A method and system for identifying and tracing power grid panoramic waveform faults based on space-time spectrum S network

Abstract

The present invention discloses a method and system for identifying and tracing the panoramic waveform fault of a power grid based on a space-time spectrum S network. The present invention includes calculating multiple characteristic values ​​according to the panoramic waveform data and splicing them to obtain fused characteristic values; inputting the time domain signal of the fused characteristic value into the trained space-time spectrum S network to obtain the fault type and fault location. The space-time spectrum S network includes a multi-level feature extraction module, a fully connected layer and a multi-task learning component. The feature extraction module includes a sequentially connected time-frequency spectrum module, a convolution layer and a pooling layer. The time-frequency spectrum module includes a convolution layer, a global feature extraction branch, a local time-frequency feature extraction branch and a feature fusion module. The features extracted by the feature extraction module are sent to the multi-task learning component through the fully connected layer to obtain the fault type and fault location of the power grid. The present invention aims to improve the accuracy and positioning capability of fault monitoring of new power systems and solve the problem that new power grid faults are difficult to identify and trace under new power systems.

Description

Technical Field

[0001] The present invention relates to the technical field of power system information measurement and data processing, and in particular to a method and system for identifying and tracing power grid panoramic waveform faults in a space-time spectrum S network. Background Art

[0002] In the new power system, the large-scale access of new energy sources such as wind, solar and storage not only brings more adjustable resources to the power system, but also injects more high-frequency disturbances and new faults into the power system. Due to the interaction of power electronic devices, this new type of fault not only includes traditional disturbance types, such as generator cutting and load shedding, but also includes some new faults, such as broadband oscillation, photovoltaic power station cutting and high-frequency switching faults. However, due to the limitations of traditional fault detection technology, traditional methods are difficult to accurately apply to complex time-varying new disturbance faults, and cannot simultaneously realize fault identification and location analysis. In response to the difficulties of fault detection in new power systems, existing methods mainly monitor faults based on synchronized phasor data from wide-area monitoring systems. This type of method is mainly data-driven, such as relative phase angle displacement and recurrent neural network methods, but this type of method has the problems of low data resolution and insufficient monitoring accuracy; another type of method is the graph neural network method, which can effectively combine the topological relationship of the power system for fault location analysis, but this method needs to rely heavily on the structural information of the power system and is difficult to cope with complex and changeable distribution networks. Moreover, relying only on low-resolution synchronized phasor data, the above methods are unable to simultaneously achieve classification and location monitoring of power grid faults. Summary of the invention

[0003] Technical problem to be solved by the present invention: In view of the above-mentioned problems in the prior art, a method and system for panoramic waveform fault identification and tracing of a power grid based on a space-time spectrum S network is provided. The present invention aims to improve the accuracy and positioning capability of fault monitoring in new power systems, and solve the problem that new power grid faults are difficult to identify and trace under new power systems.

[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0005] A method for identifying and tracing power grid panoramic waveform faults based on a time-space spectrum S network comprises the following steps:

[0006] S1, obtain the panoramic waveform data obtained by measuring the grid waveform , , is the number of sampling points;

[0007] S2, based on panoramic waveform data Calculate multiple eigenvalues ​​and concatenate them to obtain fused eigenvalues , the multiple characteristic values ​​include part or all of incremental modulation statistics, root mean square values ​​and waveform energy values;

[0008] S3, the fusion feature value The time domain signal is input into the trained space-time spectrum S network to obtain the fault type and fault location. The space-time spectrum S network is pre-trained to establish a fusion feature value The mapping relationship between the time domain signal, fault type and fault location; the spatiotemporal spectrum S network includes a multi-level feature extraction module, a fully connected layer and a multi-task learning component, the feature extraction module includes a sequentially connected time-frequency spectrum module, a convolution layer and a pooling layer, the time-frequency spectrum module includes a convolution layer, a global feature extraction branch, a local time-frequency feature extraction branch and a feature fusion module, the fusion feature value The time domain signal is sent to the convolution layer for convolution operation and then sent to the global feature extraction branch to extract global features through S transform and to the local time-frequency feature extraction branch to extract local time-frequency features through fast Fourier transform. The extracted global features and local time-frequency features are sent to the feature fusion module for feature fusion. The fused features are sent to the multi-task learning component through the fully connected layer to obtain the fault type and fault location of the power grid.

[0009] Optionally, the fusion feature value is obtained by splicing in step S2 The calculation function expression is:

[0010] ,

[0011] in, is the incremental modulation statistic, is the root mean square value, is the waveform energy value; the calculation function expression of the incremental modulation statistical value in step S2 is:

[0012] ,

[0013] In the above formula, is the incremental modulation statistic, is the next waveform value, is the current waveform value, Indicates the number of measurement nodes. represents the number of times the incremental modulation is executed; the calculation function expression of the waveform energy value in step S2 is:

[0014]

[0015] In the above formula, is the waveform energy value, is the window length, For time point The square of the waveform value, For displacement The window function at , Represents a convolution operation.

[0016] Optionally, the function expression of the convolution layer in step S3 is:

[0017] ,

[0018] In the above formula, is the signal obtained by the convolution operation of the convolution layer at time t, is the convolution kernel, for The fusion eigenvalue at the moment The time domain signal, For time shift.

[0019] Optionally, the global feature extraction branch extracts the global feature including: firstly performing an S-transform on the signal obtained by the convolution operation of the convolution layer according to the following formula:

[0020] ,

[0021] In the above formula, is the result of S transformation, is the signal obtained by the convolution operation of the convolution layer at time t, for Time frequency The Gaussian window function is are time shift and frequency respectively, is an imaginary unit, is time, and the Gaussian window function is defined as:

[0022] ,

[0023] In the above formula, for Time frequency Gaussian window function; the amplitude of the result obtained by S transformation Constructed with dimensions A two-dimensional matrix, where are the channels, width and length of the input features respectively, and the dimension is The width of the two-dimensional matrix Compressed to one dimension while maintaining the same shape, the dimension is The two-dimensional matrix is ​​processed using the enhanced block Inception to refine and filter features to obtain local time-frequency features .

[0024] Optionally, the local time-frequency feature extraction branch extracts the local time-frequency feature, including: firstly performing a fast Fourier transform on the signal obtained by the convolution operation of the convolution layer according to the following formula:

[0025] ,

[0026] In the above formula, is the feature obtained by fast Fourier transform, is the length of the input signal, the length of the input signal and the length of the input feature d equal, for The discrete form of is an imaginary unit, For time, ; Then select the one with the highest amplitude based on the features obtained by fast Fourier transform The characteristic amplitude obtained by fast Fourier transform ,Sure The amplitude of the softmax normalization function Corresponding , using the period coefficient To adjust the input The size of Periodic coefficient ,in for The discrete form of for Round up, For the selected frequency, ~ They are No. 1~ period coefficients, using The period coefficient will Discrete form of Split into length The block is adjusted to a two-dimensional spectrogram, filled with zeros to match the dimension of the segmented series, and then processed with the enhanced block Inception to refine and filter the features to obtain global features. .

[0027] Optionally, the function expression of the feature fusion module is:

[0028] ,

[0029] In the above formula, is the fused feature, is a global feature, is the local time-frequency feature, is a trainable parameter, To adjust the hyperparameters of the ratio of global features to local time-frequency features, Indicates the amplitude The softmax normalization function of is the characteristic amplitude obtained by fast Fourier transform.

[0030] Optionally, the fully connected layer includes two output linear layers corresponding to the fault detection task and the fault location task respectively, and the function expressions of the two output linear layers are:

[0031] ,

[0032] ,

[0033] In the above formula, and are the output results of the two output linear layers, and Represent two output linear layers, and are the weights of the two output linear layers, is the input to the two output linear layers, and are biases of two output linear layers respectively; the multi-task learning component includes two task prediction modules for the fault detection task and the fault location task respectively, for predicting the fault according to the output results of the two output linear layers and Mapped to output vector of fault type and fault location and , the loss function used in the training of the spatiotemporal spectrum S network is:

[0034] ,

[0035] In the above formula, is the loss function, is the network parameter of the space-time spectrum S network, and are the importance weight coefficients of the fault detection task and the fault location task, respectively. and represent the true probability and predicted probability respectively, For penalty items.

[0036] In addition, the present invention also provides a power grid panoramic waveform fault identification and tracing system of a space-time spectrum S network, including an interconnected microprocessor and a memory, wherein the microprocessor is programmed or configured to execute the power grid panoramic waveform fault identification and tracing method of the space-time spectrum S network.

[0037] In addition, the present invention also provides a computer-readable storage medium, which stores a computer program or instruction, and the computer program or instruction is programmed or configured to execute the power grid panoramic waveform fault identification and tracing method of the space-time spectrum S network through a processor.

[0038] In addition, the present invention also provides a computer program product, including a computer program or instructions, which are programmed or configured to execute the grid panoramic waveform fault identification and tracing method of the space-time spectrum S network through a processor.

[0039] Compared with the prior art, the present invention mainly has the following advantages: the present invention makes full use of the space-time spectrum to target the extraction of key features and location information of the panoramic information of the power grid, and constructs a multi-task space-time spectrum S network, specifically including panoramic information sampling of the electrical parameters of each measuring node of the active distribution network; in order to fully extract the panoramic information of multiple nodes of the power grid, a space-time spectrum mechanism based on fast Fourier transform and S transform of convolution operation is constructed, and the key features of the space-time spectrum of fault information are extracted in combination with weight fusion; in combination with the panoramic information and the location of the measurement node, a multi-task learning component based on the dynamic weight loss function is built to realize the panoramic information fault tracing; the space-time spectrum mechanism is regarded as a layer in multi-task learning, and an end-to-end space-time spectrum S network is constructed through multi-layer superposition to realize the synchronous identification and tracing of panoramic information faults. The power grid panoramic waveform fault identification and tracing method of the space-time spectrum S network of this embodiment realizes the characteristics of simplicity, multi-function, fast response and high precision, has important engineering practice significance, and solves the problem that new power grid faults are difficult to identify and trace under new power systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 Schematic diagram of the network structure of the space-time spectrum S network in an embodiment of the present invention.

[0041] Figure 2 FIG. 4 is a schematic diagram of a network structure of a time spectrum module in an embodiment of the present invention.

[0042] Figure 3 It is a panoramic waveform diagram of PV tripping under the IEEE 33 bus system in an embodiment of the present invention.

[0043] Figure 4 It is a panoramic waveform effective value diagram of PV tripping under the IEEE 33bus system in an embodiment of the present invention.

[0044] Figure 5 This is a visualization result diagram of an example of extracting deep features in an embodiment of the present invention.

[0045] Figure 6 This is a graph showing the time-frequency spectrum information result of extracting depth features in an embodiment of the present invention.

[0046] Figure 7 This is a graph showing the weight coefficient results of fault detection in an embodiment of the present invention.

[0047] Figure 8 This is a result diagram of the weight coefficient of the fault location task importance in an embodiment of the present invention.

[0048] Fig. 9 Graph showing the fault detection and positioning performance results under different noise levels in an embodiment of the present invention.

[0049] Fig.10 Graph showing the fault detection and location performance results under different training set ratios in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] like Figure 1 and Figure 2 As shown, the method for identifying and tracing the panoramic waveform fault of the power grid of the spatiotemporal spectrum S network in this embodiment includes the following steps: S1, obtaining the panoramic waveform data obtained by measuring the power grid waveform , , is the number of sampling points; S2, based on the panoramic waveform data Calculate multiple eigenvalues ​​and concatenate them to obtain fused eigenvalues The multiple characteristic values ​​include incremental modulation statistics, root mean square value and waveform energy value; S3, the fusion characteristic value The time domain signal is input into the trained space-time spectrum S network to obtain the fault type and fault location. The space-time spectrum S network is pre-trained to establish a fusion feature value The mapping relationship between the time domain signal, fault type and fault location. The space-time spectrum S network refers to a convolutional network model that fuses the S transform and the fast Fourier transform in the signal processing method into an embeddable layer. The parameters and data dimensions of the embeddable layer are trained and changed together with the convolutional network model and are portable, which facilitates the targeted extraction of time-space domain and other space-time features of the input fault signal. In this embodiment, the space-time spectrum S network includes a multi-level feature extraction module, a fully connected layer and a multi-task learning component, the feature extraction module includes a time-frequency spectrum module, a convolutional layer and a pooling layer connected in sequence, the time-frequency spectrum module includes a convolutional layer, a global feature extraction branch, a local time-frequency feature extraction branch and a feature fusion module, the fused feature value The time domain signal is sent to the convolution layer for convolution operation and then sent to the global feature extraction branch to extract global features through S transform and to the local time-frequency feature extraction branch to extract local time-frequency features through fast Fourier transform. The extracted global features and local time-frequency features are sent to the feature fusion module for feature fusion. The fused features are sent to the multi-task learning component through the fully connected layer to obtain the fault type and fault location of the power grid. Among them, the global feature extraction branch extracts global features through S transform, and the local time-frequency feature extraction branch extracts local time-frequency features through fast Fourier transform. Thus, the two feature extraction branches of the global feature extraction branch and the local time-frequency feature extraction branch realize the fusion of S transform and fast Fourier transform in the signal processing method.

[0052] In step S1 of this embodiment, the panoramic waveform data obtained by measuring the grid waveform is obtained. When the waveform measurement unit data sampling rate is higher than 5kHz, the measured high-speed panoramic waveform data are voltage waveform and current waveform data.

[0053] In step S2 of this embodiment, the fusion feature value is obtained by splicing The calculation function expression is:

[0054] ,

[0055] in, is the incremental modulation statistic, is the root mean square value, is the waveform energy value, thus for the new power system m The panoramic waveform features of the measurement nodes are spliced ​​to achieve the initial fusion of fault information and form a fusion feature value containing fault information data , and its time domain expression is .

[0056] The calculation function expression of the incremental modulation statistics in step S2 is:

[0057] ,

[0058] In the above formula, is the incremental modulation statistic, is the next waveform value, is the current waveform value, Indicates the number of measurement nodes. Indicates the number of times incremental modulation is executed (the value is 2 in this embodiment);

[0059] The calculation function expression of the waveform energy value in step S2 is:

[0060]

[0061] In the above formula, is the waveform energy value, is the window length, For time point The square of the waveform value, For displacement The window function at , Represents a convolution operation.

[0062] Figure 3 This is a panoramic waveform diagram of PV tripping under the IEEE 33 bus system in this embodiment. Figure 4 This is a panoramic waveform RMS diagram of PV tripping under the IEEE 33bus system in this embodiment. Due to the similarities between different waveform data, it is challenging to directly use the panoramic waveform to determine the fault location. Figure 4 The extracted effective value can make the feature more significant to a certain extent. However, due to the similarity of the features of the surrounding nodes, it is still challenging to detect or locate the fault using only the effective value feature. Calculate multiple eigenvalues ​​and concatenate them to obtain fused eigenvalues , various characteristic values ​​include delta modulation statistics, RMS values, and waveform energy values. Figure 5 This is a visualization result diagram of an example of extracting deep features in this embodiment. Figure 6 This is a time-frequency spectrum information result diagram of extracting deep features in this embodiment. Figure 5 and Figure 6 The results show that the initially randomly distributed contours are transformed into continuous and smooth contours. In addition, the extracted features include low-frequency, medium-frequency and high-frequency regions. This example shows that the fault features (delta modulation statistics, RMS value and waveform energy value) are concatenated to obtain the fused feature value. ) can be automatically extracted in the neural network.

[0063] The features extracted from the deep network are often ambiguous and their specific physical meanings are unclear, which may cause the loss of relative position and type features during fault identification. In order to integrate the global and local features of the panoramic waveform into the neural network, a spatiotemporal spectrum mechanism (time-frequency spectrum module) is implemented in this embodiment to extract and integrate fault-related information and physical information. The time-frequency spectrum module includes a convolution layer, a global feature extraction branch, a local time-frequency feature extraction branch, and a feature fusion module, which includes two main stages: the waveform feature extraction stage and the feature fusion stage. The first layer of the spatiotemporal spectrum mechanism (time-frequency spectrum module) is a convolution operation. The function expression of the convolution layer in step S3 of this embodiment is:

[0064] ,

[0065] In the above formula, is the signal obtained by the convolution operation of the convolution layer at time t, is the convolution kernel, for The fusion eigenvalue at the moment The time domain signal, The operation results are then input into two branches of the feature extraction stage: the global feature extraction branch and the local time-frequency feature extraction branch to capture global features and local time-frequency features.

[0066] The S-transform time-frequency analysis method is first introduced in the space-time spectrum mechanism as a layer to extract physical information in the first branch (global feature extraction branch). The advantage of the S-transform is that it can decompose deep features into continuous time and frequency components. In this embodiment, the global feature extraction branch extracts global features by first performing an S-transform on the signal obtained by the convolution operation of the convolution layer according to the following formula:

[0067] ,

[0068] In the above formula, is the result of S transformation, is the signal obtained by the convolution operation of the convolution layer at time t, for Time frequency The Gaussian window function is are time shift and frequency respectively, is an imaginary unit, is time, and the Gaussian window function is defined as:

[0069] ,

[0070] In the above formula, for Time frequency Gaussian window function; the amplitude of the result obtained by S transformation Constructed with dimensions A two-dimensional matrix, where are the channel, width and length of the input feature respectively (in this embodiment, the values ​​are 64, 8 and 180 respectively), and the dimension is The width of the two-dimensional matrix Compressed to one dimension while maintaining the same shape, the dimension is The two-dimensional matrix is ​​processed using the enhanced block Inception to refine and filter features to obtain local time-frequency features The output of the S transform is a complex domain matrix. To facilitate the calculation of the model gradient, The amplitude As output, a two-dimensional matrix representation is obtained. The input channel, width and length are represented as ,but The output dimension becomes To speed up the calculation process, the width Compressing to one dimension while keeping the same shape, the result is Among them, the enhanced block Inception is an existing module, and the enhanced block Inception is represented as , then:

[0071] .

[0072] In the second branch (local time-frequency feature extraction branch), fast Fourier transform is applied to adjust the feature according to its global characteristics. The purpose of this branch is to enhance the expressiveness of the feature by reorganizing the important feature sequence. Specifically, in this embodiment, the local time-frequency feature extraction branch extracts the local time-frequency feature by first performing fast Fourier transform on the signal obtained by the convolution operation of the convolution layer according to the following formula:

[0073] ,

[0074] In the above formula, is the feature obtained by fast Fourier transform, is the length of the input signal, the length of the input signal and the length of the input feature d equal, for The discrete form of is an imaginary unit, For time, ; Features obtained by fast Fourier transform is a complex number that generates global components from a macroscopic perspective. In order to reduce redundant information and considering that the highest amplitude component occupies the main energy, only the highest amplitude component is selected. frequencies, and select the one with the highest amplitude based on the features obtained by fast Fourier transform The characteristic amplitude obtained by fast Fourier transform ,Sure The amplitude of the softmax normalization function Corresponding , using the period coefficient To adjust the input The size of The periodic coefficients [ , , ... , ],in for The discrete form of for Round up, For the selected frequency, use The period coefficient will Discrete form of Split into length The block is adjusted to a two-dimensional spectrogram, filled with zeros to match the dimension of the segmented series, and then processed with the enhanced block Inception to refine and filter the features to obtain global features. . The reinforcement block Inception is represented as , then:

[0075] .

[0076] in, is the resized frequency sequence.

[0077] In the feature fusion stage, in order to fuse the outputs of the two branches, the global and local features are fused together. In addition, and The dimension matches, that is and Dimensions d , d / 2 d The same should be true. In this stage, a convolution layer is added to achieve dimensional conversion. In this embodiment, the function expression of the feature fusion module is:

[0078] ,

[0079] In the above formula, is the fused feature, is a global feature, is the local time-frequency feature, is a trainable parameter, To adjust the hyperparameter of the ratio of global features to local time-frequency features (the value is 2 in this embodiment), Indicates the amplitude The softmax normalization function of is the characteristic amplitude obtained by fast Fourier transform.

[0080] In this embodiment, the fully connected layer includes two output linear layers corresponding to the fault detection task and the fault location task respectively. The function expressions of the two output linear layers are:

[0081] ,

[0082] ,

[0083] In the above formula, and are the output results of the two output linear layers, and Represent two output linear layers, and are the weights of the two output linear layers, is the input to the two output linear layers, and are the biases of the two output linear layers respectively.

[0084] In this embodiment, the multi-task learning component includes two task prediction modules for the fault detection task and the fault location task respectively. The task prediction module can be implemented by a fully connected layer for predicting the fault location based on the output results of the two output linear layers. and Mapped to output vector of fault type and fault location and .

[0085] In this embodiment, the loss function used in the training of the spatiotemporal spectrum S network is:

[0086] ,

[0087] In the above formula, is the loss function, is the network parameter of the space-time spectrum S network, and are the importance weight coefficients of the fault detection task and the fault location task, respectively. and Represent the true probability and predicted probability respectively (values ​​range from 0 to 1), is the penalty term (the importance weight coefficient can be calculated based on the L1 norm or L2 norm The sum of the two is multiplied by a given penalty coefficient. To achieve fault location, Set to the value corresponding to the panoramic waveform node measured during training. Therefore, the predicted maximum The corresponding index indicates the location of the faulty node.

[0088] In order to verify the learning performance and detection performance of the grid panoramic waveform fault identification and tracing method of the space-time spectrum S network in this embodiment, in this embodiment, a distributed model including 14 photovoltaics was established based on the IEEE 123-node test feeder system. The model was built based on OpenDSS and ATP-EMTP simulation software. The proportional changes of load size, proportion of new energy, etc. in the system were set to eliminate the influence of uncertainty, and the sampling rate was set to 12000Hz; in addition, in order to verify the anti-noise performance of the method, Gaussian white noise with a signal-to-noise ratio of 30dB and 50dB was added to the data.

[0089] In a simulation example, to verify the effectiveness of the dynamic weights in the multi-task framework, the multi-task learning component was trained for 100 iterations and terminated at 80 iterations. Figure 7 This is the result diagram of the weight coefficient of fault detection in this simulation example. Figure 8 The weight coefficient result diagram of the importance of the fault location task in this simulation example is shown in Figure 2. In this case, the importance weight coefficient and Initialized to 1. Figure 7 From the weight coefficients shown, the weights drop rapidly during the initial iterations (below 40) and then stabilize after minor adjustments. Always below , indicating that the fault detection task accounts for a large proportion. Figure 8 It can be seen that as the number of iterations increases, the total loss decreases from 3 to 0.02, indicating the effectiveness of the multi-task learning component training. In addition, both loss terms are very stable and do not decrease further when the number of iterations approaches 80.

[0090] In addition, this embodiment analyzes the fault detection and location performance under different ratios of training data and noise. The results are as follows: Fig. 9 and Fig.10 shown. Fig. 9 : is a graph showing the performance results of fault detection and positioning under different noise levels in this embodiment. Fig.10 This is a graph showing the fault detection and location performance results under different training set ratios in this embodiment. Fig. 9 and Fig.10It can be seen that the fault detection results show that the accuracy increases linearly with the increase in the amount of training data. The maximum and minimum accuracies observed were 99.45% and 97.65%, respectively. At a noise level of 30 dB, the performance of the power grid panoramic waveform fault identification and tracing method of the space-time spectrum S network in this embodiment showed a similar trend. When comparing 70% and 20% training data, the accuracy differences at 50 dB and 30 dB noise levels were 1.8% and 2.24%, respectively, indicating that the power grid panoramic waveform fault identification and tracing method of the space-time spectrum S network in this embodiment has strong learning ability. For fault location, the performance differences of 50% and 70% training data were 0.16% and 0.14%, respectively. The performance difference when using 20% ​​training data was 2.28%, indicating that the location task is more effective than the detection task. Overall, the test shows that the power grid panoramic waveform fault identification and tracing method of the space-time spectrum S network in this embodiment has strong feature extraction capabilities and adaptability even with a limited number of training faults.

[0091] It can be seen that the power grid panoramic waveform fault identification and tracing method of the space-time spectrum S network in this embodiment utilizes the advantages of high resolution and strong information integrity of the power grid panoramic waveform to construct a space-time spectrum mechanism and module based on fast Fourier transform and S transform of convolution operation. The space-time spectrum module effectively integrates the signal analysis method, so that the module has the characteristics of portability, simple structure and strong feature learning ability; a multi-task learning component based on a dynamic weight loss function is built, and the space-time spectrum mechanism is regarded as a layer in multi-task learning. An end-to-end space-time spectrum S network is constructed through multi-layer superposition to realize synchronous identification and tracing of panoramic information faults. The multi-task learning component can simultaneously learn the type and location information of new power system faults end-to-end, so that the method of the present invention has the characteristics of multi-function and simple structure.

[0092] In addition, this embodiment also provides a power grid panoramic waveform fault identification and tracing system of a space-time spectrum S network, including an interconnected microprocessor and a memory, and the microprocessor is programmed or configured to execute the power grid panoramic waveform fault identification and tracing method of the space-time spectrum S network.

[0093] In addition, this embodiment also provides a computer-readable storage medium, which stores a computer program or instruction, and the computer program or instruction is programmed or configured to execute the power grid panoramic waveform fault identification and tracing method of the space-time spectrum S network through a processor.

[0094] In addition, this embodiment also provides a computer program product, including a computer program or instructions, which are programmed or configured to execute the grid panoramic waveform fault identification and tracing method of the space-time spectrum S network through a processor.

[0095] Those skilled in the art should understand that the technical solutions provided by the embodiments of the present application may be in the form of methods, systems, or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes 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 process Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including an instruction device, which implements the functions specified in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0096] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A method for identifying and tracing power grid panoramic waveform faults based on a time-space spectrum S network, characterized in that: The steps include: S1, obtain the panoramic waveform data obtained by measuring the grid waveform , , is the number of sampling points; S2, based on panoramic waveform data Calculate multiple eigenvalues ​​and concatenate them to obtain fused eigenvalues , the multiple characteristic values ​​include part or all of incremental modulation statistics, root mean square values ​​and waveform energy values; S3, the fusion feature value The time domain signal is input into the trained space-time spectrum S network to obtain the fault type and fault location. The space-time spectrum S network is pre-trained to establish a fusion feature value The mapping relationship between the time domain signal, fault type and fault location; the spatiotemporal spectrum S network includes a multi-level feature extraction module, a fully connected layer and a multi-task learning component, the feature extraction module includes a sequentially connected time-frequency spectrum module, a convolution layer and a pooling layer, the time-frequency spectrum module includes a convolution layer, a global feature extraction branch, a local time-frequency feature extraction branch and a feature fusion module, the fusion feature value The time domain signal is sent to the convolution layer for convolution operation and then sent to the global feature extraction branch to extract global features through S transform and to the local time-frequency feature extraction branch to extract local time-frequency features through fast Fourier transform. The extracted global features and local time-frequency features are sent to the feature fusion module for feature fusion. The fused features are sent to the multi-task learning component through the fully connected layer to obtain the fault type and fault location of the power grid.

2. The method for identifying and tracing power grid panoramic waveform faults based on the space-time spectrum S network according to claim 1 is characterized in that: The fusion feature value is obtained by splicing in step S2 The calculation function expression is: ， in, is the incremental modulation statistic, is the root mean square value, is the waveform energy value; the calculation function expression of the incremental modulation statistical value in step S2 is: ， In the above formula, is the incremental modulation statistic, is the next waveform value, is the current waveform value, Indicates the number of measurement nodes. Indicates the number of times the incremental modulation is executed; the calculation function expression of the waveform energy value in step S2 is: In the above formula, is the waveform energy value, is the window length, For time point The square of the waveform value, For displacement The window function at , Represents a convolution operation.

3. The method for identifying and tracing power grid panoramic waveform faults based on the space-time spectrum S network according to claim 1 is characterized in that: The function expression of the convolutional layer in step S3 is: ， In the above formula, is the signal obtained by the convolution operation of the convolution layer at time t, is the convolution kernel, for The fusion eigenvalue at the moment The time domain signal, For time shift.

4. The method for identifying and tracing power grid panoramic waveform faults based on the space-time spectrum S network according to claim 1 is characterized in that: The global feature extraction branch extracts global features by first performing an S-transform on the signal obtained by the convolution operation of the convolution layer according to the following formula: ， In the above formula, is the result of S transformation, is the signal obtained by the convolution operation of the convolution layer at time t, for Time frequency The Gaussian window function is are time shift and frequency respectively, is an imaginary unit, is time, and the Gaussian window function is defined as: ， In the above formula, for Time frequency Gaussian window function; the amplitude of the result obtained by S transformation Constructed with dimensions A two-dimensional matrix, where are the channels, width and length of the input features respectively, and the dimension is The width of the two-dimensional matrix Compressed to one dimension while maintaining the same shape, the dimension is The two-dimensional matrix is ​​processed using the enhanced block Inception to refine and filter features to obtain local time-frequency features .

5. The method for identifying and tracing power grid panoramic waveform faults based on the space-time spectrum S network according to claim 1 is characterized in that: The local time-frequency feature extraction branch extracts local time-frequency features by first performing a fast Fourier transform on the signal obtained by the convolution operation of the convolution layer according to the following formula: ， In the above formula, is the feature obtained by fast Fourier transform, is the length of the input signal, the length of the input signal and the length of the input feature d equal, for The discrete form of is an imaginary unit, For time, ; Then select the one with the highest amplitude based on the features obtained by fast Fourier transform The characteristic amplitude obtained by fast Fourier transform ,Sure The amplitude of the softmax normalization function Corresponding , using the period coefficient To adjust the input The size of Periodic coefficient ,in for The discrete form of for Round up, For the selected frequency, ~ They are No. 1~ period coefficients, using The period coefficient will Discrete form of Split into length The block is adjusted to a two-dimensional spectrogram, filled with zeros to match the dimension of the segmented series, and then processed with the enhanced block Inception to refine and filter the features to obtain global features. .

6. The method for identifying and tracing power grid panoramic waveform faults based on the space-time spectrum S network according to claim 5 is characterized in that: The function expression of the feature fusion module is: ， In the above formula, is the fused feature, is a global feature, is the local time-frequency feature, is a trainable parameter, To adjust the hyperparameters of the ratio of global features to local time-frequency features, Indicates amplitude The softmax normalization function of is the characteristic amplitude obtained by fast Fourier transform.

7. The method for identifying and tracing power grid panoramic waveform faults based on the space-time spectrum S network according to claim 1 is characterized in that: The fully connected layer includes two output linear layers corresponding to the fault detection task and the fault location task respectively. The function expressions of the two output linear layers are: ， ， In the above formula, and are the output results of the two output linear layers, and Represent two output linear layers, and are the weights of the two output linear layers, is the input to the two output linear layers, and are biases of two output linear layers respectively; the multi-task learning component includes two task prediction modules for the fault detection task and the fault location task respectively, for predicting the fault according to the output results of the two output linear layers and Mapped to output vector of fault type and fault location and , the loss function used in the training of the spatiotemporal spectrum S network is: ， In the above formula, is the loss function, is the network parameter of the space-time spectrum S network, and are the importance weight coefficients of the fault detection task and the fault location task, respectively. and represent the true probability and predicted probability respectively, For penalty items.

8. A grid panoramic waveform fault identification and tracing system based on a time-space spectrum S network, comprising an interconnected microprocessor and a memory, characterized in that: The microprocessor is programmed or configured to execute the grid panoramic waveform fault identification and tracing method of the space-time spectrum S network as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program or instruction stored therein, characterized in that: The computer program or instruction is programmed or configured to execute the method for identifying and tracing power grid panoramic waveform faults of the space-time spectrum S network described in any one of claims 1 to 7 through a processor.

10. A computer program product comprising a computer program or instructions, characterized in that The computer program or instruction is programmed or configured to execute the method for identifying and tracing power grid panoramic waveform faults of the space-time spectrum S network described in any one of claims 1 to 7 through a processor.

Images