A method for identifying power quality disturbances based on synchronous phasor measurement data

Abstract

This invention proposes a power quality disturbance identification method based on synchronous phasor measurement data, comprising: data preprocessing: acquiring the original power quality disturbance signal and normalizing it; using fast Fourier transform and signal reconstruction methods to obtain a frequency domain component map containing the frequency domain information of the original disturbance signal; combining the original disturbance signal with its corresponding frequency domain component map as the input features of a parallel fusion convolutional neural network; training process of the parallel fusion convolutional neural network: generating multiple mixed power quality disturbance signals to train a parallel fusion convolutional neural network including a one-dimensional convolutional neural network, a two-dimensional convolutional neural network, and a parallel fusion layer, and adjusting the hyperparameters of the network to obtain a fully trained parallel fusion convolutional neural network; power quality disturbance identification: directly inputting the actually acquired signal into the fully trained parallel fusion convolutional neural network to complete the classification and identification of power quality disturbances.

Description

Technical Field

[0001] This invention relates to the technical fields of power quality disturbance identification and synchronous phasor measurement, and in particular to a power quality disturbance identification method based on synchronous phasor measurement data. Background Technology

[0002] To accelerate the achievement of "dual-carbon" goals, the construction of new power systems is progressing rapidly, exhibiting a diversified development trend characterized by high proportions of new energy sources, high proportions of power electronic equipment, and AC / DC interconnection. This has led to extensive and profound changes in the power quality characteristics and distribution levels across all stages of the power generation-grid-load-storage system. The new power system is gradually integrating massive amounts of new loads with highly nonlinear and stochastic fluctuations, such as electric vehicle charging stations, distributed photovoltaic systems, and large-scale power converters. This results in new characteristics of power quality problems, including multi-point occurrence, high frequency, and time-varying nature, posing a severe challenge to power quality management. To strengthen power quality management and implement the "Interim Measures for Power Quality Management," it is urgent to conduct power quality analysis, source tracing, and prevention work based on the source and load characteristics of new power systems. Among these, power quality disturbance identification is the primary task in carrying out a series of power quality work, playing a crucial role in the overall power quality management of the power system.

[0003] Power quality disturbance identification aims to rapidly analyze short-term, high-frequency power quality sampling data and identify the type of power quality disturbance currently affecting the system, thereby effectively guiding the location, tracing, and prevention of power quality problems. With the diversified development of power systems, traditional power quality disturbances based on harmonic and voltage amplitude characteristics have gradually transformed into power quality disturbances involving the fusion of multiple disturbances in the metastable state, rapidly increasing the complexity of power quality disturbance identification. To more accurately and efficiently identify the power quality characteristics of power systems, feeder terminals based on synchronous phasor measurement technology have been applied, providing higher-quality, longer-term analysis samples for power quality disturbance identification methods and further expanding the applicable scenarios of these methods. Therefore, researching power quality disturbance methods adapted to the new characteristics of modern power systems has strong application needs and practical engineering significance.

[0004] Currently, numerous scholars have conducted research on power quality disturbance identification, primarily based on two steps: signal preprocessing and disturbance classification. In signal preprocessing, various time-frequency analysis methods, such as Fast Fourier Transform (FFT), Continuous Wavelet Transform (CWT), and Stockwell Transform (SWT), are widely used, each with its own advantages and disadvantages. FFT can only obtain frequency domain information and cannot correlate it with time domain information; CWT and SWT suffer from issues such as optimal parameter selection, sensitivity to singular values, and high computational complexity. Compared to time-frequency analysis methods, signal decomposition methods have also seen some application, such as Empirical Mode Decomposition (EMD), Variational Mode Decomposition (VMD), and Adaptive Noise Set EEM. Signal decomposition aims to decompose the original signal into multiple eigenmode functions reflecting the characteristics of power quality disturbances; however, determining the number of decompositions and evaluating the relationship between eigenmode functions and specific power quality disturbance types remains unclear. Currently, many scholars employ trial-and-error methods, subjective determination methods, or optimization solutions. However, none of these methods can provide a preprocessing framework or paradigm that requires no human intervention, making it difficult to effectively improve the efficiency of power quality disturbance identification. After signal preprocessing, further classification and identification of power quality disturbances are necessary. Traditional classification and identification methods often use mathematical classification models or introduce machine learning models such as decision trees and support vector machines. However, with the increasing number of power quality disturbance types and the intensification of the coupling phenomenon of multiple disturbances, the identification accuracy of traditional mathematical classification models and machine learning models is relatively low, and their generalization ability for complex scenarios is difficult to guarantee. Therefore, many novel classifiers based on computer vision, such as convolutional neural networks and Transformers, have gained popularity among scholars. However, currently, these methods are limited to identifying the original signal after a single transformation, with extremely limited input features. Based on the principles of power quality analysis, power quality disturbances have significant characteristics in both the time and frequency domains. Therefore, power quality disturbance identification methods that take into account both time and frequency domain characteristics urgently need improvement and refinement. Summary of the Invention

[0005] To address the shortcomings and deficiencies of existing technologies, this invention provides a power quality disturbance identification method based on synchronous phasor measurement data, the main design features of which include:

[0006] 1) A frequency domain image representation method for synchronous phasor measurement data is proposed, which introduces sufficient frequency domain information into the power quality disturbance identification method;

[0007] 2) A novel parallel fusion convolutional neural network that takes into account both time-frequency domain features is proposed, which can effectively utilize the time-frequency domain features of the input to achieve high-accuracy power quality disturbance identification.

[0008] The specific technical solution adopted is as follows:

[0009] A method for identifying power quality disturbances based on synchronous phasor measurement data includes:

[0010] Data preprocessing: Obtain the original power quality disturbance signal and normalize it; use Fast Fourier Transform and signal reconstruction methods to obtain the frequency domain component map containing the frequency domain information of the original disturbance signal; combine the original disturbance signal and its corresponding frequency domain component map as the input features of the parallel fusion convolutional neural network;

[0011] Training process of parallel fusion convolutional neural network: Generate various mixed power quality perturbation signals to train the parallel fusion convolutional neural network including one-dimensional convolutional neural network, two-dimensional convolutional neural network and parallel fusion layer, and adjust the hyperparameters of the network to obtain a fully trained parallel fusion convolutional neural network.

[0012] Power quality disturbance identification: The actual collected signals are directly input into a well-trained parallel fusion convolutional neural network to complete the classification and identification of power quality disturbances.

[0013] Furthermore, the original power quality disturbance signal is a voltage timing signal or a current timing signal.

[0014] Furthermore, the original power quality disturbance signal u(t) or i(t) containing time-domain information is compared with the frequency-domain component map TF obtained by processing it and containing frequency-domain information. u(t) or TF i(t) By combining these features, an input feature set containing both time-frequency features is constructed for parallel fusion of convolutional neural networks for feature extraction and classification.

[0015] In this process, the original power quality disturbance signal u(t) or i(t) is directly used as the input to the time-domain features of the parallel fusion convolutional neural network.

[0016] Furthermore, for the original power quality disturbance signal (u(t)), a fast Fourier transform is first applied to transform it to obtain a frequency domain sequence reflecting the power quality disturbance signal.

[0017]

[0018] Among them, W fft The matrix is ​​the Fast Fourier Transform matrix, and W is an L×N Vandermonde matrix; N =e (-2πi) / N is the multiplier of the Fast Fourier Transform; L is the length of the frequency domain sequence, and N is the length of the time domain sequence;

[0019] Then the frequency domain sequence The component matrix is ​​reconstructed into a component matrix with time-frequency domain characteristics, and the specific process is shown in equation (2).

[0020]

[0021] in, This is the output frequency domain component matrix; Transform a one-dimensional frequency domain sequence into a corresponding L×L diagonal matrix; It is the generalized inverse matrix of the Fast Fourier Transform;

[0022] Then the component matrix is ​​normalized and converted into a corresponding grayscale image for recognition by a two-dimensional convolutional neural network, as shown in equation (3).

[0023]

[0024] Where TF(m,n) is the output two-dimensional frequency domain component map; round is the rounding function; max and min are the operations to find the maximum and minimum values ​​of the frequency domain component matrix.

[0025] If the original disturbance signal is a current i(t), the same method is used to complete the preprocessing and subsequent training.

[0026] Furthermore, the parallel fusion convolutional neural network includes: a one-dimensional convolutional neural network for extracting temporal features, a two-dimensional convolutional neural network for extracting frequency domain features, and a parallel fusion layer for fusing features.

[0027] Furthermore, the one-dimensional convolutional neural network consists of two sets of serially connected convolutional layers, pooling layers, and batch normalization layers. It takes the original perturbation signal as input and sequentially passes it through the one-dimensional convolutional layer, one-dimensional pooling layer, and batch normalization layer to extract one-dimensional temporal features.

[0028] Furthermore, the two-dimensional convolutional neural network takes a two-dimensional frequency domain component map as input, and the input dimension depends on the size of the original signal after transformation. Its frequency domain features are extracted sequentially through a two-dimensional convolutional layer, a two-dimensional pooling layer, and a batch normalization layer.

[0029] Furthermore, the output dimension of the two-dimensional convolutional neural network is the same as that of the one-dimensional convolutional neural network, and can be obtained by scaling through fully connected layers.

[0030] Furthermore, the parallel fusion layer horizontally concatenates one-dimensional and two-dimensional features, and then sequentially inputs the concatenated features into a fully connected layer, a bidirectional long short-term memory network layer, a fully connected layer, and a Softmax activation function to complete the power quality disturbance identification. Among them, the two fully connected layers are used to scale the signal dimension and construct the probabilistic relationship between input and output; the bidirectional long short-term memory network layer is used to extract the temporal dependency relationship between the time and frequency domains of the signals; and the Softmax activation function is used to transform the output features into multiple labels corresponding to the power quality disturbance types.

[0031] Furthermore, a synchronous phasor measurement device is provided, equipped with a fully trained parallel fusion convolutional neural network obtained according to the method described above, for classifying and identifying power quality disturbances. That is, the fully trained parallel fusion convolutional neural network is embedded into the synchronous phasor measurement device.

[0032] Compared to existing technologies, this invention and its preferred embodiment are applicable to any scenario equipped with a synchronous phasor measurement device capable of voltage and current sampling. It can acquire voltage or current waveform data according to the actual scenario requirements, perform signal preprocessing and classification on the data, and derive the type of power quality disturbance reflecting the current waveform quality. This effectively guides power quality management personnel in conducting source tracing analysis and prevention of power quality problems. While this method, based on synchronous phasor measurement data and fully utilizing the time-frequency domain characteristics of the original signal, slightly increases the computational load, it provides more accurate power quality disturbance identification results. Furthermore, the provided algorithm can achieve power quality disturbance classification and identification in actual scenarios without any manual intervention. After pre-parameter settings are completed during the training process, it can adapt to power quality disturbance classification and identification in real-world scenarios. When applied to new scenarios, only a small amount of data is needed for automatic updates, demonstrating significant engineering application value and providing services for the safe operation, maintenance management, and efficiency improvement of power grid companies. Attached Figure Description

[0033] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0034] Figure 1 This is a schematic diagram of the power quality disturbance identification framework according to an embodiment of the present invention. Detailed Implementation

[0035] In the following, specific embodiments of this application will be described in detail with reference to the accompanying drawings. Based on these detailed descriptions, those skilled in the art will be able to clearly understand and implement this application. Without departing from the principles of this application, features from various embodiments can be combined to obtain new implementations, or certain features from some embodiments can be substituted to obtain other preferred implementations.

[0036] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0037] To make the features and advantages of this patent more apparent and understandable, specific embodiments are provided below for detailed explanation: 1. Preprocessing stage based on synchronous phasor measurement data

[0038] Synchronous phasor measurement utilizes high sampling frequencies to measure voltage and current data from the actual power grid, simultaneously performing phasor calculations to obtain the magnitude and phase of sinusoidal quantities. Under a unified time standard, all measured data are time-stamped, allowing for direct comparison between data points. Therefore, the data possesses high quality and long duration characteristics, effectively adapting to current methods for identifying numerous power quality disturbances. Furthermore, since multiple data segments can be directly compared, the power quality disturbance identification results can be directly applied to scenarios such as interference source identification and harmonic source tracing, possessing significant practical engineering value.

[0039] In this embodiment of the invention, the preprocessing step based on synchronous phasor measurement data aims to combine the original power quality disturbance signal (u(t) or i(t)) containing time-domain information with the frequency component map (TF) obtained by processing it and containing frequency-domain information. u(t) or TF i(t) The input signal and image containing both time and frequency features are combined to construct an input signal and image that can be used for feature extraction and classification by a parallel fusion convolutional neural network. Therefore, the following mainly introduces the construction process of the frequency domain component map containing frequency domain information. The original power quality disturbance signal (u(t) or i(t)) can be directly used as a time domain feature input to the parallel fusion convolutional neural network.

[0040] This embodiment takes a voltage signal input as an example. For the original power quality disturbance signal (u(t)), a fast Fourier transform is first used to transform it, which yields a frequency domain sequence reflecting the power quality disturbance signal. As shown in the following formula:

[0041]

[0042] Among them, W fft The matrix is ​​the Fast Fourier Transform matrix, and W is an L×N Vandermonde matrix; N =e (-2πi) / N is the multiplier of the Fast Fourier Transform; L is the length of the frequency domain sequence, and N is the length of the time domain sequence.

[0043] One-dimensional frequency domain sequence Although it can effectively reflect the frequency domain steady-state information of the original disturbance signal, the cross-correlation between this sequence and the original power quality disturbance signal (u(t)) is low. Directly using this sequence as frequency domain information input will lead to problems such as feature confusion or network degradation during time-frequency domain feature fusion. Therefore, it is necessary to use the frequency domain sequence as input. To avoid problems such as feature confusion, the frequency domain sequence is further reconstructed into a component matrix with time-frequency domain characteristics, thereby improving the cross-correlation between the two feature inputs and improving the learning effect of the neural network. The specific process is shown in Equation (2).

[0044]

[0045] in, This is the output frequency domain component matrix; Transform a one-dimensional frequency domain sequence into a corresponding L×L diagonal matrix; It is the generalized inverse matrix of the Fast Fourier Transform.

[0046] Furthermore, the component matrix is ​​normalized and converted into a corresponding grayscale image for recognition by the two-dimensional convolutional neural network, as shown in equation (3).

[0047]

[0048] Where TF(m,n) is the output two-dimensional frequency domain component map; round is the rounding function; max and min are operations to find the maximum and minimum values ​​of the frequency domain component matrix; through this transformation, the frequency domain component matrix can be converted into a grayscale image with

[0255] pixel intensity, which can significantly enhance the signal characteristics of the two-dimensional frequency domain components. Therefore, by combining the one-dimensional voltage time series signal and the two-dimensional frequency domain component map and inputting them into the neural network, features can be adaptively extracted and disturbance recognition can be completed through parallel fusion convolutional neural networks.

[0049] 2 Power quality disturbance identification based on parallel fusion convolutional neural network

[0050] The parallel fusion convolutional neural network consists of three parts: a one-dimensional convolutional neural network for extracting time-domain features, a two-dimensional convolutional neural network for extracting frequency-domain features, and a parallel fusion layer for fusing features. To simplify the analysis, this embodiment assumes that the sampling frequency domain for synchronous phasor measurements is 10kHz and the analysis time window for power quality disturbances is 0.2s, i.e., 2000 sampling points. Therefore, the parallel fusion convolutional neural network is designed as follows:

[0051] 1) One-dimensional convolutional neural networks for extracting temporal features

[0052] A one-dimensional convolutional neural network takes the original perturbed signal u(t) as input, with an input dimension of 2000. It sequentially extracts one-dimensional temporal features through one-dimensional convolutional layers, one-dimensional pooling layers, and batch normalization layers. Specifically, the one-dimensional convolutional layers use multiple sets of one-dimensional convolutional kernels to perform convolution operations on the original signal, effectively extracting the temporal features of the one-dimensional signal; the one-dimensional normalization layers downsample the multiple sets of signals obtained from the convolutional layers to extract effective and obvious temporal features; and the batch normalization layers normalize the input features, thus avoiding overfitting and network degradation. The one-dimensional convolutional neural network used in this embodiment consists of two sets of serially connected convolutional layers, pooling layers, and batch normalization layers, with an output dimension of 1000.

[0053] 2) Two-dimensional convolutional neural networks for extracting frequency domain features

[0054] Two-dimensional convolutional neural networks (2D convolutional neural networks) take a two-dimensional frequency domain component map TF(m,n) as input. The input dimension depends on the size of the original signal after transformation. If a two-sided spectrum is used to represent the two-dimensional frequency domain component map of the original signal, its size will be 2000*2000. Different processing methods can be used to simplify this two-dimensional frequency domain component map for different problems, thereby achieving network simplification and efficiency improvement. Frequency domain feature extraction also needs to be completed through two-dimensional convolutional layers, two-dimensional pooling layers, and batch normalization layers, with the specific details as above. To ensure easier feature interaction, the output dimension of the two-dimensional convolutional neural network is also 1000 by reasonably setting the convolutional block and pooling size.

[0055] 3) Parallel fusion layer for fusing features

[0056] The parallel fusion layer first performs a concatenation operation, horizontally concatenating one-dimensional and two-dimensional features. Then, these features are sequentially input into a fully connected layer, a bidirectional long short-term memory (LSTM) network layer, another fully connected layer, and a Softmax activation function to complete power quality disturbance identification. Specifically, two fully connected layers are used to scale the signal dimension and construct the probabilistic relationship between input and output; the bidirectional LSM network layer is used to extract the time-frequency domain temporal dependencies between signals; and the Softmax activation function transforms the output features into multiple labels corresponding to different power quality disturbance types.

[0057] As can be seen from the above three network blocks, the time-series features are extracted entirely by a one-dimensional convolutional neural network, while the frequency-domain features are extracted entirely by a two-dimensional convolutional neural network. These two types of features are ultimately fused into a parallel fusion layer, and further explored through a fully connected layer and a bidirectional long short-term memory network layer to uncover new time-frequency domain mapping relationships. This ultimately achieves efficient and accurate power quality disturbance identification. The detailed network architecture is as follows: Figure 1 As shown.

[0058] 3. Algorithm Flow

[0059] Based on the above design, the specific process of the power quality disturbance identification method based on synchronous phasor measurement data proposed in this embodiment of the invention is as follows:

[0060] 1) Data preprocessing: First, the original power quality disturbance signal (u(t) or i(t)) is acquired and normalized; second, the frequency domain component map (TF) containing frequency domain information is calculated according to the fast Fourier transform of equations (1)-(3) and the signal reconstruction method. u(t) or TF i(t) Finally, the original power quality disturbance signal and the frequency domain component map are combined as the input features of the convolutional neural network.

[0061] 2) Training process of parallel fused convolutional neural network: A parallel convolutional neural network adapted to the above two feature inputs is designed, which includes a one-dimensional convolutional neural network, a two-dimensional convolutional neural network, and a fused layer. It is trained using various mixed power quality perturbation measured and simulated signals. The hyperparameters of the neural network are adjusted through trial and error, and a fully trained parallel fused convolutional neural network is output.

[0062] 3) Power quality disturbance identification: The actual collected signals are directly input into a well-trained parallel fusion convolutional neural network or the network is embedded into a synchronous phasor measurement device, so as to quickly and automatically complete the classification and identification of power quality disturbances.

[0063] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied 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.

[0064] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0065] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0066] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0067] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0068] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0069] This patent is not limited to the above-described preferred embodiment. Anyone can derive other forms of power quality disturbance identification method based on synchronous phasor measurement data under the guidance of this patent. All equivalent changes and modifications made within the scope of this patent application shall fall within the scope of this patent.

Claims

1. A method for identifying power quality disturbances based on synchronous phasor measurement data, characterized in that, include: Data preprocessing: Acquire the raw power quality disturbance signal and normalize it; The frequency domain component map containing the frequency domain information of the original perturbation signal is obtained by using Fast Fourier Transform and signal reconstruction methods; the original perturbation signal and its corresponding frequency domain component map are combined as the input features of the parallel fusion convolutional neural network. Training process of parallel fusion convolutional neural network: Generate various mixed power quality perturbation signals to train the parallel fusion convolutional neural network including one-dimensional convolutional neural network, two-dimensional convolutional neural network and parallel fusion layer, and adjust the hyperparameters of the network to obtain a fully trained parallel fusion convolutional neural network. Power quality disturbance identification: The actual collected signals are directly input into a well-trained parallel fusion convolutional neural network to complete the classification and identification of power quality disturbances; For the original power quality disturbance signal (u(t)), a fast Fourier transform is first applied to transform it to obtain a frequency domain sequence reflecting the power quality disturbance signal. Among them, W fft The matrix is ​​the Fast Fourier Transform matrix, and W is an L×N Vandermonde matrix; N =e (-2πi) / N is the multiplier of the Fast Fourier Transform; L is the length of the frequency domain sequence, and N is the length of the time domain sequence; Then the frequency domain sequence The component matrix is ​​reconstructed into a component matrix with time-frequency domain characteristics, and the specific process is shown in equation (2). in, This is the output frequency domain component matrix; Transform a one-dimensional frequency domain sequence into a corresponding L×L diagonal matrix; It is the generalized inverse matrix of the Fast Fourier Transform; Then the component matrix is ​​normalized and converted into a corresponding grayscale image for recognition by a two-dimensional convolutional neural network, as shown in equation (3). Where TF(m,n) is the output two-dimensional frequency domain component map; round is the rounding function; max and min are the operations to find the maximum and minimum values ​​of the frequency domain component matrix.

2. The power quality disturbance identification method based on synchronous phasor measurement data according to claim 1, characterized in that: The original power quality disturbance signal is a voltage timing signal or a current timing signal.

3. The power quality disturbance identification method based on synchronous phasor measurement data according to claim 1, characterized in that: The original power quality disturbance signal u(t) or i(t) containing time-domain information is compared with the frequency component map TF obtained by processing it and containing frequency-domain information. u(t) or TF i(t) By combining these features, an input feature set containing both time-frequency features is constructed for parallel feature extraction and classification recognition of convolutional neural networks.

4. The power quality disturbance identification method based on synchronous phasor measurement data according to claim 1, characterized in that: The parallel fusion convolutional neural network includes: a one-dimensional convolutional neural network for extracting temporal features, a two-dimensional convolutional neural network for extracting frequency domain features, and a parallel fusion layer for fusing features.

5. The power quality disturbance identification method based on synchronous phasor measurement data according to claim 4, characterized in that: The one-dimensional convolutional neural network consists of two sets of serially connected convolutional layers, pooling layers, and batch normalization layers. It takes the original perturbation signal as input and extracts one-dimensional temporal features by sequentially passing through the one-dimensional convolutional layer, the one-dimensional pooling layer, and the batch normalization layer.

6. The power quality disturbance identification method based on synchronous phasor measurement data according to claim 4, characterized in that: The two-dimensional convolutional neural network takes a two-dimensional frequency domain component map as input, and the input dimension depends on the size of the original signal after transformation. The extraction of its frequency domain features is accomplished sequentially through a two-dimensional convolutional layer, a two-dimensional pooling layer, and a batch normalization layer.

7. The power quality disturbance identification method based on synchronous phasor measurement data according to claim 6, characterized in that: The output dimension of the two-dimensional convolutional neural network is the same as that of the one-dimensional convolutional neural network, and is obtained by scaling through a fully connected layer.

8. The power quality disturbance identification method based on synchronous phasor measurement data according to claim 4, characterized in that: The parallel fusion layer horizontally concatenates one-dimensional and two-dimensional features, and then sequentially inputs the concatenated features into a fully connected layer, a bidirectional long short-term memory network layer, another fully connected layer, and a Softmax activation function to complete the power quality disturbance identification. Among them, the two fully connected layers are used to scale the signal dimension and construct the probabilistic relationship between input and output; the bidirectional long short-term memory network layer is used to extract the time-frequency domain temporal dependency relationship between signals; and the Softmax activation function is used to transform the output features into multiple labels corresponding to the power quality disturbance types.

9. A synchronous phasor measurement device, characterized in that, A fully trained parallel fusion convolutional neural network, obtained by any one of the methods described in claims 1-8, is provided for the classification and identification of power quality disturbances.

Images