the first tensor operation is input into the multilayer perceptron to obtain the output of the multilayer perceptron; the output of the multilayer perceptron and the time-varying features are then subjected to tensor operations to obtain the result of the second tensor operation.

[0045] The result of the second tensor operation is mapped through an activation function to obtain the fused features.

[0046] Preferably, it also includes determining whether the error prediction deviation at the current moment exceeds a preset deviation threshold;

[0047] If it exceeds, then based on the error prediction deviation, the second fixed length... Second order constant The correction is made, and the second-order sliding window value is recalculated based on the corrected second-order fixed length and second-order constant.

[0048] On the other hand, the present invention also proposes an error state prediction system for a DC voltage transformer, the system comprising:

[0049] The data acquisition module is used to acquire the error data of the target DC voltage transformer; and transmit the error data to the data segmentation module;

[0050] The data segmentation module is used to segment error data using a two-order sliding window to obtain a time series matrix;

[0051] The feature extraction module is used to extract features from the time series matrix using the Fast Fourier Transform to obtain time-varying features and time-invariant features;

[0052] The feature fusion module is used to fuse time-varying features and time-invariant features using a multilayer perceptron to obtain fused features;

[0053] The error prediction module is used to input time-varying features, time-invariant features, and fused features into the error prediction model built based on deep learning to obtain the error prediction state of the target DC voltage transformer.

[0054] In another aspect, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in the present invention.

[0055] In another aspect, the present invention also provides a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the method described in the present invention.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] 1. This invention provides a method and system for predicting the error state of a DC voltage transformer. Through an adaptive two-order sliding window mechanism, the first-order sliding window dynamically adjusts the window value based on the information entropy of the error data (measuring data uncertainty) and the autocorrelation matrix (reflecting data correlation); the second-order window dynamically adjusts the window parameters based on the real-time error prediction deviation and cumulative amount, combined with a feedback mechanism to correct the window parameters. This improves the adaptability to time-varying and non-stationary error data, enhances the ability to capture key features (such as long-term trends and short-term fluctuations) at different time scales, and improves the accuracy and utilization of data segmentation.

[0058] 2. This invention provides a method and system for predicting the error state of a DC voltage transformer. It extracts features by fusing the frequency domain and time domain, uses fast Fourier transform to convert the error data to the frequency domain, and combines convolutional neural network to extract the dominant spectrum and other spectra. This achieves the separation of stable features and dynamic features, breaks through the limitations of traditional methods that rely only on time domain features or single spectrum features, and fully explores the multi-dimensional characteristics of error data.

[0059] 3. This invention provides a method and system for predicting the error state of a DC voltage transformer. Based on a multilayer perceptron, tensor operations and activation function mapping are performed on time-varying and time-invariant features. In conjunction with deep learning models such as LSTM, three types of features (time-varying, time-invariant, and fused) are jointly learned to capture the complex correlation between features. This solves the problem of loss of feature correlation caused by traditional feature fusion, improves the efficiency of complementary utilization of different features, and improves the accuracy of error prediction state judgment. Attached Figure Description

[0060] Figure 1 This is a detailed flowchart of an embodiment of the present invention. Detailed Implementation

[0061] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] It should be understood that the step numbers used in the text are for ease of description only and are not intended to limit the order in which the steps are performed.

[0063] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0064] The terms “comprising” and “including” indicate the presence of the described feature, whole, step, operation, element and / or component, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or collections thereof.

[0065] The term “and / or” refers to any combination of one or more of the associated listed items, as well as all possible combinations, and includes these combinations.

[0066] Example 1:

[0067] See details Figure 1 This invention provides a method for predicting the error state of a DC voltage transformer, the method comprising:

[0068] S1. Obtain the error data of the target DC voltage transformer;

[0069] S2. The error data is segmented using a two-order sliding window; the two-order sliding window includes a first-order segmentation and a second-order segmentation.

[0070] S21. The first-order segmentation specifically involves calculating the information entropy and autocorrelation matrix of the error data to obtain the first-order sliding window value, where:

[0071] The information entropy of the error data is calculated using the following formula:

[0072] ;

[0073] ;

[0074] In the formula, Representing error data Information entropy; This indicates that the error data belongs to the first... The probability of each error interval; Indicates the first The index value of each error interval; Indicates the first One error data point; Indicates the first One error data point; Indicates the length of the error data; Indicates the first The index value of each error data;

[0075] The autocorrelation matrix of the error data is calculated using the following formula:

[0076] ;

[0077] ;

[0078] In the formula, Representing error data The autocorrelation matrix; Expressing expectations; Representing error data The conjugate transpose of; Representing error data The complex conjugate; Indicates the transpose operation;

[0079] The value of the first-order sliding window is obtained by the following formula:

[0080] ;

[0081] In the formula, This represents the value of the first-order sliding window; This indicates the preset first-order fixed length; This represents a pre-defined first-order constant; This represents the floor function; The sign that indicates positive or negative;

[0082] When the information entropy is lower than the preset information entropy threshold and the autocorrelation is stronger than the preset autocorrelation threshold, it means that the first-order sliding window value will increase, and the calculation formula for the first-order sliding window value takes the "+" sign; conversely, it means that the first-order sliding window value will decrease, and the calculation formula for the first-order sliding window value takes the "-" sign.

[0083] The first-order segmentation of the error data based on the first-order sliding window value is specifically as follows: ... Each error data point is divided into... A length of The first-order subsequence is represented as ,in , Indicates the number of first-order subsequences;

[0084] If the length of the last first-order subsequence is insufficient If there are 1 or more, pad with 0 to the length. ;

[0085] S22. The second-order segmentation specifically involves adjusting the second-order sliding window value based on the accumulated error, expressed by the formula:

[0086] ;

[0087] ;

[0088] ;

[0089] In the formula, Indicates the value of the second-order sliding window; express Error prediction deviation at any given time; express Error prediction status at any given time; express The actual error at time; express The cumulative amount of error at any given time; express The cumulative amount of error at any given time; Indicates the error weight; This indicates a preset second-order fixed length; This represents a pre-defined second-order constant;

[0090] when 'time' represents the initial moment, at which point... This indicates the preset error prediction state, which is the error prediction state at the initial moment.

[0091] Each length is The first-order subsequence is further divided into A length of The second-order subsequence is represented as ;

[0092] in This indicates that the first-order subsequence The second-order subsequences formed after segmentation, and so on. This indicates that the first-order subsequence The second-order subsequence formed after segmentation;

[0093] The combination of all second-order subsequences into a time series matrix is ​​represented as follows: ;

[0094] If the length of the last second-order subsequence is insufficient If there are 1 or more, pad with 0 to the length. ;

[0095] S23, also includes determining whether the error prediction deviation at the current moment exceeds a preset deviation threshold;

[0096] If it exceeds, then based on the error prediction deviation, the second fixed length... Second order constant The correction is made, and the second-order sliding window value is recalculated based on the corrected second-order fixed length and second-order constant.

[0097] In this embodiment, the correction process for the second-order fixed length and the second-order constant is expressed by the following formula:

[0098] ;

[0099] ;

[0100] In the formula, This indicates the corrected second-order fixed length; Indicates the correction factor; This represents the corrected second-order constant; Indicates the error prediction deviation;

[0101] S3. Use the Fast Fourier Transform to extract features from the time series matrix to obtain time-varying and time-invariant features, expressed by the following formula:

[0102] ;

[0103] ;

[0104] In the formula, Represents time-invariant features; Indicates time-varying characteristics; Represents a convolutional neural network; Represents the filter function; Represents the Fast Fourier Transform function; Represents a time series matrix; Indicates the dominant spectrum; Indicates other spectrum;

[0105] The calculation process for the dominant spectrum and other spectra is expressed by the following formula:

[0106] ;

[0107] ;

[0108] ;

[0109] In the formula, This represents the amplitude corresponding to the frequency of each sampling point in the time series matrix; The function representing the mean; This represents the amplitude extraction function; Indicates taking the first The largest spectral frequency is used as the dominant spectrum; This indicates taking spectral frequencies other than the dominant spectrum, i.e., other spectrum frequencies; This represents the preset spectral constant;

[0110] Time-invariant features refer to relatively stable, slowly changing, or persistent features in error data, corresponding to dominant and stable frequency components in the data, specifically including slowly changing features such as low-frequency signals and the inherent vibration frequency of equipment.

[0111] Time-varying characteristics refer to the dynamic changes and rapid fluctuations in error data, corresponding to minor and unstable frequency components in the data, specifically including short-term fluctuations such as high-frequency noise and sudden interference.

[0112] The error of a DC voltage transformer is mainly affected by harmonics and noise, and it changes over time. Therefore, time-varying characteristics can be obtained based on other spectra. The dominant spectrum, on the other hand, is the original signal itself, which does not change much. Therefore, time-invariant characteristics can be obtained based on the dominant spectrum.

[0113] S4. Use a multilayer perceptron to fuse time-varying features and time-invariant features to obtain fused features;

[0114] Perform tensor operations on the time-invariant features to obtain the result of the first tensor operation.

[0115] The result of the first tensor operation is input into the multilayer perceptron to obtain the output of the multilayer perceptron; the output of the multilayer perceptron and the time-varying features are then subjected to tensor operations to obtain the result of the second tensor operation.

[0116] The result of the second tensor operation is mapped using an activation function;

[0117] The fusion features are obtained, expressed by the formula:

[0118] ;

[0119] In the formula, Indicates fusion characteristics; Indicates the activation function; This represents a multilayer perceptron; This represents the Hadamard product, used for tensor operations;

[0120] S5. Input the time-varying features, time-invariant features and fused features into the error prediction model built based on deep learning to obtain the error prediction state of the target DC voltage transformer, wherein the error prediction state includes normal state and abnormal state.

[0121] If the error prediction status is greater than the preset error prediction threshold, it indicates an abnormal state and triggers an alarm; otherwise, it indicates a normal state.

[0122] In this embodiment, an LSTM model is selected as the error prediction model to obtain the error prediction state of the target DC voltage transformer, which is expressed by the formula:

[0123] ;

[0124] In the formula, Indicates the error prediction status; Represents the LSTM model;

[0125] In one specific embodiment, a DC fast charging station deploys a DC voltage transformer (PT) to monitor the 750V DC bus voltage at a sampling frequency of 1kHz;

[0126] Obtain error data ={0.12, 0.15, 0.18, 0.22, 0.25, 0.31, 0.28, 0.26, 0.24, 0.23};

[0127] set up =0.5, =2, =1, =1, =0.7, =0.8;

[0128] The first-order segmentation is performed as follows:

[0129] Divide the error into three equally spaced intervals: [0.10, 0.17), [0.17, 0.24), and [0.24, 0.31].

[0130] Calculate the information entropy of the error data:

[0131] = [0.2×log2(2 / 10)+0.3×log2(3 / 10)+0.5×log2(5 / 10)]=1.57;

[0132] Calculate the autocorrelation matrix of the error data =0.00286;

[0133] because =1.57 < preset information entropy threshold of 1.5 and =0.00286 > the preset autocorrelation threshold of 0.002. At this point, taking the "+" sign yields the first-order sliding window value. =5;

[0134] because =6, resulting in the first-order subsequences {0.12, 0.15, 0.18, 0.22, 0.25}, {0.15, 0.18, 0.22, 0.25, 0.31}, {0.15, 0.18, 0.22, 0.25, 0.31}...{0.28, 0.26, 0.24, 0.23, 0};

[0135] The second-order segmentation is performed as follows:

[0136] Calculate the error prediction deviation =0.29-0.31=-0.02;

[0137] Calculate the cumulative error =0.7×0.01+0.3×0.02=0.013;

[0138] Calculate the second-order sliding window value =1+(0.013 / 0.01)×1=3;

[0139] Take the first-order subsequence {0.12, 0.15, 0.18, 0.22, 0.25}, divide it into subsequences of length 3 to obtain the second-order subsequences {0.12, 0.15, 0.18}, {0.15, 0.18, 0.22}, {0.18, 0.22, 0.25}, and so on to obtain all the second-order subsequences;

[0140] Perform a Fast Fourier Transform (FFT) on each second-order subsequence: FFT([0.12, 0.15, 0.18]) = [-0.45, -0.015 + 0.026j, -0.015 - 0.026j];

[0141] Calculate the amplitude corresponding to the frequency of each sampling point in the time series matrix. =[0.45, 0.028, 0.029], since 0.45 2 / (0.45) 2 +0.028 2 +0.029 2 Since 0.998 > 0.8, the dominant spectrum is [0.45], and the other spectra are [0.028, 0.029].

[0142] Obtain time-invariant features =0.44, time-varying characteristic is =0.028;

[0143] Obtain fusion features =0.5006;

[0144] Obtain the error prediction state =0.73;

[0145] judge If the error prediction threshold of 0.73 is greater than the preset error prediction threshold of 0.7, it indicates that the current DC voltage transformer is in an abnormal state, triggering an early warning.

[0146] Example 2:

[0147] This embodiment provides an error state prediction system for a DC voltage transformer, the system comprising:

[0148] The data acquisition module is used to acquire the error data of the target DC voltage transformer; and transmit the error data to the data segmentation module;

[0149] The data segmentation module is used to segment error data using a two-order sliding window to obtain a time series matrix;

[0150] The feature extraction module is used to extract features from the time series matrix using the Fast Fourier Transform to obtain time-varying features and time-invariant features;

[0151] The feature fusion module is used to fuse time-varying features and time-invariant features using a multilayer perceptron to obtain fused features;

[0152] The error prediction module is used to input time-varying features, time-invariant features, and fused features into the error prediction model built based on deep learning to obtain the error prediction state of the target DC voltage transformer.

[0153] Example 3:

[0154] This embodiment proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the method described in any embodiment of the present invention.

[0155] Example 4:

[0156] This embodiment proposes a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the method described in any embodiment of the present invention.

[0157] In this application embodiment, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, A and B simultaneously, or B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of singular or plural items. For example, at least one of a, b, and c can represent: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, and c can be single or multiple.

[0158] Those skilled in the art will recognize that the units and algorithm steps described in the embodiments disclosed herein can be implemented using electronic hardware, computer software, or a combination of electronic hardware and software. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0159] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0160] In the several embodiments provided in this application, any function, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0161] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for predicting the error state of a DC voltage transformer, characterized in that, The method includes: Obtain the error data of the target DC voltage transformer; The error data is segmented using a two-order sliding window to obtain the time series matrix; The two-order sliding window includes a first-order segmentation and a second-order segmentation, wherein the first-order segmentation specifically involves calculating the information entropy and autocorrelation matrix of the error data to obtain the first-order sliding window value, wherein: The information entropy of the error data is calculated using the following formula: ； ； In the formula, Representing error data Information entropy; This indicates that the error data belongs to the first... The probability of each error interval; Indicates error data; Indicates the first The index value of each error interval; Indicates the first One error data point; Indicates the first One error data point; Indicates the length of the error data; Indicates the first The index value of each error data; The autocorrelation matrix of the error data is calculated using the following formula: ； ； In the formula, Representing error data The autocorrelation matrix; Expressing expectations; Representing error data The conjugate transpose of; Representing error data The complex conjugate; Indicates the transpose operation; The value of the first-order sliding window is obtained by the following formula: ； In the formula, This represents the value of the first-order sliding window; This indicates the preset first-order fixed length; This represents a pre-defined first-order constant; This represents the floor function; The sign that indicates positive or negative; The first-order segmentation of the error data based on the first-order sliding window value is specifically as follows: ... Each error data point is divided into... A length of The first-order subsequence, where , Indicates the number of first-order subsequences; The second-order segmentation specifically involves adjusting the second-order sliding window value based on the accumulated error, as expressed by the formula: ； ； ； In the formula, Indicates the value of the second-order sliding window; express Error prediction deviation at any given time; express Error prediction status at any given time; express The actual error at time; express The cumulative amount of error at any given time; express The cumulative amount of error at any given time; Indicates the error weight; This indicates a preset second-order fixed length; This represents a pre-defined second-order constant; Each length is The first-order subsequence is further divided into A length of The second-order subsequence; Combine all second-order subsequences into a time series matrix; The time series matrix is ​​subjected to fast Fourier transform to extract features, resulting in time-varying and time-invariant features; A multilayer perceptron is used to fuse time-varying and time-invariant features to obtain fused features; By inputting time-varying features, time-invariant features, and fused features into an error prediction model built on deep learning, the error prediction state of the target DC voltage transformer is obtained.

2. The error state prediction method for a DC voltage transformer according to claim 1, characterized in that, Feature extraction of the time series matrix is ​​performed using the Fast Fourier Transform, expressed by the following formula: ； ； In the formula, Represents time-invariant features; Indicates time-varying characteristics; Represents a convolutional neural network; Represents the filter function; Represents the Fast Fourier Transform function; Represents a time series matrix; Indicates the dominant spectrum; Indicates other spectrum.

3. The error state prediction method for a DC voltage transformer according to claim 2, characterized in that, The calculation process for the dominant spectrum and other spectra is expressed by the following formula: ； ； ； In the formula, This represents the amplitude corresponding to the frequency of each sampling point in the time series matrix; The function representing the mean; This represents the amplitude extraction function; Indicates taking the first Greater than or equal to The spectral frequency is used as the dominant spectrum; Indicates taking the first Less than The spectral frequency is used as other spectra; This represents the preset spectral constant.

4. The error state prediction method for a DC voltage transformer according to claim 1, characterized in that, The time-varying and time-invariant features are fused using a multilayer perceptron, specifically as follows: Perform tensor operations on the time-invariant features to obtain the result of the first tensor operation. The result of the first tensor operation is input into the multilayer perceptron to obtain the output of the multilayer perceptron; the output of the multilayer perceptron and the time-varying features are then subjected to tensor operations to obtain the result of the second tensor operation. The result of the second tensor operation is mapped through an activation function to obtain the fused features.

5. The error state prediction method for a DC voltage transformer according to claim 1, characterized in that, It also includes determining whether the error prediction deviation at the current moment exceeds a preset deviation threshold; If it exceeds, then based on the error prediction deviation, the second fixed length... Second order constant The correction is made, and the second-order sliding window value is recalculated based on the corrected second-order fixed length and second-order constant.

6. An error state prediction system for a DC voltage transformer, characterized in that, The system includes: The data acquisition module is used to acquire the error data of the target DC voltage transformer; and transmit the error data to the data segmentation module; The data segmentation module is used to segment error data using a two-order sliding window to obtain a time series matrix; The two-order sliding window includes a first-order segmentation and a second-order segmentation, wherein the first-order segmentation specifically involves calculating the information entropy and autocorrelation matrix of the error data to obtain the first-order sliding window value, wherein: The information entropy of the error data is calculated using the following formula: ； ； In the formula, Representing error data Information entropy; This indicates that the error data belongs to the first... The probability of each error interval; Indicates error data; Indicates the first The index value of each error interval; Indicates the first One error data point; Indicates the first One error data point; Indicates the length of the error data; Indicates the first The index value of each error data; The autocorrelation matrix of the error data is calculated using the following formula: ； ； In the formula, Representing error data The autocorrelation matrix; Expressing expectations; Representing error data The conjugate transpose of; Representing error data The complex conjugate; Indicates the transpose operation; The value of the first-order sliding window is obtained by the following formula: ； In the formula, This represents the value of the first-order sliding window; This indicates the preset first-order fixed length; This represents a pre-defined first-order constant; This represents the floor function; The sign that indicates positive or negative; The first-order segmentation of the error data based on the first-order sliding window value is specifically as follows: ... Each error data point is divided into... A length of The first-order subsequence, where , Indicates the number of first-order subsequences; The second-order segmentation specifically involves adjusting the second-order sliding window value based on the accumulated error, as expressed by the formula: ； ； ； In the formula, Indicates the value of the second-order sliding window; express Error prediction deviation at any given time; express Error prediction status at any given time; express The actual error at time; express The cumulative amount of error at any given time; express The cumulative amount of error at any given time; Indicates the error weight; This indicates a preset second-order fixed length; This represents a pre-defined second-order constant; Each length is The first-order subsequence is further divided into A length of The second-order subsequence; Combine all second-order subsequences into a time series matrix; The feature extraction module is used to extract features from the time series matrix using the Fast Fourier Transform to obtain time-varying features and time-invariant features; The feature fusion module is used to fuse time-varying features and time-invariant features using a multilayer perceptron to obtain fused features; The error prediction module is used to input time-varying features, time-invariant features, and fused features into the error prediction model built based on deep learning to obtain the error prediction state of the target DC voltage transformer.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements an error state prediction method for a DC voltage transformer as described in any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements an error state prediction method for a DC voltage transformer as described in any one of claims 1 to 5.

Images