Novel broadband oscillation wide-area monitoring method for power system

Abstract

The invention discloses a novel broadband oscillation wide-area monitoring method for an electric power system. The method comprises the following steps: acquiring electric signal waveforms of key nodes of the electric power system at each substation; each substation performs modal decomposition and feature extraction on the electric signal waveform, identifies the broadband oscillation type to obtain a preliminary pre-judgment result, and stores a corresponding broadband oscillation electric signal, namely a to-be-compressed broadband oscillation electric signal; each substation performs secondary sub-sampling on the broadband oscillation electric signal to be compressed to obtain a compressed signal, and uploads the compressed signal and the preliminary pre-judgment result to the master station; the master station performs feature extraction on the compressed signal by using a broadband oscillation classification and recognition model, and obtains a final broadband oscillation type by comparing preliminary pre-judgment results; and the master station constructs a broadband oscillation signal reconstruction model, the input quantity is the compressed signals uploaded by the substations, the output quantity is the reconstructed broadband oscillation signals, and the master station comprehensively analyzes the reconstructed broadband oscillation signals, so that the global monitoring of the broadband oscillation wide area of the novel power system is effectively realized.

Description

Technical Field

[0001] The present invention relates to the technical field of wide - area monitoring of power systems, and in particular to a novel wide - area monitoring method for broadband oscillations in a power system. Background Art

[0002] In recent years, the grid - connection ratio of renewable energy and power electronic devices has increased significantly. In the future, the "dual - high" characteristics of a new power system dominated by new energy will be more prominent. Compared with traditional power systems, in "dual - high" power systems, "source - grid - load" are all developing towards the direction of power electronics. The broadband oscillations caused by the interaction between a large number of renewable energy units and existing power generation equipment, transmission networks, power loads and themselves, as well as the resulting engineering safety accidents, have become increasingly frequent. The oscillations in the new power system exhibit characteristics such as wide frequency bands, high noise, multi - modality, strong time - variability, strong non - linearity and spatio - temporal distribution characteristics. Moreover, the inducing causes are complex and diverse, seriously threatening the stability of the power system and the safe and economic operation of electrical equipment, and severely restricting the effective consumption of renewable energy such as wind power and photovoltaic power. Therefore, it is of great significance to effectively monitor broadband oscillations and provide a data basis for mechanism analysis, source tracing and positioning, online monitoring and early warning, and preventive control of broadband oscillations.

[0003] Existing broadband oscillation monitoring methods still have many limitations in practical applications. First, the phasor measurement unit (PMU) device is based on the Nyquist sampling theorem and can only monitor partial sub - synchronous / supersynchronous oscillations in the range of 10 - 40 Hz and 60 - 90 Hz. For sub - synchronous / supersynchronous oscillations with frequencies higher than low - frequency oscillations but within twice the power frequency range, accurate parameters of broadband oscillations cannot be directly obtained from the phasor data collected by the PMU. Second, due to the limitations of communication bandwidth and the Nyquist sampling theorem, the master station can only obtain oscillation signals within half of the uplink frequency from the slave station, and can only effectively monitor and analyze partial sub - synchronous / supersynchronous oscillation signals within 50 Hz of the power frequency, unable to obtain the complete oscillation information of broadband oscillations and difficult to effectively monitor the entire broadband oscillation. In addition, at the master - station dispatching end, since different types of broadband oscillations may occur simultaneously, there is a lack of a wide - area monitoring and identification method for multi - source oscillations, unable to accurately trace the oscillation source, analyze the propagation path and its influence range, and difficult to provide support for the online monitoring and preventive control protection of the power grid, affecting the safe and stable operation of the power grid. Summary of the Invention

[0004] The purpose of the present invention is to overcome the disadvantages and deficiencies of the prior art, and provide a novel wide - area monitoring method for broadband oscillations in a power system, which can effectively achieve global monitoring of broadband oscillations in a new power system.

[0005] To achieve the above object, the technical solution provided by the present invention is: a wide-area monitoring method for broadband oscillations in a new power system, comprising the following steps:

[0006] S1. At each substation, use the synchronized measurement unit PMU of the wide-area measurement system to collect the electrical signal waveforms of key nodes in the power system, and store the collected electrical signal waveforms as time series;

[0007] S2. Each substation performs modal decomposition and feature extraction on the electrical signal waveforms, extracts the oscillation characteristics of the electrical signal waveforms, identifies the broadband oscillation types, obtains a preliminary prediction result, and saves the corresponding broadband oscillation electrical signals, namely the broadband oscillation electrical signals to be compressed;

[0008] S3. Each substation uses the compressive sensing theory to perform secondary sub-sampling on the broadband oscillation electrical signals to be compressed to obtain compressed signals, and uploads the compressed signals and the preliminary prediction results to the master station through a communication network;

[0009] S4. At the master station, use the broadband oscillation classification and recognition model based on deep dictionary learning to extract features from the uploaded compressed signals, and obtain the final broadband oscillation types by comparing with the preliminary prediction results uploaded by the substation;

[0010] S5. At the master station, use compressive sensing to construct a broadband oscillation signal reconstruction model, with the input being the compressed signals uploaded by each substation and the output being the reconstructed broadband oscillation signals. The master station performs comprehensive analysis on the reconstructed broadband oscillation signals to achieve global monitoring of broadband oscillations in the new power system over a wide area.

[0011] Further, in step S1, the electrical signal waveforms refer to the electrical signal waveforms of each time period obtained by real-time sampling using the synchronized measurement unit PMU of the wide-area measurement system, and the electrical signals include line current, node voltage, active power, reactive power, and rotor angle.

[0012] Further, step S2 includes the following steps:

[0013] S21. Use the adaptive modal decomposition method to perform modal decomposition on the electrical signal waveforms to obtain the main modal components of the electrical signal waveforms, and perform denoising processing on the main modal components by combining correlation analysis, kurtosis criterion, and soft threshold to obtain the denoised main modal components;

[0014] S22. Use the Hilbert transform to extract features from the denoised main modal components to obtain oscillation feature quantities, and the oscillation feature quantities include oscillation amplitude and oscillation frequency;

[0015] S23. Based on the preset broadband oscillation frequency threshold and amplitude threshold, determine whether the obtained oscillation amplitude and oscillation frequency are greater than the preset broadband oscillation frequency threshold and amplitude threshold. If both the oscillation amplitude and oscillation frequency simultaneously meet the condition of being greater than or equal to the preset broadband oscillation frequency threshold and amplitude threshold, then it is identified as a broadband oscillation, obtaining a preliminary prediction result, and saving the corresponding broadband oscillation electrical signal, that is, the broadband oscillation electrical signal to be compressed.

[0016] Further, the step S3 includes the following steps:

[0017] S31. At the substation, based on the broadband oscillation electrical signal to be compressed, use the sparsity - adaptive K - SVD dictionary learning method to design the substation classification sparse basis, that is, the substation classification sparse transform domain, denoted as ψ = [ψ1, ψ2, …, ψ m , … ψ K , ψ is the set of classification sparse bases corresponding to various broadband oscillations, ψ m is the classification sparse basis corresponding to the m - th type of broadband oscillation, m = 1, 2, …, K; where the sparsity - adaptive K - SVD dictionary learning method enables the sparsity, that is, the number of non - zero coefficients in the representation of each signal on the dictionary, to be adaptively adjusted according to the characteristics of the signal;

[0018] S32. Use the Logistic chaotic mapping method shown in the following formula (1) to construct a random measurement matrix Φ that is uncorrelated with the substation classification sparse basis: First, determine the matrix dimension as M×N, and select appropriate initial values and bifurcation parameters. Then, generate a chaotic sequence of length M×N through the Logistic chaotic mapping method shown in the following formula (1), and fill this chaotic sequence into the M×N matrix in sequence to obtain the original random measurement matrix Φ'. Then, perform normalization processing on the original random measurement matrix Φ' to obtain the random measurement matrix Φ that is uncorrelated with the substation classification sparse basis. This random measurement matrix Φ has good numerical stability and can effectively project high - dimensional signals into a low - dimensional space in compressive sensing;

[0019] h n+1 = μh n (1 - h n )(1)

[0020] In the formula, h n+1 、h n are the chaotic variable values of the (n + 1) - th and n - th iterations respectively, and μ is the bifurcation parameter;

[0021] S33. Based on a certain substation classification sparse basis Ψ m 、random measurement matrix Φ and the compressive sensing theory shown in the following formula (2), compress the broadband oscillation electrical signal to be compressed, that is, perform secondary subsampling, to obtain a compressed signal y, that is, a one - dimensional observation signal;

[0022] y = Φx = ΦΨ m s = Θs (2)

[0023] Wherein, x is the original signal with a length of N, that is, the broadband oscillation electrical signal to be compressed. M represents the number of rows of the observation matrix, that is, the dimension of the compressed signal y, and satisfies N >> M; S is the coefficient vector of the original signal x in a certain substation classification sparse basis Ψ m and is sparse, with a sparsity of G, and satisfies G << N; Θ = ΦΨ, which is an M×N-dimensional sensing matrix.

[0024] Furthermore, in step S4, at the master station, a broadband oscillation classification and recognition model is constructed using deep dictionary learning, and the broadband oscillation classification and recognition model is used to extract oscillation feature quantities from the uploaded compressed signal to identify the broadband oscillation type; First, the master station normalizes the one-dimensional observation signals of different broadband oscillation types received from each substation after compression, and sets classification labels according to the preliminary prediction results of the broadband oscillation type to form a one-dimensional observation signal sample data set; Then, the Gram angular field method is used to convert the one-dimensional observation signals in the one-dimensional observation signal sample data set into two-dimensional Gram feature maps, and corresponding classification labels are set for each two-dimensional Gram feature map; The two-dimensional Gram feature maps and the corresponding classification labels are used to construct a two-dimensional image sample data set, which is divided into a training set and a test set;

[0025] Based on the deep learning theory and the dictionary learning theory, a deep dictionary learning method is proposed for constructing a broadband oscillation classification and recognition model. The model structure includes:

[0026] Input module: The input is two-dimensional Gram feature maps of various categories, providing basic data for subsequent deep dictionary learning;

[0027] Classification deep learning dictionary module: It contains deep learning dictionary modules of multiple categories. The input is the two-dimensional Gram feature maps of various categories of the input module, and the output is the updated deep learning dictionaries of various categories; Each category's deep learning dictionary module is composed of multiple cascaded deep dictionary learning modules. Each deep dictionary learning module includes a sparse coding layer and a dictionary update layer. Through multi-layer dictionaries, local and global features of the two-dimensional Gram feature maps are gradually learned and extracted to improve the feature extraction ability; The input of each deep dictionary learning module is the two-dimensional Gram feature map, and the output is the updated deep learning dictionaries of various categories; Among them, the input of the sparse coding layer is the data output by the previous layer, that is, the input module or the previous deep dictionary learning module, and the dictionary matrix of this layer. By optimizing the solution, the sparse representation coefficients are output, thereby extracting the key features of the input signal; The input of the dictionary update layer is the sparse representation coefficients output by the sparse coding layer. Using the dictionary learning algorithm based on nonlinear transformation, the dictionary matrix of this layer is updated to obtain the updated dictionary of this layer, thereby improving the nonlinear expression ability and suppressing the noise and non-significant features in the signal;

[0028] Sparse representation coefficient calculation module: The input is the original data of each two-dimensional Gram feature map and the updated deep learning dictionaries of various categories. Using the sparse representation method, calculate the sparse representation coefficients of each two-dimensional Gram feature map under the deep learning dictionaries of various categories, and use the signal reconstruction algorithm to obtain the reconstructed two-dimensional Gram feature maps under the deep learning dictionaries of various categories, and compare them with the original data. Finally, output the reconstruction residuals of each two-dimensional Gram feature map under the deep learning dictionaries of various categories. The smaller the error, the higher the matching degree with this category;

[0029] Classification decision-making module: The input is the reconstruction residuals of each two-dimensional Gram feature map under the deep learning dictionaries of various categories. Select the category with the smallest reconstruction residual as the sample category, and the output is the category label of each two-dimensional Gram feature map;

[0030] Finally, use the test set divided from the two-dimensional image sample data set to test the broadband oscillation classification and recognition model, further optimize and adjust the model parameters, and obtain the broadband oscillation classification and recognition model with the optimal performance.

[0031] Furthermore, the step S5 includes the following steps:

[0032] S51. Based on the one-dimensional observation signal sample data set, the master station uses the K-SVD dictionary learning method with adaptive sparsity to design the master station classification sparse basis, that is, the master station classification sparse transform domain;

[0033] S52. The master station uses the random measurement matrix generation model constructed by the Logistic chaotic mapping method shown in formula (1) to reconstruct the random measurement matrix, and multiplies it by the master station classification sparse basis to obtain the sensing matrix;

[0034] S53. Based on the compressive sensing theory and the sensing matrix, construct a broadband oscillation signal reconstruction model, and the model structure is as follows:

[0035] Input layer: The input is the one-dimensional observation signal;

[0036] Signal reconstruction layer: Use the signal reconstruction algorithm based on sparse representation to reconstruct the one-dimensional observation signal, and obtain the reconstructed sparse signal through the deep learning dictionaries of various categories and the sparse representation coefficients;

[0037] Signal recovery layer: Recover the reconstructed sparse signal and output the recovered broadband oscillation signal, that is, the reconstructed broadband oscillation signal;

[0038] Finally, use the training set and the test set to train and test the broadband oscillation signal reconstruction model, optimize the model parameters, and obtain the broadband oscillation signal reconstruction model with the optimal performance;

[0039] S54. Use the broadband oscillation signal reconstruction model with the optimal performance. The input is the compressed signals uploaded by each sub-station, and the output is the reconstructed broadband oscillation signal.

[0040] S55. The master station conducts comprehensive analysis based on the reconstructed broadband oscillation signal, including time-frequency analysis, wide-area monitoring, mechanism analysis, disturbance traceability and positioning, data mining, dispatching decision-making, and preventive protection and control.

[0041] Furthermore, the signal reconstruction algorithms based on sparse representation include matching pursuit algorithm, orthogonal matching pursuit algorithm, regularized matching pursuit algorithm, basis pursuit algorithm, and alternating direction method of multipliers.

[0042] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0043] In the sub-station, the present invention uses the adaptive mode decomposition method to decompose the waveform of the broadband oscillation electrical signal, obtains the main mode components of the waveform of the broadband oscillation electrical signal, and combines correlation analysis, kurtosis criterion, and soft threshold to denoise the main mode components, obtaining the denoised main mode components. The Hilbert transform is used to extract the oscillation feature quantities from the denoised main mode components, and combined with a preset threshold, the broadband oscillation is identified to obtain a preliminary prediction result. The adaptive mode decomposition method and the Hilbert transform can quickly obtain the oscillation features and types of various electrical signals, quickly obtain a preliminary prediction result, improve the signal quality transmitted to the master station, and reduce the computing power and storage requirements of the sub-station industrial control computer.

[0044] In the sub-station, the present invention uses the designed classification sparse basis and random measurement matrix with the functions of atomic self-update and self-optimization to perform dimensionality reduction processing on the detected broadband oscillation data, greatly improving the signal compression ratio, reducing the storage and transmission burden brought by the fixed sparse basis, improving the data compression and transmission efficiency, being conducive to realizing the transmission of high-frequency broadband oscillation data under the existing communication bandwidth and PMU data transmission frequency, and ensuring the real-time performance and effectiveness of the data.

[0045] The present invention proposes a method for broadband oscillation identification and reconstruction based on Gram angle field, compressive sensing, and deep dictionary learning. The master station can directly use the compressed data to quickly and accurately extract the multi-modal oscillation features of the broadband oscillation and identify the types. While ensuring the multi-oscillation type and multi-modal feature discrimination ability, the extracted fault features have good hierarchy and physical meaning. At the same time, based on the compressed data uploaded by each sub-station, the signal reconstruction algorithm can be used to accurately reconstruct the original broadband oscillation signal, providing support for subsequent oscillation traceability, suppression, and power grid stability control, etc. Description of the Drawings

[0046] Figure 1 It is the overall flowchart of the method of the present invention. Detailed implementation manners

[0047] The present invention will be further described in detail below in conjunction with embodiments and the accompanying drawings, but the implementation manners of the present invention are not limited thereto.

[0048] As Figure 1 shown, this embodiment discloses a wide-area monitoring method for broadband oscillation in a new type of power system, and the specific situation is as follows:

[0049] S1. At each substation, use the synchronized measurement unit (PMU) of the wide-area measurement system to collect the electrical signal waveforms of the key nodes of the power system, and store the collected electrical signal waveforms as time series; wherein, the electrical signal waveforms refer to the electrical signal waveforms of each period obtained by real-time sampling through the synchronized measurement unit (PMU) of the wide-area measurement system, and the electrical signals include line current, node voltage, active power, reactive power, and rotor angle.

[0050] S2. Each substation performs modal decomposition and feature extraction on the electrical signal waveforms, extracts the oscillation characteristics of the electrical signal waveforms, identifies the broadband oscillation type, obtains a preliminary prediction result, and saves the corresponding broadband oscillation electrical signals, that is, the broadband oscillation electrical signals to be compressed, including the following steps:

[0051] S21. Use the adaptive modal decomposition method to perform modal decomposition on the electrical signal waveforms to obtain the main modal components of the electrical signal waveforms, and perform denoising processing on the main modal components by combining correlation analysis, kurtosis criterion, and soft threshold to obtain the denoised main modal components;

[0052] S22. Use the Hilbert transform to perform feature extraction on the denoised main modal components to obtain oscillation feature quantities, and the oscillation feature quantities include oscillation amplitude and oscillation frequency;

[0053] S23. Based on the preset broadband oscillation frequency threshold and amplitude threshold, judge whether the obtained oscillation amplitude and oscillation frequency are greater than the preset broadband oscillation frequency threshold and amplitude threshold. If both the oscillation amplitude and oscillation frequency are greater than or equal to the preset broadband oscillation frequency threshold and amplitude threshold, it is identified as broadband oscillation, obtain a preliminary prediction result, and save the corresponding broadband oscillation electrical signals, that is, the broadband oscillation electrical signals to be compressed.

[0054] S3. Each substation uses the compressive sensing theory to perform secondary sub-sampling on the broadband oscillation electrical signals to be compressed to obtain compressed signals, and uploads the compressed signals and the preliminary prediction results to the master station through the communication network, including the following steps:

[0055] S31. At the substation, based on the broadband oscillation electrical signals to be compressed, use the sparsity-adaptive K-SVD dictionary learning method to design the substation classification sparse basis, that is, the substation classification sparse transform domain, denoted as ψ[ψ1, ψ2,…, ψm ,…ψ K , where ψ is the set of classification sparse bases corresponding to various broadband oscillations, and ψ[[ID=]4] m is the classification sparse base corresponding to the m-th type of broadband oscillation, where m = 1, 2, …, K; among them, the sparsity-adaptive K-SVD dictionary learning method enables the sparsity, that is, the number of non-zero coefficients in the representation of each signal on the dictionary, to be adaptively adjusted according to the characteristics of the signal;

[0056] S32. Use the Logistic chaotic mapping method shown in the following formula (1) to construct a random measurement matrix Φ that is uncorrelated with the substation classification sparse base: First, determine the matrix dimension as M×N, and select appropriate initial values and bifurcation parameters. Then, generate a chaotic sequence of length M×N through the Logistic chaotic mapping method shown in the following formula (1), and fill this chaotic sequence into the M×N matrix in sequence to obtain the original random measurement matrix Φ'. Then, perform normalization processing on the original random measurement matrix Φ' to obtain the random measurement matrix Φ that is uncorrelated with the substation classification sparse base. This random measurement matrix Φ has good numerical stability and can effectively project high-dimensional signals into a low-dimensional space in compressive sensing;

[0057] h n+1 = μh n (1 - h n ) (1)

[0058] In the formula, h n+1 and h n are the chaotic variable values of the (n + 1)-th and n-th iterations respectively, and μ is the bifurcation parameter;

[0059] S33. Based on a substation classification sparse base Ψ m , a random measurement matrix Φ, and the compressive sensing theory shown in the following formula (2), compress the broadband oscillation electrical signal to be compressed, that is, perform secondary subsampling, to obtain a compressed signal y, that is, a one-dimensional observation signal;

[0060] y = Φx = ΦΨ m s = Θs(2)

[0061] In the formula, x is the original signal of length N, that is, the broadband oscillation electrical signal to be compressed, Φ is an M×N-dimensional random measurement matrix, M represents the number of rows of the observation matrix, that is, the dimension of the compressed signal y, and satisfies N >> M; S is the coefficient vector of the original signal x under a substation classification sparse base Ψ m and is sparse, with a sparsity of G, and satisfies G << N; Θ = ΦΨ, which is an M×N-dimensional sensing matrix.

[0062] S4. At the master station, use the broadband oscillation classification and recognition model based on deep dictionary learning to extract oscillation feature quantities from the uploaded compressed signals, identify the broadband oscillation types, and obtain the final broadband oscillation types by comparing with the preliminary prediction results uploaded by the slave stations, as follows:

[0063] First, the master station normalizes the one-dimensional observation signals of different broadband oscillation types received from each slave station after compression, and sets classification labels according to the preliminary prediction results of the broadband oscillation types to form a one-dimensional observation signal sample data set; then, uses the Gram angular field method to convert the one-dimensional observation signals in the one-dimensional observation signal sample data set into two-dimensional Gram feature maps, and sets corresponding classification labels for each two-dimensional Gram feature map; constructs a two-dimensional image sample data set from the two-dimensional Gram feature maps and the corresponding classification labels, and divides it into a training set and a test set;

[0064] Based on deep learning theory and dictionary learning theory, a deep dictionary learning method is proposed to construct a broadband oscillation classification and recognition model. The model structure includes:

[0065] Input module: The input is two-dimensional Gram feature maps of various categories, providing basic data for subsequent deep dictionary learning;

[0066] Classification deep learning dictionary module: It contains deep learning dictionary modules of multiple categories. The input is the two-dimensional Gram feature maps of various categories in the input module, and the output is the updated deep learning dictionaries of various categories; each category's deep learning dictionary module is composed of multiple cascaded deep dictionary learning modules. Each deep dictionary learning module includes a sparse coding layer and a dictionary update layer, and gradually learns to extract local and global features of the two-dimensional Gram feature maps through multiple layers of dictionaries to improve the feature extraction ability; the input of each deep dictionary learning module is the two-dimensional Gram feature map, and the output is the updated deep learning dictionaries of various categories; among them, the input of the sparse coding layer is the data output by the previous layer, that is, the input module or the previous deep dictionary learning module, and the dictionary matrix of this layer. By optimizing the solution, the sparse representation coefficients are output, so as to extract the key features of the input signal; the input of the dictionary update layer is the sparse representation coefficients output by the sparse coding layer. Use the dictionary learning algorithm based on nonlinear transformation to update the dictionary matrix of this layer to obtain the updated dictionary of this layer, so as to improve the nonlinear expression ability and suppress the noise and non-significant features in the signal;

[0067] Sparse representation coefficient calculation module: The input is the original data of each Gram two-dimensional feature map and the updated deep learning dictionaries of various categories. Using the sparse representation method, calculate the sparse representation coefficients of each Gram two-dimensional feature map under the deep learning dictionaries of various categories, and use the signal reconstruction algorithm to obtain the reconstructed Gram two-dimensional feature maps under the deep learning dictionaries of various categories. Compare with the original data, and finally output the reconstruction residuals of each Gram two-dimensional feature map under the deep learning dictionaries of various categories. The smaller the error, the higher the matching degree with this category;

[0068] Classification decision module: The input is the reconstruction residuals of each Gram two-dimensional feature map under the deep learning dictionaries of various categories. Select the category with the smallest reconstruction residual as the sample category, and the output is the category label of each Gram two-dimensional feature map;

[0069] Finally, use the test set divided from the two-dimensional image sample data set to test the broadband oscillation classification and recognition model, further optimize and adjust the model parameters, and obtain the broadband oscillation classification and recognition model with the best performance.

[0070] S5. At the master station, use compressive sensing to construct a broadband oscillation signal reconstruction model. The input quantity is the compressed signals uploaded by each sub-station, and the output quantity is the reconstructed broadband oscillation signals. The master station comprehensively analyzes the reconstructed broadband oscillation signals to achieve global monitoring of the broadband oscillation in the wide area of the new power system, including the following steps:

[0071] S51. Based on the one-dimensional observation signal sample data set, the master station uses the sparse degree adaptive K-SVD dictionary learning method to design the master station classification sparse basis, that is, the master station classification sparse transform domain;

[0072] S52. The master station uses the random measurement matrix generation model constructed by the Logistic chaotic mapping method shown in formula (1) to reconstruct the random measurement matrix, and multiplies it by the master station classification sparse basis to obtain the sensing matrix;

[0073] S53. Based on the compressive sensing theory and the sensing matrix, construct a broadband oscillation signal reconstruction model. The model structure is as follows:

[0074] Input layer: The input is the one-dimensional observation signal;

[0075] Signal reconstruction layer: Use the signal reconstruction algorithm based on sparse representation to reconstruct the one-dimensional observation signal. Through the deep learning dictionaries of various categories and the sparse representation coefficients, obtain the reconstructed sparse signal; The signal reconstruction algorithms based on sparse representation include matching pursuit algorithm, orthogonal matching pursuit algorithm, regularized matching pursuit algorithm, basis pursuit algorithm, and alternating direction multiplier method, etc.;

[0076] Signal recovery layer: Recovers the reconstructed sparse signal and outputs the recovered broadband oscillation signal, i.e., the reconstructed broadband oscillation signal.

[0077] Finally, use the training set and the test set to train and test the broadband oscillation signal reconstruction model, optimize the model parameters, and obtain the broadband oscillation signal reconstruction model with the optimal performance.

[0078] S54. Use the broadband oscillation signal reconstruction model with the optimal performance. The input is the compressed signal uploaded by each sub-station, and the output is the reconstructed broadband oscillation signal.

[0079] S55. The master station conducts comprehensive analysis based on the reconstructed broadband oscillation signal, including time-frequency analysis, wide-area monitoring, mechanism analysis, disturbance source tracing and positioning, data mining, dispatching decision-making, and preventive protection and control, etc.

[0080] The above embodiments are the preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications made without departing from the spirit and principle of the present invention shall be equivalent replacement methods and shall be included in the protection scope of the present invention.

Claims

1. A wide-area monitoring method for broadband oscillations in a new type of power system, characterized in that, Including the following steps: S1. At each substation, use the synchronous measurement unit PMU of the wide area measurement system to collect the electrical signal waveforms of the key nodes of the power system, and store the collected electrical signal waveforms as time series; S2. Each substation performs modal decomposition and feature extraction on the electrical signal waveforms, extracts the oscillation characteristics of the electrical signal waveforms, identifies the wideband oscillation types, obtains the preliminary prediction results, and saves the corresponding wideband oscillation electrical signals, that is, the wideband oscillation electrical signals to be compressed; S3. Each substation uses the compressive sensing theory to perform secondary sub-sampling on the wideband oscillation electrical signals to be compressed, obtains the compressed signals, and uploads the compressed signals and the preliminary prediction results to the master station through the communication network; S4. At the master station, use the wideband oscillation classification and recognition model based on deep dictionary learning to extract the features of the uploaded compressed signals, and obtain the final wideband oscillation types by comparing with the preliminary prediction results uploaded by the substation; S5. At the master station, use compressive sensing to construct a wideband oscillation signal reconstruction model, with the input being the compressed signals uploaded by each substation and the output being the reconstructed wideband oscillation signals. The master station performs comprehensive analysis on the reconstructed wideband oscillation signals to achieve global monitoring of wideband oscillations in the new power system; 2. A wide - area monitoring method for wide - frequency oscillation in a new - type power system according to claim 1, characterized in that, In step S1, the electrical signal waveforms refer to the electrical signal waveforms of each time period obtained by real-time sampling through the synchronous measurement unit PMU of the wide area measurement system, and the electrical signals include line current, node voltage, active power, reactive power, and rotor angle; 3. A wide - area monitoring method for broadband oscillations in a new power system according to claim 2, characterized in that, The step S2 includes the following steps: S21. Use the adaptive modal decomposition method to perform modal decomposition on the electrical signal waveforms, obtain the main modal components of the electrical signal waveforms, and perform denoising processing on the main modal components by combining correlation analysis, kurtosis criterion, and soft threshold to obtain the denoised main modal components; S22. Use the Hilbert transform to perform feature extraction on the denoised main modal components to obtain oscillation feature quantities, and the oscillation feature quantities include oscillation amplitude and oscillation frequency; S23. Based on the preset wideband oscillation frequency threshold and amplitude threshold, judge whether the obtained oscillation amplitude and oscillation frequency are greater than the preset wideband oscillation frequency threshold and amplitude threshold. If both the oscillation amplitude and oscillation frequency are greater than or equal to the preset wideband oscillation frequency threshold and amplitude threshold, it is identified as a wideband oscillation, obtain the preliminary prediction results, and save the corresponding wideband oscillation electrical signals, that is, the wideband oscillation electrical signals to be compressed; 4. A wide - area monitoring method for broadband oscillations in a novel power system according to claim 3, characterized in that, The step S3 includes the following steps: S31. At the substation, based on the broadband oscillating electrical signal to be compressed, design the substation classification sparse basis by using the sparsity adaptive K-SVD dictionary learning method, that is, the substation classification sparse transform domain, denoted as ψ = [ψ1, ψ2, …, ψ m , … ψ K , where ψ is the set of classification sparse bases corresponding to various broadband oscillations, and ψ m is the classification sparse basis corresponding to the m-th type of broadband oscillation, m = 1, 2, …, K; among them, the sparsity adaptive K-SVD dictionary learning method enables the sparsity, that is, the number of non-zero coefficients in the representation of each signal on the dictionary, to be adaptively adjusted according to the characteristics of the signal. S32. Use the Logistic chaotic mapping method shown in the following formula (1) to construct a random measurement matrix Φ that is uncorrelated with the substation classification sparse basis: First, determine the matrix dimension as M×N, and select appropriate initial values and bifurcation parameters. Then, generate a chaotic sequence of length M×N through the Logistic chaotic mapping method shown in the following formula (1), and sequentially fill this chaotic sequence into the M×N matrix to obtain the original random measurement matrix Φ'. Then, perform normalization processing on the original random measurement matrix Φ' to obtain the random measurement matrix Φ that is uncorrelated with the substation classification sparse basis. This random measurement matrix Φ has good numerical stability and can effectively project high-dimensional signals into a low-dimensional space in compressive sensing; h n+1 = μh n (1 - h n )(1) where h n+1 and h n are the chaotic variable values of the (n + 1)-th and n-th iterations respectively, and μ is the bifurcation parameter; S33. Based on the classification sparse basis Ψ of a certain substation m , and according to the compressive sensing theory shown in the following formula (2) with the random measurement matrix Φ, compress the broadband oscillatory electrical signal to be compressed, that is, perform secondary subsampling, to obtain a compressed signal y, that is, a one-dimensional observation signal; y = Φx = ΦΨ m s = Θs (2) Wherein, x is the original signal with a length of N, that is, the broadband oscillating electrical signal to be compressed, M represents the number of rows of the observation matrix, that is, the dimension of the compressed signal y, and N >> M; S is the coefficient vector of the original signal x under a certain sub-station classification sparse basis Ψ m and is sparse, with a sparsity of G, and G << N; Θ = ΦΨ, which is an M×N dimensional sensing matrix.

5. A wide-area monitoring method for broadband oscillation in a novel power system according to claim 4, characterized in that, In step S4, at the master station, use deep dictionary learning to construct a broadband oscillation classification and recognition model, and use this broadband oscillation classification and recognition model to extract oscillation feature quantities from the uploaded compressed signals to identify the broadband oscillation types. First, the master station performs normalization processing on the one-dimensional observation signals of different broadband oscillation types received from each substation after compression, and sets classification labels according to the preliminary prediction results of the broadband oscillation types to form a one-dimensional observation signal sample data set. Then, use the Gram angular field method to convert the one-dimensional observation signals in the one-dimensional observation signal sample data set into two-dimensional Gram feature maps, and set corresponding classification labels for each two-dimensional Gram feature map. Construct a two-dimensional image sample data set from the two-dimensional Gram feature maps and the corresponding classification labels, and divide it into a training set and a test set; Based on the deep learning theory and the dictionary learning theory, a deep dictionary learning method is proposed for constructing a broadband oscillation classification and recognition model. The model structure includes: Input module: The input is two-dimensional Gram feature maps of various categories, providing basic data for subsequent deep dictionary learning; Classification deep learning dictionary module: It contains deep learning dictionary modules of multiple categories. The input is two-dimensional Gram feature maps of various categories in the input module, and the output is the updated deep learning dictionaries of various categories. Each category's deep learning dictionary module consists of multiple cascaded deep dictionary learning modules. Each deep dictionary learning module includes a sparse coding layer and a dictionary update layer. Through multi-layer dictionaries, gradually learn to extract local and global features of two-dimensional Gram feature maps to improve the feature extraction ability. The input of each deep dictionary learning module is a two-dimensional Gram feature map, and the output is the updated deep learning dictionaries of various categories. Among them, the input of the sparse coding layer is the data output by the previous layer, that is, the output of the input module or the previous deep dictionary learning module, and the dictionary matrix of this layer. By optimizing the solution, the sparse representation coefficients are output, thereby extracting the key features of the input signal. The input of the dictionary update layer is the sparse representation coefficients output by the sparse coding layer. Use the dictionary learning algorithm based on nonlinear transformation to update the dictionary matrix of this layer to obtain the updated dictionary of this layer, thereby improving the nonlinear expression ability and suppressing the noise and non-significant features in the signal; Sparse representation coefficient calculation module: The input is the original data of each two-dimensional Gram feature map and the updated deep learning dictionaries of various categories. Using the sparse representation method, calculate the sparse representation coefficients of each two-dimensional Gram feature map under the deep learning dictionaries of various categories, and use the signal reconstruction algorithm to obtain the reconstructed two-dimensional Gram feature maps under the deep learning dictionaries of various categories, and compare them with the original data. Finally, output the reconstruction residuals of each two-dimensional Gram feature map under the deep learning dictionaries of various categories. The smaller the error, the higher the matching degree with this category; Classification decision module: The input is the reconstruction residuals of each two-dimensional Gram feature map under the deep learning dictionaries of various categories. Select the category with the smallest reconstruction residual as the sample category, and the output is the category label of each two-dimensional Gram feature map; Finally, use the test set divided from the two-dimensional image sample data set to test the broadband oscillation classification and recognition model, further optimize and adjust the model parameters, and obtain the broadband oscillation classification and recognition model with the best performance.

6. A wide - area monitoring method for broadband oscillation of a new power system according to claim 5, characterized in that, The step S5 includes the following steps: S51. Based on the one-dimensional observation signal sample data set, the master station designs the master station classification sparse basis, that is, the master station classification sparse transform domain, using the sparse degree adaptive K-SVD dictionary learning method; S52. The master station uses the random measurement matrix generation model reconstructed by the Logistic chaotic mapping method shown in formula (1) to generate a random measurement matrix, and multiplies it by the master station classification sparse basis to obtain the sensing matrix; S53. Based on the compressive sensing theory and the sensing matrix, construct a broadband oscillation signal reconstruction model, and the model structure is as follows: Input layer: The input is the one-dimensional observation signal; Signal reconstruction layer: Use the signal reconstruction algorithm based on sparse representation to reconstruct the one-dimensional observation signal, and obtain the reconstructed sparse signal through the deep learning dictionaries of various categories and the sparse representation coefficients; Signal recovery layer: Recover the reconstructed sparse signal and output the recovered broadband oscillation signal, that is, the reconstructed broadband oscillation signal; Finally, use the training set and the test set to train and test the broadband oscillation signal reconstruction model, optimize the model parameters, and obtain the broadband oscillation signal reconstruction model with the best performance; S54. Use the broadband oscillation signal reconstruction model with the best performance, with the input being the compressed signals uploaded by each slave station and the output being the reconstructed broadband oscillation signals; S55. The master station conducts comprehensive analysis based on the reconstructed broadband oscillation signals, including time-frequency analysis, wide-area monitoring, mechanism analysis, disturbance source tracing and positioning, data mining, dispatching decision-making, and preventive protection and control.

7. A wide - area monitoring method for broadband oscillation of a new - type power system according to claim 6, characterized in that, The signal reconstruction algorithms based on sparse representation include the matching pursuit algorithm, the orthogonal matching pursuit algorithm, the regularized matching pursuit algorithm, the basis pursuit algorithm, and the alternating direction method of multipliers.

Images