Intelligent identification inversion and reproduction method for typical faults of transformer substation

Abstract

The invention discloses an intelligent identification inversion and reproduction method for typical faults of a transformer substation, which comprises the following steps of: calculating a time domain discrete characteristic value by extracting time-frequency domain characteristics of fault recording analog quantity signals; fault recording switching value characteristic parameters including a rising time sequence position coefficient, a falling time sequence position coefficient and an action time proportion coefficient are extracted, and the two types of characteristic values are combined to form an analog-switch fusion characteristic (ASFF); a DC1DCNN-AlexNet model of which the voting weight is optimized by an attention mechanism is constructed; and the two types of ASFF data are respectively input into a DC1DCNN-AlexNet-Attention voting model, and the characteristics of different fault modes are learned and identified by training the DC1DCNN + AlexNet-Attention model. And according to the identified fault type, reproducing a corresponding acousto-optic fault phenomenon.

Description

Technical Field

[0001] The invention belongs to the technical field of substation fault diagnosis, and in particular relates to a method for intelligent identification, inversion and reproduction of typical substation faults. Background Art

[0002] In the operation and maintenance of the power system, substations are important hubs, and the reliability of their equipment is directly related to the safety and stability of the entire power network. However, substations often face the risk of compound faults during operation. This places higher demands on fault detection and processing, and traditional single fault diagnosis methods often fail to meet actual needs. Moreover, after the fault diagnosis is completed, there is no effective way to reproduce the physical phenomenon of the fault as a reference.

[0003] In traditional fault detection methods, most rely on empirical or single-feature-based information analysis, which often cannot meet the needs of accurate identification of complex or compound faults. With the rapid development of artificial intelligence technologies, such as support vector machines, random forests, neural networks, etc., it is possible to learn from multi-dimensional faults and identify the correspondence between fault recordings and specific fault types. Machine learning models not only have strong self-learning capabilities, but also can improve the accuracy and efficiency of fault detection, allowing operation and maintenance personnel to locate and analyze faults more quickly. After successfully detecting a fault, the dynamic characteristics of the fault can be reproduced by building a physical model and performing numerical simulation. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a method for intelligent identification, inversion and reproduction of typical substation faults. The method can effectively use various types of substation fault data to diagnose substation faults and reproduce the physical phenomena when the fault occurs.

[0005] Technical solution: A method for intelligent identification, inversion and reproduction of typical substation faults of the present invention comprises:

[0006] Step 1, extracting the time-frequency domain characteristics of the fault recording analog signal;

[0007] Step 2, extracting the fault recording switch quantity characteristics;

[0008] Step 3: Build and train a 1DCNN model with the time-frequency domain features of the fault recording analog signal as input and the fault type as output, and use the trained 1DCNN model to identify the fault type;

[0009] Step 4: Build and train an AlexNet model with the fault recording switch quantity feature as input and the fault type as output, and use the trained AlexNet model to identify the fault type;

[0010] Step 5: Based on the identification results of the trained 1DCNN model and AlexNet model, a soft voting method is used to obtain the final prediction result of the fault type;

[0011] Step 6: Reproduce the corresponding fault phenomenon according to the final prediction result of the fault type.

[0012] Furthermore, in step 1, extracting the time-frequency domain characteristics of the fault recording analog signal includes: performing empirical mode decomposition on the original signal, and taking the mean, variance, root mean square, peak factor, pulse factor and margin factor of the instantaneous amplitude and instantaneous frequency of each IMF component as the time-frequency domain characteristics of the fault recording analog signal.

[0013] Furthermore, step 1 specifically includes:

[0014] Step 101, using empirical mode decomposition (EMD) to decompose the original signal:

[0015]

[0016] Where X(t) is the original signal, t represents time, c i (t) is the i-th IMF component, r I (t) is the residual component;

[0017] Step 102, perform Hilbert transform on each IMF component:

[0018]

[0019] Definition and c i (t) The associated complex analytic signal z i (t):

[0020]

[0021] Where a i (t) represents c i The amplitude of (t), φ i (t) represents c i The phase of (t);

[0022] Step 103, calculate the instantaneous amplitude and instantaneous frequency of each IMF component:

[0023]

[0024] Where IA i (t) represents c i (t) is the instantaneous amplitude, IF i (t) represents c i (t) the instantaneous frequency;

[0025] Step 104, calculate the mean, variance, root mean square, peak factor, pulse factor and margin factor of the instantaneous amplitude and instantaneous frequency of each IMF component as the time-frequency domain characteristics of the fault recording analog signal.

[0026] Furthermore, in step 2, the fault recording switch quantity characteristics include the rising timing position coefficient, the falling timing position coefficient and the action time proportion coefficient of the switch quantity.

[0027] Furthermore, in step 2, extracting the fault recording switch quantity feature includes:

[0028] Step 201, calculate the jth switch value S j The rising time sequence position coefficient η j1 :

[0029]

[0030] where t j1 Indicates the time of the jth switch rising edge, t b1 and t e1 Respectively represent the time of the first rising edge and the last rising edge of all switching quantities;

[0031] Step 202, calculate the jth switch value S j Falling time position coefficient η j2 :

[0032]

[0033] Where t j2 Indicates the time of the jth switching value falling edge, t b2 and t e2 Respectively represent the time of the first falling edge and the last falling edge of all switching quantities;

[0034] Step 203, calculate the jth switch value S j The action time ratio λ j :

[0035]

[0036] Furthermore, in step 3, the construction of 1DCNN includes:

[0037] Step 301, constructing a one-dimensional convolutional neural network including a convolutional layer, a pooling layer and a fully connected layer;

[0038] Step 302, adding batch normalization between the convolution layer and the pooling layer to speed up the convergence speed;

[0039] Step 303, introduce regularization technology with Dropout layer to reduce overfitting and improve generalization ability.

[0040] Furthermore, in step 5, the soft voting method is used to obtain the final prediction result of the fault type:

[0041] O=α1*O 1DCNN +α2*O AlexN

[0042] In the formula, O represents the final prediction result, O 1DCNN and O AlexN They represent the recognition results of the trained 1DCNN model and AlexNet model, α1 and α2 represent O 1DCNN and O AlexNet The weight of .

[0043] Furthermore, the attention mechanism is used to update the weights α1 and α2:

[0044] α1=softmax(W1*O 1DCNN +b1)

[0045] α2=softmax(W2*O AlexN +b2)

[0046] W1 and W2 are learnable parameters, b1 and b2 are bias terms, and are updated in real time as training progresses.

[0047] The present invention also provides a computer-readable storage medium storing one or more programs, wherein the one or more programs include instructions, which, when executed by a computing device, cause the computing device to perform the method as described above.

[0048] The present invention also provides an electronic device, comprising one or more processors, one or more memories and one or more programs, wherein the one or more programs are stored in the one or more memories and are configured to be executed by the one or more processors, and the one or more programs include instructions for executing the method as described above.

[0049] Compared with the prior art, the present invention has the following significant advantages:

[0050] (1) The present invention utilizes the particularity that the fault switch quantity data only contains two states, "0" and "1", and the conversion between the two states is instantaneous action without an intermediate state. The three parameters of the switch quantity data, namely the rising sequential position coefficient (RSPC), the falling sequential position coefficient (FSPC) and the action time proportion coefficient (ATPC), are extracted and combined with the time-frequency domain characteristics of the fault analog signal to construct the analog-switch fusion feature (ASFF) to fully reflect the information when the substation fault occurs.

[0051] (2) The present invention constructs a dual-channel DC1DCNN-AlexNet-Attention model, which consists of two independent feature processing modules and a shared decision module, and can effectively fuse and process analog-switch fusion features (ASFF) to accurately perform multi-category classification tasks for various types of substation faults.

[0052] (3) The present invention performs principal component analysis (PCA) on the two types of data input by DC1DCNN-AlexNet-Attention to reduce the dimension of the data and screen key features to optimize the input of the model; compared with the traditional model, the voting mechanism of this model can avoid the overfitting problem that may exist in a single model and accelerate the convergence speed of the model. The generalization ability of the model is improved, and the optimization of the attention mechanism is adopted before the output of the model to update the optimal parameters in real time. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 The figure is a flow chart of the substation fault diagnosis method of the present invention.

[0054] Figure 2 This is the main wiring diagram of the simulated substation of the present invention.

[0055] Figure 3 This is a flow chart of fault recording switch quantity feature extraction of the present invention.

[0056] Figure 4 This is a simplified time axis for the switch quantity model of the present invention.

[0057] Figure 5 This is the structural diagram of 1DCNN.

[0058] Figure 6 This is the structural diagram of AlexNet.

[0059] Figure 7 These are the accuracy curve and loss function curve of the model training set and validation set.

[0060] Figure 8 This is the structural diagram of the multi-dimensional comprehensive display system of sound, light and electricity.

[0061] Fig. 9 This is an example of a physical reproduction phenomenon. DETAILED DESCRIPTION

[0062] In order to make the objectives, technical solutions and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0063] like Figure 1The figure shows a typical substation fault intelligent identification, inversion and reproduction method described in this embodiment, which includes the following steps:

[0064] Step 1: extract the time-frequency domain features of the fault recording analog signal. First, perform empirical mode decomposition on the signal, then perform Hilbert transform on the decomposed signal, and finally calculate the time-domain discrete eigenvalues.

[0065] Specifically, the time-frequency domain features of the fault recording analog signal are extracted including:

[0066] Step 101, using empirical mode decomposition (EMD) to decompose the original signal to obtain multiple IMF components,

[0067]

[0068] Where X(t) is the original signal, t represents time, c i (t) is the i-th IMF component, r I (t) is the residual component.

[0069] Step 102, after all intrinsic mode functions (IMFs) are derived, Hilbert transform is performed on each IMF.

[0070]

[0071] This can be defined as the signal c i (t) The associated complex analytic signal z i (t),

[0072]

[0073] Where τ represents the integral variable, a i (t) represents c i The amplitude of (t), φ i (t) represents c i The phase of (t).

[0074] Step 103, calculate the instantaneous amplitude (IA) and instantaneous frequency (IF) of each IMF,

[0075]

[0076] Step 104, calculate the mean, variance, root mean square, peak factor, pulse factor and margin factor of the instantaneous amplitude IA and instantaneous frequency IF of each IMF component as the time-frequency domain characteristics of the fault recording analog signal.

[0077] In signal processing, mean, variance, and RMS are important parameters that describe the statistical characteristics of a signal. They are usually used to analyze characteristics such as amplitude distribution, stability, and energy of a signal.

[0078] The mean is the average value of all values ​​in a data set, which can measure the average level of the data. The mean of a signal is the average amplitude of the signal in the time domain, and the square of the mean is the static component of the signal.

[0079] Variance is defined as the mean of the sum of the squares of the differences between each data point in the sample set and its mean. It represents the dynamic component of the signal energy and reflects the volatility of the signal by quantifying the deviation between the signal amplitude and the mean.

[0080] The root mean square (RMS) is the square root of the variance, also known as the effective value. It provides another measure of the fluctuation of the signal amplitude, and the unit is the same as the original unit of the signal. The three dimensional eigenvalues ​​of the signal in the time domain are closely related and provide a basis for signal analysis. The mean provides the average amplitude of the signal; the variance measures the dispersion of the signal amplitude relative to the mean and is used to evaluate the stability of the signal; the root mean square value is used to measure the energy of the signal.

[0081] Peak factor, pulse factor and margin factor are three common dimensionless characteristic values ​​of signals in the time domain. They are often used to describe the extreme degree of the peak value of the signal in the waveform, and can well reflect the change and severity of the impact current when the substation line is short-circuited.

[0082] The peak factor is defined as the ratio of the peak value of a signal to the root mean square value (RMS). It is used to measure the difference between the peak value of a signal and the average level of the signal. It represents the extreme degree of the peak value of the signal in the waveform. The larger the peak factor, the more significant the peak value of the signal.

[0083] Impulse Factor is defined as the ratio of the peak value of a signal to its rectified average value, where the rectified average value is the average value of the absolute value of the signal. The difference between it and the peak factor lies mainly in the denominator. Since the rectified average value of the same signal is less than the effective value (RMS), the impulse factor is greater than the peak factor.

[0084] The margin factor is defined as the ratio of the peak value of the signal to the root square amplitude. The root square amplitude corresponds to the root mean square value (RMS) of the signal and can be calculated by the square of the square root mean of the signal. Therefore, the margin factor corresponds to the kurtosis factor.

[0085] Among them, the specific expressions of mean, variance, root mean square, peak factor, pulse factor and margin factor are shown in Table 1.

[0086] Table 1 Analog signal characteristic parameters

[0087]

[0088] Where x[n] represents the amplitude of the nth sampling point of IA or IF, μ is the average value of the sampling points, and N is the total number of samples.

[0089] Step 2: Extract the fault recording switch characteristic parameters: rising sequence position coefficient (RSPC), falling sequence position coefficient (FSPC) and action time proportion coefficient (ATPC), and combine the two types of characteristic quantities to form an analog-switch fusion feature (ASFF).

[0090] Specifically, the jth switch quantity S j The rising time sequence position coefficient η j1 for:

[0091]

[0092] where t j1 Indicates the time of the jth switch rising edge, t b1 and t e1 Respectively represent the time of the first rising edge and the last rising edge of all switching quantities;

[0093] The jth switch quantity S j Falling time position coefficient η j2 for:

[0094]

[0095] Where t j2 Indicates the time of the jth switching value falling edge, t b2 and t e2 Respectively represent the time of the first falling edge and the last falling edge of all switching quantities;

[0096] Calculate the jth switch quantity S j The action time ratio λ j for:

[0097]

[0098] The two types of feature quantities in steps 1 and 2 above are combined to form an analog-switch fusion feature (ASFF). Based on ASFF, a mixed data set containing various types and modes of faults is constructed to comprehensively evaluate the model's ability to identify and classify different line fault types. The data set covers a variety of fault types that may be encountered in substations, including single-phase grounding, two-phase short circuit, two-phase grounding, and three-phase short circuit faults, ensuring that the model can handle the diversity in practical applications. On this basis, by setting different fault parameters, adding metallic and non-metallic categories, the distance factor of the fault point from the substation, the fault duration, current transformer saturation and other conditions, the adaptability and robustness of the model to the complex and changeable conditions encountered in the actual operation of the substation are comprehensively evaluated.

[0099] Step 3: Build the DC1DCNN-AlexNet model, including:

[0100] Build and train a 1DCNN model with the time-frequency domain features of the fault recording analog signal as input and the fault type as output, and use the trained 1DCNN model to identify the fault type;

[0101] Build and train the AlexNet model with fault recording switch quantity features as input and fault type as output, and use the trained AlexNet model to identify the fault type;

[0102] The two types of feature quantities of ASFF are respectively input into the identification results of the trained 1DCNN model and AlexNet model, and the soft voting method is used to obtain the final prediction result of the fault type.

[0103] Specifically, the soft voting method is used to obtain the final prediction result of the fault type:

[0104] O=α1*O 1D +α2*O AlexN

[0105] In the formula, O represents the final prediction result, O 1DCNN and O AlexNet They represent the recognition results of the trained 1DCNN model and AlexNet model, α1 and α2 represent O 1DCN and O AlexN The weight of .

[0106] Furthermore, the attention mechanism is used to update the weights α1 and α2:

[0107] α1=softmax(W1*O CNN +b1)

[0108] α2=softmax(W2*O AlexNet +b2)

[0109] W1 and W2 are learnable parameters, b1 and b2 are bias terms, and they are updated in real time as the training progresses.

[0110] Step 5: Reproduce the corresponding acoustic, optical and electrical fault phenomenon according to the identified fault type.

[0111] In an example, a typical 220KV substation in a city is selected. The main electrical wiring diagram of the substation is as follows: Figure 2 As shown in the figure, the main transformer adopts a three-phase auto-coupling on-load tap-changing power transformer with a connection group of YNa0d11; the 220KV partial outgoing line is two-circuit, using double busbar connection; the 110KV partial outgoing line is two-circuit, using double busbar connection; the 10KV partial outgoing line is two-circuit, using single busbar segmented connection; the station transformer capacity is 350kVA. According to the main electrical connection diagram of the substation, the DDRTS digital dynamic real-time simulation system is used to build the simulation circuit model of the substation. Construct the substation virtual secondary protection system. First, obtain the data results of the primary modeling calculation from the DDRTS electromagnetic transient simulation platform interface, process the basic data through Fourier transform and the real algorithm of the protection device, input the voltage and current with time stamp into the virtual digital protection device model, and simulate the operation of the virtual protection device in actual production through the startup criterion of the virtual protection device combined with the logic circuit of the protection device. Based on the simulation model of the primary system and the secondary protection system mentioned above, the fault setting is carried out, and the line fault data set, bus fault data set and transformer fault data set are obtained after running the simulation. Substation failure can be divided into three stages: electrical quantity change, protection device action, and circuit breaker tripping. When the fault is triggered, the voltage, current and other electrical quantities of the faulty component will change suddenly, and then the relay protection device related to the faulty component will analyze and judge the fault symptom electrical quantity monitored to make the protection action, and then send a tripping signal to the corresponding circuit breaker to drive the circuit breaker to trip, and isolate the faulty component from the substation to eliminate the fault. By setting the simulation operation to run different parameters of the line, including the transition resistance, fault point location and fault duration, different types of line faults can be obtained. Since there are thousands of simulation files, after the simulation is completed, all ".cfg" and ".dat" fault recording files are saved and renamed in the form of numbers plus suffixes. When Python imports these files, it reads the file name in the form of "path + number + suffix" string concatenation, and traverses all files by building a loop function, parsing and converting the data format before saving it in a table form.

[0112] Step 2, extract the time-frequency domain features of the fault recording analog signal. First, perform adaptive complete empirical mode decomposition on the signal, then perform Hilbert transform on the decomposed signal, and finally calculate the time-domain discrete eigenvalues.

[0113] Specifically, assuming that each protection action and switch quantity change twice within a period of time, the start and end time nodes of all switch quantity actions are arranged on the time axis in the order of action change, such as Figure 3 As shown, t j1 Indicates the time when the jth switch starts to act, that is, the time node of the rising edge, t j2 Indicates its end action, that is, the time node of the falling edge, t b1 (t b2 ) and t e1 (t e2 ) respectively represent the time nodes of the first rising edge (falling edge) and the last rising edge (falling edge) of all switching quantities. Thus, the jth switching quantity S is defined j The rising time sequence position coefficient η j1 and the decreasing timing position coefficient η j2 The expressions are as follows:

[0114]

[0115] From the above formula, we can see that η j1 The first rising edge of all switching values ​​t b1 Distance S j Rising edge t j1 The length of time is the same as the last rising edge t e1 The ratio of time length, similarly η i2 It can be understood as the first falling edge t of all switching quantities b2 Distance S j Falling edge t j2 The length of time is the same as the last falling edge t e2 The ratio of time length. The earlier the switch starts, the smaller the rising timing position coefficient RSPC is, and vice versa; the earlier the end action time is, the smaller the falling timing position coefficient FSPC is, and vice versa.

[0116] Step 202, calculate the jth switch value S j Action time proportion coefficient (ATPC)λ j It indicates the ratio of the total duration of the action to the entire protection time. The expression is as follows:

[0117]

[0118] Since some switching quantities have two-stage opening and closing actions, their RSPC, FSPC and ATPC coefficients need to be calculated twice respectively. Therefore, each switching quantity has 6 feature dimensions, and the other three coefficients of the switching quantity that only acts once are defined as 0. Secondly, for some switching quantities with continuous actions, the time node t j2is unknown, so the time node of the last falling edge is t e2 Set as the simulation end time, based on the assumption that there is still 1 unit of time after the simulation ends, and the states of all switch quantities at this time point are 0.

[0119] Step 203, the seven characteristic parameter values ​​of the analog signal are combined with the three characteristic quantities of RSPC, FSPC and ATPC of the switch quantity to form an analog-switch fusion feature (ASFF). Figure 4 shown.

[0120] Specifically, the implementation process of step 3 includes:

[0121] Step 301, build a 1DCNN model and an AlexNet model; the 1DCNN channel is used to process the analog features of the fault recorder ASFF. It starts from an input layer and receives the feature vector after PCA dimension reduction. 64 convolution kernels are used to extract higher-level features. Batch normalization (BN) is added between the convolution layer and the pooling layer of the traditional 1D-CNN to speed up the convergence of the model. The structural diagram of the 1DCNN model is shown in Figure 5 As shown in the figure. The AlexNet channel is used to process the switch quantity features of the fault recording ASFF. Multiple convolutional layers and pooling layers are introduced to perform multiple dimensionality reduction processing on the input features. Regularization technology with a Dropout layer is introduced to reduce overfitting. The structural diagram of the AlexNet model is shown in the figure. Figure 6 shown.

[0122] Step 302: Connect the output layers of the 1DCNN model and the AlexNet model to the input of the attention mechanism, and optimize their voting weights by the attention mechanism. Set the initial weights α1 and a2 to 0.5. During the training process, the output results of each training are voted according to the weights. The attention mechanism updates the weights one by one according to the training labels to optimize the recognition ability of the model. The output of 1DCNN is O 1DCnn , the output of AlexNet is O Ale , then the output O of DC1DCNN-AlexNet is

[0123] O=α1*O 1DCN +α2*O AlexNet

[0124] Use the attention mechanism to update weights α1, α2:

[0125] α1=softmax(W1*O 1DCNN +b1)

[0126] α2=softmax(W2*O AlexNet +b2))

[0127] Among them, W1 and W2 are learnable parameters, b1 and b2 are bias terms, which are updated in real time as the training progresses.

[0128] Specifically, the implementation process of step 4 includes:

[0129] Step 401, construct a mixed data set containing various types and modes of faults to comprehensively evaluate the model's ability to identify and classify different line fault types. The data set covers a variety of fault types that may be encountered in substations, including single-phase grounding, two-phase short circuit, two-phase grounding, and three-phase short circuit faults, ensuring that the model can handle the diversity in practical applications. On this basis, by setting different fault parameters, adding metallic and non-metallic categories, factors such as the distance between the fault point and the substation, fault duration, current transformer saturation, etc., to comprehensively evaluate its adaptability and robustness to complex and changing situations encountered during the actual operation of the substation.

[0130] Step 402, the two types of ASFF data are input into the two input channels of the DC1DCNN-AlexNet-Attention model for training and verification. The training and verification results are shown in Figure 7 The dataset was trained 5 times independently, and the four evaluation indicators were calculated for each training. The results are shown in Table 2.

[0131] Table 2 Evaluation indicators

[0132] Precision Recall Accuracy F1 score 1 97.11％ 96.19％ 95.83％ 95.80％ 2 94.50％ 94.31％ 93.33％ 93.15％ 3 95.21％ 95.71％ 96.67％ 95.00％ 4 95.62％ 95.01％ 95.00％ 94.38％ 5 96.05％ 93.73％ 93.75％ 93.86％

[0133] Specifically, the implementation process of step 5 includes:

[0134] Step 501, after the fault type is identified, the fault type is input into the sound, light and electricity multi-dimensional integrated display system. The structure of the sound, light and electricity multi-dimensional integrated display system is as follows: Figure 8 shown.

[0135] Step 502, the sound, light and electricity multi-dimensional integrated display system inputs the fault type into the physical reproduction module, analyzes the physical phenomenon of the fault, completes the inversion reproduction of the fault, and reproduces the physical phenomenon of the corresponding fault. Fig. 9 shown.

[0136] Based on the same technical solution, the present invention also discloses a computer-readable storage medium storing one or more programs, wherein the one or more programs include instructions, and when the instructions are executed by a computing device, the computing device executes the above-mentioned intelligent identification, inversion and reproduction method of typical substation faults.

[0137] Based on the same technical solution, the present invention also discloses a computing device, including one or more processors, one or more memories and one or more programs, wherein the one or more programs are stored in the one or more memories and are configured to be executed by the one or more processors, and the one or more programs include instructions for executing the above-mentioned substation typical fault intelligent identification, inversion and reproduction method.

[0138] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0139] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0140] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0141] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0142] The above embodiments are only for illustrating the technical idea of ​​the present invention, and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the present invention.

Claims

1. A method for intelligent identification, inversion and reproduction of typical faults in substations, characterized in that: The method comprises: Step 1, extracting the time-frequency domain characteristics of the fault recording analog signal; Step 2, extracting the fault recording switch quantity characteristics; Step 3: Build and train a 1DCNN model with the time-frequency domain features of the fault recording analog signal as input and the fault type as output, and use the trained 1DCNN model to identify the fault type; Step 4: Build and train an AlexNet model with the fault recording switch quantity feature as input and the fault type as output, and use the trained AlexNet model to identify the fault type; Step 5: Based on the identification results of the trained 1DCNN model and AlexNet model, a soft voting method is used to obtain the final prediction result of the fault type; Step 6: Reproduce the corresponding fault phenomenon according to the final prediction result of the fault type.

2. The method for intelligent identification, inversion and reproduction of typical substation faults according to claim 1 is characterized in that: In step 1, extracting the time-frequency domain features of the fault recording analog signal includes: performing empirical mode decomposition on the original signal, and taking the mean, variance, root mean square, peak factor, pulse factor and margin factor of the instantaneous amplitude and instantaneous frequency of each IMF component as the time-frequency domain features of the fault recording analog signal.

3. The method for intelligent identification, inversion and reproduction of typical substation faults according to claim 1 or 2, characterized in that: Step 1 specifically includes: Step 101, using empirical mode decomposition (EMD) to decompose the original signal: Where X(t) is the original signal, t represents time, c i (t) is the i-th IMF component, r I (t) is the residual component; Step 102, perform Hilbert transform on each IMF component: Definition and c i (t) The associated complex analytic signal z i (t): Where a i (t) represents c i The amplitude of (t), φ i (t) represents c i The phase of (t); Step 103, calculate the instantaneous amplitude and instantaneous frequency of each IMF component: Where IA i (t) represents c i (t) is the instantaneous amplitude, IF i (t) represents c i (t) the instantaneous frequency; Step 104, calculate the mean, variance, root mean square, peak factor, pulse factor and margin factor of the instantaneous amplitude and instantaneous frequency of each IMF component as the time-frequency domain characteristics of the fault recording analog signal.

4. The method for intelligent identification, inversion and reproduction of typical substation faults according to claim 1 is characterized in that: In step 2, the fault recording switch quantity characteristics include the rising timing position coefficient, the falling timing position coefficient and the action time proportion coefficient of the switch quantity.

5. The method for intelligent identification, inversion and reproduction of typical substation faults according to claim 1 is characterized in that: In step 2, extracting the fault recording switch quantity features includes: Step 201, calculate the jth switch value S j The rising time sequence position coefficient η j1 : where t j1 Indicates the time of the jth switch rising edge, t b1 and t e1 Respectively represent the time of the first rising edge and the last rising edge of all switching quantities; Step 202, calculate the jth switch value S j Falling time position coefficient η j2 : Where t j2 Indicates the time of the jth switching value falling edge, t b2 and t e2 Respectively represent the time of the first falling edge and the last falling edge of all switching quantities; Step 203, calculate the jth switch value S j The action time ratio λ j :

6. The method for intelligent identification, inversion and reproduction of typical substation faults according to claim 1 is characterized in that: In step 3, the construction of 1DCNN includes: Step 301, constructing a one-dimensional convolutional neural network including a convolutional layer, a pooling layer and a fully connected layer; Step 302, adding batch normalization between the convolution layer and the pooling layer to speed up the convergence speed; Step 303, introduce regularization technology with Dropout layer to reduce overfitting and improve generalization ability.

7. The method for intelligent identification, inversion and reproduction of typical substation faults according to claim 1 is characterized in that: In step 5, the soft voting method is used to obtain the final prediction result of the fault type: O=α1*O 1DCN +α2*O AlexNet In the formula, O represents the final prediction result, O 1DCNN and O AlexN They represent the recognition results of the trained 1DCNN model and AlexNet model, α1 and α2 represent O 1DCNN and O AlexNet The weight of .

8. The method for intelligent identification, inversion and reproduction of typical substation faults according to claim 7 is characterized in that: Update weights α1 and α2 using the attention mechanism: α1=softmax(W1*O 1DCNN +b1) <h2 style=";text-align:left;direction:ltr">α2 = softmax(W2*O<h2 style=";text-align:left;direction:ltr"> AlexNet <h2 style=";text-align:left;direction:ltr"> +b2) W1 and W2 are learnable parameters, b1 and b2 are bias terms, and they are updated in real time as the training progresses.

9. A computer-readable storage medium storing one or more programs, wherein the one or more programs include instructions, characterized in that: When the instructions are executed by a computing device, the computing device is caused to perform the method according to any one of claims 1 to 8.

10. An electronic device, characterized in that: The method comprises one or more processors, one or more memories and one or more programs, wherein the one or more programs are stored in the one or more memories and are configured to be executed by the one or more processors, and the one or more programs include instructions for executing the method as claimed in any one of claims 1 to 8.

Images