Wideband oscillation identification method and device for power system

Abstract

The invention discloses an electric power system broadband oscillation identification method and device, and can be used in the technical field of electric power systems. The method comprises the following steps: acquiring an electric signal of a power system based on a Nyquist sampling theorem, and preprocessing the electric signal to obtain a preprocessed electric signal; wavelet transformation is carried out on the preprocessed electric signals, a wavelet feature matrix is constructed according to a wavelet transformation result, and the wavelet feature matrix comprises wavelet time-frequency energy distribution, a scale energy spectrum and a marginal spectrum; and inputting the wavelet feature matrix into a recognition model constructed based on a convolutional neural network, so that the recognition model performs broadband oscillation recognition on the power system and outputs a recognition result, the recognition model is obtained by training the convolutional neural network according to a wavelet feature matrix constructed based on an electric signal of the power system in a training set.

Description

Technical Field

[0001] This application relates to the field of power system technology, specifically to a method and apparatus for identifying broadband oscillations in power systems. Background Technology

[0002] Broadband oscillations in power systems refer to oscillations with frequencies ranging from a few hertz to several kilohertz, typically caused by the interaction between the rapid control response of power electronic equipment and grid impedance. These oscillations can threaten system stability, leading to equipment damage or power outages. Traditional oscillation analysis methods have limitations in non-steady-state signal processing, failing to simultaneously provide localized information in both the time and frequency domains. Therefore, a method is needed to accurately identify the frequency, amplitude, and time-domain characteristics of broadband oscillations.

[0003] Existing technologies typically employ two methods for oscillation identification: wavelet transform, which decomposes the signal into multiple scales using wavelet basis functions to extract oscillation features; and neural network methods, which utilize deep learning models to directly learn oscillation patterns from signal data.

[0004] Existing broadband oscillation identification methods have the following limitations:

[0005] While wavelet transform-based methods possess time-frequency localization analysis capabilities, their performance heavily relies on the manual selection of wavelet basis functions and scaling parameters, lacking adaptive adjustment capabilities. They are also highly sensitive to noise and inherently lack pattern recognition capabilities, thus failing to achieve end-to-end intelligent identification.

[0006] While neural network-based methods possess excellent nonlinear fitting capabilities, they typically require large amounts of high-quality labeled data for training and have limited ability to extract time-frequency structural features of oscillating signals. Especially when using unsupervised or reconstructive models such as autoencoders, which focus on data compression and reconstruction rather than regression tasks oriented towards physical parameters, the model output lacks clear physical meaning, exhibits poor interpretability, and is difficult to directly support oscillation mechanism analysis and control decisions.

[0007] Existing methods mostly focus on qualitative discrimination and cannot simultaneously provide quantitative modal parameters such as frequency, amplitude, and damping ratio. This makes it difficult to apply the analysis results to the accurate sensing and suppression control of broadband oscillations, thus limiting their decision support value in engineering practice.

[0008] In summary, existing methods have failed to effectively balance adaptive feature learning and physical interpretability, and have not fully utilized the inherent complementary advantages of wavelet time-frequency structures and convolutional neural networks in feature extraction. Summary of the Invention

[0009] To address at least some of the aforementioned technical problems, embodiments of this application provide a method and apparatus for identifying broadband oscillations in power systems.

[0010] This application provides a method for identifying broadband oscillations in a power system, comprising:

[0011] The electrical signals of the power system are acquired based on the Nyquist sampling theorem, and the electrical signals are preprocessed to obtain the preprocessed electrical signals.

[0012] The preprocessed electrical signal is subjected to wavelet transform, and a wavelet feature matrix is ​​constructed based on the wavelet transform result. The wavelet feature matrix includes wavelet time-frequency energy distribution, scale energy spectrum, and marginal spectrum.

[0013] The wavelet feature matrix is ​​input into the recognition model constructed based on the convolutional neural network, so that the recognition model can perform broadband oscillation recognition on the power system and output the recognition result. The recognition model is obtained by training the convolutional neural network based on the wavelet feature matrix constructed based on the electrical signal of the power system in the training set.

[0014] In some embodiments, the preprocessing of the electrical signal to obtain a preprocessed electrical signal includes:

[0015] The electrical signal is decomposed into approximation coefficients and detail coefficients, wherein the approximation coefficients and the detail coefficients correspond to the low-frequency trend and high-frequency details of the electrical signal, respectively.

[0016] The detail coefficients are then subjected to threshold denoising to obtain the denoised detail coefficients.

[0017] The signal is reconstructed based on the approximation coefficients and the denoised detail coefficients to obtain the reconstructed electrical signal;

[0018] The reconstructed electrical signal is standardized to obtain a preprocessed electrical signal.

[0019] In some embodiments, the detail coefficients are thresholded and denoised based on the following soft thresholding function to obtain the denoised detail coefficients:

[0020]

[0021] in,

[0022] Indicates the detail coefficient after noise reduction;

[0023] Indicates detail coefficients;

[0024] Indicates the threshold parameter.

[0025] , The noise standard deviation is estimated using the median of the detail coefficients. Indicates the number of data points.

[0026] In some embodiments, wavelet transform is performed on the preprocessed electrical signal based on the following formula:

[0027]

[0028] in,

[0029] Indicates the wavelet transform result;

[0030] This represents the scaling parameter, which controls the scaling of the wavelet function;

[0031] This represents the preprocessed electrical signal;

[0032] This represents the translation parameter, which controls the position of the wavelet function;

[0033] Represents the complex conjugate of the wavelet function;

[0034] This represents the scaling factor, used to ensure energy normalization of wavelet functions at different scales.

[0035] In some embodiments, the structure of the convolutional neural network is as follows:

[0036] The input layer is used to receive the wavelet time-frequency feature matrix;

[0037] The first convolutional layer, the expression for the first convolutional layer is: ,in, This is the output of the first convolutional layer. For input variables, b1 is the convolution kernel, b1 is a constant coefficient, and ReLU is the activation function;

[0038] The second convolutional layer, the expression for the second convolutional layer is: ,in, is the output of the second convolutional layer, W2 is the convolutional kernel of the second convolutional layer, and b2 is a constant coefficient;

[0039] Pooling layer, the expression for pooling layer is: ;

[0040] The fully connected layer consists of three levels. , , ,

[0041]

[0042] Where W3, W4, and W5 are convolution kernels, and b3, b4, and b5 are constant coefficients;

[0043] The output layer, the expression for the output layer is: ,in, Indicates the output variable. This represents the estimated oscillation frequency, reflecting the speed of the oscillation. This represents the estimated amplitude of the oscillation, characterizing the intensity of the oscillation; This represents the estimated damping ratio, describing the damping characteristics of the oscillation.

[0044] In some embodiments, the following weighted loss function is used when training the convolutional neural network:

[0045]

[0046] in, , , The weighting coefficients and the loss functions for each component are as follows:

[0047] Frequency estimation loss ;

[0048] Amplitude estimation loss ;

[0049] Damping ratio estimated loss ;

[0050] in, , , These are the actual values ​​of frequency, amplitude, and damping ratio, respectively, and N is the number of data points.

[0051] In some embodiments, a sliding window processing strategy is used to collect electrical signals from the power system, wherein the length of the analysis window is greater than a set threshold, and the sliding step size is set to be less than the length of the analysis window to ensure that adjacent analysis windows overlap.

[0052] In some embodiments, the identification result includes oscillation frequency, amplitude, and damping ratio; the method further includes:

[0053] Based on the established reliability detection rules, the reliability of the oscillation frequency, amplitude, and damping ratio is detected.

[0054] This application also provides a power system broadband oscillation identification device, comprising:

[0055] The acquisition and preprocessing module is used to acquire electrical signals of the power system based on the Nyquist sampling theorem and preprocess the electrical signals to obtain preprocessed electrical signals.

[0056] The conversion module is used to perform wavelet transform on the preprocessed electrical signal and construct a wavelet feature matrix based on the wavelet transform result. The wavelet feature matrix includes wavelet time-frequency energy distribution, scale energy spectrum and marginal spectrum.

[0057] The recognition module is used to input the wavelet feature matrix into a recognition model constructed based on a convolutional neural network, so that the recognition model can perform broadband oscillation recognition on the power system and output the recognition result. The recognition model is obtained by training the convolutional neural network based on the wavelet feature matrix constructed based on the electrical signals of the power system in the training set.

[0058] This application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the methods described in any of the above embodiments.

[0059] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the methods described in any of the above embodiments.

[0060] This application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the methods described in any of the above embodiments.

[0061] The power system broadband oscillation identification method and apparatus provided in this application acquires electrical signals from the power system based on the Nyquist sampling theorem and preprocesses the electrical signals to obtain preprocessed electrical signals. Wavelet transform is performed on the preprocessed electrical signals, and a wavelet feature matrix is ​​constructed based on the wavelet transform results. The wavelet feature matrix includes wavelet time-frequency energy distribution, scale energy spectrum, and marginal spectrum. The wavelet feature matrix is ​​input into an identification model constructed based on a convolutional neural network, enabling the identification model to identify broadband oscillations in the power system and output identification results. The identification model is obtained by training the convolutional neural network using the wavelet feature matrix constructed from the electrical signals of the power system in a training set. This solves the problems of low accuracy and poor adaptability in broadband oscillation identification in existing technologies. By combining the time-frequency analysis capability of wavelet transform with the self-learning characteristics of neural networks, high-precision and adaptive broadband oscillation identification is achieved. Attached Figure Description

[0062] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0063] Figure 1 This is a flowchart illustrating a broadband oscillation identification method for power systems provided in an embodiment of this application.

[0064] Figure 2 This is a partial flowchart illustrating a broadband oscillation identification method for power systems provided in an embodiment of this application.

[0065] Figure 3 This is a flowchart illustrating a broadband oscillation identification method for power systems provided in an embodiment of this application.

[0066] Figure 4 This is a schematic diagram of the structure of a power system broadband oscillation identification device provided in an embodiment of this application.

[0067] Figure 5 This is a schematic diagram of the physical structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0068] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the embodiments of this application will be further described in detail below with reference to the accompanying drawings. Here, the illustrative embodiments and their descriptions are used to explain this application, but are not intended to limit this application. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be arbitrarily arranged.

[0069] The terms “first,” “second,” etc., used in this document are not intended to specifically refer to order or sequence, nor are they used to limit this application; they are merely used to distinguish elements or operations described using the same technical terms.

[0070] The terms “include,” “including,” “have,” “contain,” etc., used in this article are all open-ended terms, meaning that they include but are not limited to.

[0071] The term "and / or" as used in this document includes any or all of the items mentioned.

[0072] Figure 1 This is a flowchart illustrating a power system broadband oscillation identification method provided in an embodiment of this application. Figure 1 As shown in the embodiment of this application, a power system broadband oscillation identification method includes:

[0073] S1. Acquire electrical signals from the power system based on the Nyquist sampling theorem, and preprocess the electrical signals to obtain preprocessed electrical signals;

[0074] In step S1, the broadband oscillation signal of the power system is characterized by a wide frequency range and rapid amplitude changes, thus requiring high-precision acquisition equipment. A synchronous phasor measurement unit is preferred as the data acquisition source, and its sampling frequency setting must strictly adhere to the Nyquist sampling theorem:

[0075]

[0076] in, Sampling frequency, This represents the highest oscillation frequency to be detected. Considering that wideband oscillations in power systems typically cover a frequency range of 5Hz to 2000Hz, the recommended sampling frequency is [not specified]. To ensure signal integrity.

[0077] The electrical signals may include current and / or voltage signals. The field environment of power systems is complex, and the collected signals often contain various noise interferences. Therefore, it is necessary to preprocess the collected electrical signals.

[0078] like Figure 2 As shown, in some embodiments, the preprocessing of the electrical signal to obtain a preprocessed electrical signal includes:

[0079] S11. Decompose the electrical signal into approximation coefficients and detail coefficients, wherein the approximation coefficients and the detail coefficients correspond to the low-frequency trend and high-frequency details of the electrical signal, respectively.

[0080] In step S11, the electrical signal can be decomposed into approximation coefficients and detail coefficients using the following formula (2):

[0081]

[0082] Where x represents the acquired electrical signal; CA represents the approximation coefficient, and CD represents the detail coefficient, corresponding to the low-frequency trend and high-frequency details of the signal, respectively; This indicates that the Daubechies 4 wavelet basis functions are used for decomposition.

[0083] S12. Perform threshold denoising on the detail coefficients to obtain the denoised detail coefficients;

[0084] In step S12, an improved soft thresholding function can be used to handle the detail coefficients:

[0085]

[0086] in, The threshold parameter λ, representing the detail coefficients after denoising, can be adaptively determined using the following formula (4):

[0087]

[0088] in, The noise standard deviation is estimated using the median of the detail coefficients. This refers to the signal length, also known as the number of data points. This threshold selection method can effectively suppress noise while preserving the valid signal.

[0089] S13. Reconstruct the signal based on the approximation coefficients and the denoised detail coefficients to obtain the reconstructed electrical signal;

[0090] In step S13, the signal can be reconstructed using the following formula (5):

[0091]

[0092] in, This represents the reconstructed electrical signal, which retains the original oscillation characteristics while significantly reducing the impact of noise.

[0093] S14. The reconstructed electrical signal is standardized to obtain a preprocessed electrical signal.

[0094] In step S14, to eliminate the influence of dimensions and improve the training efficiency of the neural network, the reconstructed electrical signal can be Z-score normalized:

[0095]

[0096] in, This represents the standardized electrical signal, that is, the pre-processed electrical signal. Indicates the signal mean. This represents the standard deviation of the signal. The standardized electrical signal has zero mean and unit variance, which helps improve the model's convergence speed.

[0097] S2. Perform wavelet transform on the preprocessed electrical signal, and construct a wavelet feature matrix based on the wavelet transform result. The wavelet feature matrix includes wavelet time-frequency energy distribution, scale energy spectrum, and marginal spectrum.

[0098] In step S2, feature extraction is achieved through continuous wavelet transform. The Morlet wavelet can be used as the basis function, as it exhibits good localization properties in both the time and frequency domains, making it particularly suitable for analyzing non-stationary oscillating signals. Its expression is shown below:

[0099]

[0100] in, Using wavelet basis functions, by scaling and translating this mother wavelet, we can analyze the characteristics of the signal at different times and frequencies; This is a normalization constant used to ensure the invertibility of the wavelet transform; It is a complex exponential term. It is a natural constant. It is the imaginary unit. It is the center frequency. It is a Gaussian window function.

[0101] The formula for calculating continuous wavelets is:

[0102]

[0103] in, Indicates the wavelet transform result; The scaling parameter controls the scaling of the wavelet function; These are translation parameters that control the position of the wavelet function; Represents the complex conjugate of the wavelet function; scale factor This is used to ensure the energy normalization of wavelet functions at different scales.

[0104] To achieve accurate conversion from scale parameters to actual frequencies, the following mapping relationship is established:

[0105]

[0106] in, The actual physical frequency of the signal to be analyzed. The center frequency of the Morlet wavelet; This represents the sampling interval. Through this mapping, abstract scale parameters can be transformed into frequency values ​​with explicit physical meaning.

[0107] Subsequently, based on the wavelet transform results, a feature matrix containing multidimensional features in the time-frequency domain is constructed. First, the time-frequency energy distribution is calculated:

[0108]

[0109] Time-frequency energy distribution matrix It reflects the distribution of signal energy on the time-frequency plane and can clearly show the time-varying characteristics of the oscillation mode. After calculating E(a, b), the scale a of the vertical axis is converted into the actual frequency using equation (9). Thus, E(a, b) becomes a time-frequency graph with the actual frequency f as the vertical axis and time b as the horizontal axis.

[0110] Next, the scale energy spectrum is extracted:

[0111]

[0112] Scale energy spectrum It reflects the overall energy distribution of different frequency components, which helps to identify the dominant oscillation mode.

[0113] Simultaneously calculate the marginal spectrum:

[0114]

[0115] Marginal spectrum It demonstrates the variation of signal energy over time and can capture the onset and decay of oscillations.

[0116] S3. Input the wavelet feature matrix into the recognition model constructed based on the convolutional neural network, so that the recognition model can perform broadband oscillation recognition on the power system and output the recognition result. The recognition model is obtained by training the convolutional neural network based on the wavelet feature matrix constructed based on the electrical signal of the power system in the training set.

[0117] In step S3, in some embodiments, a convolutional neural network structure specifically designed for broadband oscillation identification is used, which fully considers the characteristics of oscillating signals. First, the input layer receives a wavelet feature matrix with a size of M×N, where M corresponds to the frequency resolution and N corresponds to the number of time points.

[0118] The convolutional layer employs a two-layer design. The first convolutional layer is represented as follows: The second convolutional layer is represented as Its expression is as follows:

[0119]

[0120] in, This is the output of the first convolutional layer. The kernel is used to extract local time-frequency features through convolution operations; b1 is the input variable, and ReLU is the activation function used to introduce nonlinear transformation.

[0121]

[0122] in, is the output of the second convolutional layer, W2 is the convolutional kernel of the second convolutional layer, and b2 is a constant coefficient. 64 3×3 convolutional kernels are used to further extract high-level features.

[0123] The pooling layer uses max pooling:

[0124]

[0125] This operation reduces data dimensionality while maintaining feature expressiveness, thereby enhancing the model's generalization ability.

[0126] The fully connected layer consists of three levels:

[0127]

[0128] W3, W4, and W5 are convolution kernels, and b3, b4, and b5 are constant coefficients. Furthermore, to simultaneously obtain multiple key parameters of the oscillation, the output layer employs a multi-branch design, namely:

[0129]

[0130] in, For output variables, This is an estimate of the oscillation frequency, reflecting the speed of the oscillation. This is an estimate of the oscillation amplitude, characterizing the intensity of the oscillation; This is the estimated damping ratio, describing the damping characteristics of the oscillation.

[0131] This multi-output design avoids the complexity of training multiple models separately and improves recognition efficiency.

[0132] To address the characteristics of multi-output tasks, a weighted loss function is designed:

[0133]

[0134] in, , , The weighting coefficients need to be determined through extensive experimental optimization. Their configuration should ensure the accuracy of frequency estimation while also considering the learning effect of other parameters. The loss functions for each component are defined as follows:

[0135]

[0136] in, , , Here, N represents the true values ​​of frequency, amplitude, and damping ratio, and N is the number of data points.

[0137] Subsequently, the Adam optimization algorithm is adopted, and its update rule is as follows:

[0138]

[0139] in, For the current first moment estimate, For the current second-order moment estimate, For the updated parameters, i.e., the first convolutional layer , The second convolutional layer , and fully connected layers , , , , , . This is an estimate of the first moment from the previous moment. This is an estimate of the second moment from the previous moment. The first-order moment decay rate, The second-order moment decay rate, This is the numerical stability constant. Parameter settings can be based on empirical values: Initial learning rate Finally, comprehensive training quality control is achieved through batch training, dynamic learning rate, early stopping mechanism, and regularization measures.

[0140] In some embodiments, a sliding window processing strategy is used to acquire electrical signals from the power system. The length of the analysis window is greater than a set threshold, and the sliding step size is less than the length of the analysis window to ensure that adjacent analysis windows overlap. For example, to achieve online monitoring, a sliding window processing strategy is used with a window length of 1 second and a sliding step size of 0.1 seconds. This configuration ensures that each analysis window contains a sufficient oscillation period while maintaining good real-time performance, and the overlapping window design avoids information loss.

[0141] In some embodiments, the identification results include oscillation frequency, amplitude, and damping ratio; the method further includes: detecting the reliability of the oscillation frequency, amplitude, and damping ratio based on a set reliability detection rule.

[0142] For example, multi-dimensional detection thresholds can be set to ensure the reliability of the identification results from three aspects: frequency validity, amplitude significance, and damping rationality. Specifically:

[0143]

[0144] Figure 3 This is a flowchart illustrating a power system broadband oscillation identification method provided in an embodiment of this application. Figure 3As shown in the embodiments of this application, the power system broadband oscillation identification method acquires power system voltage or current signals and preprocesses them using a three-level denoising strategy, including wavelet denoising, threshold denoising, and signal reconstruction. Wavelet transform is performed on the preprocessed signals to extract the time-frequency feature matrix as the input variable of a CNN model. A neural network model is constructed based on the CNN, with the wavelet time-frequency feature matrix as input and the oscillation frequency, amplitude, and damping ratio as output. The neural network model is trained, and parameters are optimized by combining a loss function and the Adam optimization algorithm. Various training quality control strategies are employed to improve the model's performance. The trained model is then used to identify broadband oscillations in real-time signals and outputs the identification results of the oscillation frequency, amplitude, and damping ratio. Thus:

[0145] At the mechanistic level, a deep fusion of wavelet transform and convolutional neural network (CNN) is achieved. Wavelet transform generates a time-frequency matrix with clear physical meaning through multi-resolution analysis, providing a structured representation of oscillation features; while CNN automatically extracts spatial patterns and deep features from this matrix based on its convolutional structure. The combination of the two forms a complementary advantage: wavelet transform solves the problem of weak physical interpretability of CNN, while CNN makes up for the shortcomings of wavelet transform in adaptive recognition, jointly realizing end-to-end intelligent identification of key parameters of broadband oscillations.

[0146] A multi-output regression architecture for broadband oscillations is proposed, overcoming the qualitative limitations of traditional binary classification models. This architecture can simultaneously estimate frequency, amplitude, phase, and damping ratio, achieving a complete quantitative description of the oscillation modes and providing a direct basis for subsequent accurate analysis and control decisions.

[0147] While improving recognition performance, the model effectively balances its adaptability and physical interpretability. By utilizing CNN to adaptively extract deep features from the wavelet time-frequency matrix, the model possesses high-precision recognition capabilities. Simultaneously, since the input is a wavelet time-frequency matrix with clear physical meaning, and the output is directly interpretable physical parameters, the entire model's decision-making process exhibits good interpretability, significantly enhancing its practical value and credibility in engineering practice.

[0148] Furthermore, the power system broadband oscillation identification method provided in this application, through a complete three-level denoising process of "wavelet denoising - threshold denoising - signal reconstruction" and its adaptive improved soft thresholding function, ensures effective preservation of oscillation characteristics under strong noise. The specific data structure and combination method of using the time-frequency feature matrix generated by wavelet transform as input to the convolutional neural network achieves complementary advantages. Moreover, the multi-branch CNN network architecture, which simultaneously and directly outputs multiple parameters such as oscillation frequency, amplitude, and damping ratio, realizes the transformation from qualitative to quantitative analysis. The combined application of the weighted loss function and the Adam optimizer can simultaneously optimize the estimation accuracy of frequency, amplitude, and damping ratio.

[0149] As can be seen, the power system broadband oscillation identification method provided in this application solves the problems of low identification accuracy and poor adaptability of broadband oscillations in the prior art. By combining the time-frequency analysis capability of wavelet transform and the self-learning characteristics of neural networks, high-precision and adaptive broadband oscillation identification is achieved.

[0150] Furthermore, in the signal processing stage, S-transform or Hilbert-Huang transform can be used instead of wavelet transform to generate time-frequency features, but wavelet transform achieves a better balance between computational efficiency and time-frequency resolution. In the pattern recognition stage, machine learning methods such as support vector machines or decision trees can be used, or unsupervised architectures such as autoencoders can be selected. However, convolutional neural networks (CNNs) have significant advantages in extracting high-dimensional spatial features from time-frequency images due to their local connectivity and weight sharing mechanism. Their supervised, feature-extraction-oriented design concept also differs from the autoencoder's focus on data reconstruction, making it more suitable for the multi-parameter regression task requirements in this scheme.

[0151] Based on the same inventive concept, embodiments of this application also provide a power system broadband oscillation identification device.

[0152] Figure 4 This is a schematic diagram of the structure of a power system broadband oscillation identification device provided in an embodiment of this application, as shown below. Figure 4 As shown, the device includes:

[0153] The acquisition and preprocessing module 21 is used to acquire electrical signals of the power system based on the Nyquist sampling theorem and preprocess the electrical signals to obtain preprocessed electrical signals.

[0154] The conversion module 22 is used to perform wavelet transform on the preprocessed electrical signal and construct a wavelet feature matrix based on the wavelet transform result. The wavelet feature matrix includes wavelet time-frequency energy distribution, scale energy spectrum and marginal spectrum.

[0155] The recognition module 23 is used to input the wavelet feature matrix into the recognition model constructed based on the convolutional neural network, so that the recognition model can perform broadband oscillation recognition on the power system and output the recognition result. The recognition model is obtained by training the convolutional neural network based on the wavelet feature matrix constructed based on the electrical signal of the power system in the training set.

[0156] The power system broadband oscillation identification device provided in this application embodiment solves the problems of low broadband oscillation identification accuracy and poor adaptability in the prior art. By combining the time-frequency analysis capability of wavelet transform and the self-learning characteristics of neural networks, high-precision and adaptive broadband oscillation identification is achieved.

[0157] The embodiments of the apparatus provided in this application can be used to execute the processing flow of the above method embodiments, and will not be repeated here. Please refer to the detailed description of the above method embodiments.

[0158] Figure 5 This is a schematic diagram of the physical structure of an electronic device provided in an embodiment of this application, as shown below. Figure 5 As shown, the electronic device may include a processor 301, a communications interface 302, a memory 303, and a communication bus 304, wherein the processor 301, the communications interface 302, and the memory 303 communicate with each other via the communication bus 304. The processor 301 may call logical instructions in the memory 303 to execute the methods described in any of the above embodiments.

[0159] Furthermore, the logical instructions in the aforementioned memory 303 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0160] This embodiment discloses a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions, and when the program instructions are executed by a computer, the computer can perform the methods provided in the above-described method embodiments.

[0161] This embodiment provides a computer-readable storage medium storing a computer program that causes the computer to perform the methods provided in the above-described method embodiments.

[0162] 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.

[0163] 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.

[0164] 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.

[0165] 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.

[0166] In the description of this specification, the references to terms such as "an embodiment," "a specific embodiment," "some embodiments," "for example," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0167] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of this application. It should be understood that the above descriptions are merely specific embodiments of this application and are not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for identifying broadband oscillations in a power system, characterized in that, include: The electrical signals of the power system are acquired based on the Nyquist sampling theorem, and the electrical signals are preprocessed to obtain the preprocessed electrical signals. The preprocessed electrical signal is subjected to wavelet transform, and a wavelet feature matrix is ​​constructed based on the wavelet transform result. The wavelet feature matrix includes wavelet time-frequency energy distribution, scale energy spectrum, and marginal spectrum. The wavelet feature matrix is ​​input into the recognition model constructed based on the convolutional neural network, so that the recognition model can perform broadband oscillation recognition on the power system and output the recognition result. The recognition model is obtained by training the convolutional neural network based on the wavelet feature matrix constructed based on the electrical signal of the power system in the training set.

2. The method according to claim 1, characterized in that, The preprocessing of the electrical signal to obtain a preprocessed electrical signal includes: The electrical signal is decomposed into approximation coefficients and detail coefficients, wherein the approximation coefficients and the detail coefficients correspond to the low-frequency trend and high-frequency details of the electrical signal, respectively. Threshold denoising is performed on the detail coefficients to obtain the denoised detail coefficients; The signal is reconstructed based on the approximation coefficients and the denoised detail coefficients to obtain the reconstructed electrical signal; The reconstructed electrical signal is standardized to obtain a preprocessed electrical signal.

3. The method according to claim 2, characterized in that, The detail coefficients are then denoised using the following soft thresholding function to obtain the denoised detail coefficients: in, Indicates the detail coefficient after noise reduction; Indicates detail coefficients; Indicates the threshold parameter. , The noise standard deviation is estimated using the median of the detail coefficients. Indicates the number of data points.

4. The method according to claim 1, characterized in that, The preprocessed electrical signal is subjected to wavelet transform based on the following formula: in, Indicates the wavelet transform result; This represents the scaling parameter, which controls the scaling of the wavelet function; This represents the preprocessed electrical signal; This represents the translation parameter, which controls the position of the wavelet function; Represents the complex conjugate of the wavelet function; This represents the scaling factor, used to ensure energy normalization of wavelet functions at different scales.

5. The method according to claim 1, characterized in that, The structure of the convolutional neural network is as follows: The input layer is used to receive the wavelet time-frequency feature matrix; The first convolutional layer, the expression for the first convolutional layer is: ,in, This is the output of the first convolutional layer. For input variables, b1 is the convolution kernel, b1 is a constant coefficient, and ReLU is the activation function; The second convolutional layer, the expression for the second convolutional layer is: ,in, is the output of the second convolutional layer, W2 is the convolutional kernel of the second convolutional layer, and b2 is a constant coefficient; Pooling layer, the expression for pooling layer is: ; The fully connected layer consists of three levels. , , , Where W3, W4, and W5 are convolution kernels, and b3, b4, and b5 are constant coefficients; The output layer, the expression for the output layer is: ,in, Indicates the output variable. This represents the estimated oscillation frequency, reflecting the speed of the oscillation. This represents the estimated amplitude of the oscillation, characterizing the intensity of the oscillation; This represents the estimated damping ratio, describing the damping characteristics of the oscillation.

6. The method according to claim 5, characterized in that, When training the convolutional neural network, the following weighted loss function is used: in, , , The weighting coefficients and the loss functions for each component are as follows: Frequency estimation loss ; Amplitude estimation loss ; Damping ratio estimated loss ; in, , , These are the actual values ​​of frequency, amplitude, and damping ratio, respectively, and N is the number of data points.

7. The method according to claim 1, characterized in that, A sliding window processing strategy is used to collect electrical signals from the power system. The length of the analysis window is greater than a set threshold, and the sliding step size is set to be less than the length of the analysis window to ensure that adjacent analysis windows overlap.

8. The method according to claim 1, characterized in that, The identification results include oscillation frequency, amplitude, and damping ratio; the method further includes: Based on the established reliability detection rules, the reliability of the oscillation frequency, amplitude, and damping ratio is detected.

9. A power system broadband oscillation identification device, characterized in that, include: The acquisition and preprocessing module is used to acquire electrical signals of the power system based on the Nyquist sampling theorem and preprocess the electrical signals to obtain preprocessed electrical signals. The conversion module is used to perform wavelet transform on the preprocessed electrical signal and construct a wavelet feature matrix based on the wavelet transform result. The wavelet feature matrix includes wavelet time-frequency energy distribution, scale energy spectrum and marginal spectrum. The recognition module is used to input the wavelet feature matrix into a recognition model constructed based on a convolutional neural network, so that the recognition model can perform broadband oscillation recognition on the power system and output the recognition result. The recognition model is obtained by training the convolutional neural network based on the wavelet feature matrix constructed based on the electrical signals of the power system in the training set.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method as described in any one of claims 1 to 8.

Images