A method and apparatus for modal continuity identification of multimodal oscillation data

Abstract

This invention discloses a method and apparatus for modal continuity identification of multimodal oscillation data. The invention includes: extracting features from multiple frames of broadband oscillation data to obtain modal features for each frame of broadband oscillation data; clustering based on the modal features to obtain multiple clusters for each frame of broadband oscillation data, and determining the cluster center variation characteristics for each frame of broadband oscillation data; obtaining the modal matching threshold for each frame's clusters based on the cluster center variation characteristics; and performing modal continuity matching using the clusters and corresponding modal matching thresholds to obtain the modal continuity identification result. This invention achieves real-time and accurate modal continuity identification of broadband oscillation data with frequency density, strong time-varying characteristics, and multimodal features by performing intra-frame modal clustering on a time scale for each modal feature and combining it with a threshold prediction model with preset hyperparameters for adaptive threshold prediction.

Description

Technical Field

[0001] This invention relates to the field of power system technology, and in particular to a method and apparatus for identifying the mode continuity of multimodal oscillation data. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] With the large-scale grid connection of new energy power generation such as wind power and photovoltaic power, the power grid operating environment has become increasingly complex. Due to the intermittency and uncertainty of new energy power generation, oscillation phenomena in the power system have become more common, exhibiting characteristics of dense frequency, strong time variation, and coexistence of multiple modes. For example, the grid connection of photovoltaic inverters and wind turbines introduces multiple oscillation modes such as subsynchronous oscillation and supersynchronous oscillation, making the grid signal exhibit more complex dynamic characteristics.

[0004] In actual measurement processes, broadband measurement devices can acquire oscillation signals in power systems at different time frames. However, due to noise interference, measurement errors, signal aliasing, and other factors, the modal characteristics of different time frames, such as frequency, amplitude, and phase, will show slight changes. This makes it difficult to directly determine whether the modes in different time frames are the same mode. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a modal continuity identification method, system method, and apparatus for modal oscillation data, with the aim of achieving real-time and accurate modal continuity identification for oscillation measurement data with frequency-dense, highly time-varying, and multimodal characteristics.

[0006] To achieve the above objective, according to a first aspect of the present invention, a method for identifying the modal continuity of multimodal oscillation data is provided, comprising:

[0007] Feature extraction is performed on multiple frames of wideband oscillation data to obtain the modal features of multiple modes in each frame of wideband oscillation data.

[0008] Based on the modal features, multiple modes are clustered to obtain multiple clusters of broadband oscillation data for each frame, and the cluster center variation characteristics of broadband oscillation data for each frame are determined.

[0009] Adaptive threshold prediction is performed based on the cluster center change characteristics to obtain the modality matching threshold of the cluster in each frame;

[0010] Modal continuity matching is performed using the clusters and the corresponding modal matching thresholds to obtain modal continuity recognition results.

[0011] Furthermore, the modal characteristics of the mode include at least the frequency, 0th-order amplitude, 1st-order amplitude, 0th-order phase, 1st-order phase, 0th-order damping ratio coefficient, and 1st-order damping ratio coefficient corresponding to the mode;

[0012] Clustering is performed on multiple modes based on the modal features to obtain multiple clusters of broadband oscillation data for each frame, and the cluster center variation features of broadband oscillation data for each frame are determined, including:

[0013] Obtain the modal features of the broadband oscillation data of the current frame and the previous frame, and construct the corresponding feature vector of the mode based on the modal features;

[0014] The feature vector is used to determine the respective clusters and modes contained in the broadband oscillation data of the current frame and the previous frame, and the cluster center is determined according to the feature vector of the modes contained in the cluster.

[0015] The cluster center change characteristics of the current frame are determined based on the cluster centers of the current frame and the previous frame, respectively.

[0016] Furthermore, based on the cluster center change characteristics, adaptive threshold prediction is performed to obtain the modality matching threshold of the clusters in each frame, including:

[0017] A prediction input vector is constructed based on the cluster center change characteristics of each frame within a preset time window.

[0018] The predicted input vector is input into a pre-trained threshold prediction model, and the threshold prediction model outputs the modality matching threshold of the cluster.

[0019] Furthermore, the first The wideband oscillation data of the frame Feature vectors of each modality for:

[0020] ,

[0021] in, , , , The first Frame wideband oscillation data The frequency of the first mode, the first Amplitude value, first Phase order and the first Damping ratio coefficient, , , For the first The number of modes in the frame-wideband oscillation data;

[0022] The clustering of broadband oscillation data in the current frame and the previous frame, and the modes they contain, are determined using the feature vectors. The cluster center is then determined based on the feature vectors of the modes contained in the clusters, including:

[0023] An improved k-means clustering algorithm is used to perform clustering on the first... The eigenvectors of each mode of the frame wideband oscillation data Clustering was performed to obtain the results. Each of the aforementioned clusters and its respective cluster center ,in, , For the first Each cluster contains a set of all the aforementioned modalities. .

[0024] Furthermore, the cluster center variation characteristics include the first... The cluster center weighted shift of the frame wideband oscillation data Weighted shift of cluster centers of each cluster ,

[0025] ;

[0026] in, , , , , The first Frame wideband oscillation data The frequency of the cluster center and the first Amplitude, phase, and damping ratio coefficient , , , The first Frame wideband oscillation data The frequency of the cluster center and the first Amplitude, phase, and damping ratio coefficient , , and Frequency and the number of Weighting coefficients for amplitude, phase, and damping ratio.

[0027] Furthermore, based on the cluster center change characteristics, adaptive threshold prediction is performed to obtain the modality matching threshold of the clusters in each frame, including:

[0028] Construct a prediction input vector based on the cluster center change characteristics of each frame within a preset time window. The predicted input vector ,in, Indicates the preset time window size. ;

[0029] The predicted input vector An adaptive threshold prediction is performed by inputting a pre-trained LSTM model, and the modality matching threshold of the cluster is output by the LSTM model.

[0030] Furthermore, the predicted first The wideband oscillation data of the frame Modality The modality matching threshold of the c-th cluster is ;

[0031] Modal continuity matching of adjacent frames is performed using the clusters and the corresponding modal matching thresholds to obtain modal continuity recognition results, including:

[0032] Calculate the first The wideband oscillation data of the frame Modality and the The wideband oscillation data of the frame Modality Weighted distance ;

[0033] Compare the weighted distances The modality matching threshold is Size;

[0034] In response to the weighted distance Less than the modal matching threshold Then the mode With the mode continuous.

[0035] According to a second aspect of the present invention, a modal continuity identification device for multimodal oscillation data is provided, comprising:

[0036] The feature extraction module is used to extract features from multiple frames of broadband oscillation data to obtain the modal features of each mode of the broadband oscillation data in each frame.

[0037] The clustering processing module is used to cluster multiple modes according to the modal features to obtain multiple clusters of broadband oscillation data in each frame, and to determine the cluster center variation features of broadband oscillation data in each frame.

[0038] The threshold prediction module is used to perform adaptive threshold prediction based on the cluster center change characteristics to obtain the modality matching threshold of the cluster in each frame.

[0039] The continuity judgment module is used to perform modal continuity matching using the cluster and the corresponding modal matching threshold to obtain the modal continuity recognition result.

[0040] According to a third aspect of the present invention, a computer-readable storage medium is provided having a program stored thereon that, when executed by a processor, implements the steps in the modal continuity identification method for multimodal oscillation data as described in the first aspect of the present invention.

[0041] According to a fourth aspect of the present invention, a terminal device is provided, including a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the modal continuity identification method for multimodal oscillation data as described in the first aspect of the present invention.

[0042] According to a fifth aspect of the present invention, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the steps in the modal continuity identification method for multimodal oscillation data as described in the first aspect of the present invention.

[0043] The embodiments of the present invention have the following advantages or beneficial effects:

[0044] This invention addresses the limitations of current modality recognition and matching methods by performing intra-frame modality clustering on a time scale for each modality and combining it with a threshold prediction model based on preset hyperparameters for adaptive threshold prediction. Compared to methods that use a fixed frequency deviation threshold to determine whether two modalities are continuous, this invention achieves real-time and accurate modality continuity recognition for broadband oscillation data with frequency-dense, highly time-varying, and multimodal characteristics.

[0045] The embodiments of the present invention cluster each frame of broadband oscillation data and obtain the typical cluster change characteristics of the clusters for threshold prediction, which reduces the computational complexity and computational load of the threshold prediction model and can meet the real-time and accuracy requirements of modal continuity identification of broadband oscillation data with dense frequency, strong time variation and multimodal characteristics.

[0046] The embodiments of the present invention use cluster center change features to construct a prediction input vector for adaptive threshold prediction, which solves the problem of continuity misjudgment caused by using a single fixed threshold that is too strict or too lenient due to the dynamic change characteristics of power grid oscillation.

[0047] The improved k-means clustering algorithm used in this embodiment of the invention can adapt to the frequency-dense, time-varying, and multimodal characteristics of broadband oscillation signals in power grids. It has good recognition accuracy in high-dimensional space, and can process newly emerging oscillation modes with less training data. It has low computational complexity and can be applied to low-latency real-time processing scenarios in actual power grid monitoring.

[0048] This invention addresses the common problems faced by traditional modal identification methods when dealing with frequency-dense, highly time-varying, and multimodal oscillatory data, such as insufficient time correlation modeling capabilities, difficulty in identifying fuzzy modal transitions, and weak robustness to disturbances. Based on the strong time series modeling and nonlinear mapping capabilities of the pre-trained LSTM model, the method automatically learns modal evolution characteristics through the LSTM model, effectively improving the accuracy and stability of modal continuity identification. Therefore, the method and apparatus of this invention can be applied to modal identification in complex dynamic power systems.

[0049] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description

[0050] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0051] Figure 1 This is a schematic diagram of the main flow of the modal continuity identification method for multimodal oscillation data according to an embodiment of the present invention;

[0052] Figure 2 This is a schematic diagram of the main modules of the modal continuity identification device for multimodal oscillation data according to an embodiment of the present invention;

[0053] Figure 3 This is a schematic diagram of the composition of a terminal device according to an embodiment of the present invention. Detailed Implementation

[0054] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0055] Example 1

[0056] Figure 1 This is a schematic diagram of the main flow of a method for identifying the modal continuity of multimodal oscillation data according to an embodiment of the present invention, as shown below. Figure 1 As shown, the modal continuity identification method for multimodal oscillation data in this embodiment of the present invention includes the following steps S101 to S104.

[0057] Step S101: Feature extraction is performed on the multi-frame broadband oscillation data to obtain the modal features of each mode of the broadband oscillation data in each frame.

[0058] Step S102: Cluster the multiple modes according to the modal features to obtain multiple clusters of broadband oscillation data for each frame, and determine the cluster center variation features of broadband oscillation data for each frame.

[0059] Step S103: Based on the cluster center change characteristics, perform adaptive threshold prediction to obtain the modality matching threshold of the cluster in each frame.

[0060] Step S104: Modality continuity matching is performed using the cluster and the corresponding modality matching threshold to obtain the modality continuity recognition result.

[0061] It is understood that the broadband oscillation data in this embodiment of the invention is data obtained by collecting broadband oscillation signals of the power system according to a preset time scale. Broadband oscillation data has characteristics of high frequency density, strong time variation, and multimodal operation. The method of this embodiment of the invention performs continuity identification of multimodal oscillation data, which refers to whether the modes in two adjacent frames of broadband oscillation data belong to the same mode. Each frame of broadband oscillation data is obtained by truncating and sampling the broadband oscillation signal according to a preset time length. The broadband oscillation signal includes typical electrical quantities such as voltage phasors and current phasors. A mode is a dynamic response pattern of a certain voltage or current. The modal characteristics of a mode include frequency, amplitude and its rate of change, phase and its rate of change, and damping ratio coefficient and its rate of change. In some embodiments of the present invention, the extracted modal features include at least the frequency, amplitude (0th order amplitude) and first rate of change of amplitude (1st order amplitude), phase (0th order phase) and first rate of change of phase (1st order phase), damping ratio coefficient (0th order damping ratio coefficient) and first rate of change of damping ratio coefficient (1st order damping ratio coefficient) corresponding to the mode.

[0062] Specifically, in step S101 of this embodiment, the first... The modal characteristics of the frame wideband oscillation data include the first Frame number The frequency of each mode , No. Frame number Amplitude of each mode (0th order amplitude), first rate of change of amplitude (First-order amplitude), second-order rate of change of amplitude (Second-order amplitude), third rate of change of amplitude (Third-order amplitude), the Frame number Phase of each mode (0th order phase), first phase change rate (First-order phase), second-order rate of change of phase (Second-order phase), third-order phase change rate (3rd phase), the Frame number Damping ratio coefficient of each mode (0th order damping ratio coefficient), first-order rate of change of damping ratio coefficient (First-order damping ratio coefficient), second-order rate of change of damping ratio coefficient (Second-order damping ratio coefficient), third rate of change of damping ratio coefficient (0th order damping ratio coefficient) , For the first The number of modalities detected in the frame.

[0063] Understandably, clustering in step S102 is an unsupervised learning method used to divide a dataset into multiple subsets (clusters), such that data objects within the same cluster have high similarity, while data objects between different clusters have low similarity. Specifically, in this embodiment and some embodiments of the present invention, step S102 includes steps S102a and S102c.

[0064] Step S102a: Obtain the modal features of the broadband oscillation data of the current frame and the previous frame, and construct the feature vector of the corresponding mode based on the modal features.

[0065] Specifically, in the embodiments of the present invention, the first The wideband oscillation data of the frame Feature vectors of each modality The frequency amplitude and phase of a mode are fundamental characteristics of a mode. In order to distinguish between modes with dense frequency, strong time variation and quantity in broadband oscillation data, this embodiment of the invention also selects the damping ratio coefficient and the first and higher order rates of change of the amplitude / phase / damping ratio coefficient as characteristic components. The first rate of change of the amplitude / phase / damping ratio coefficient is used to represent the instantaneous change trend, such as the sudden drop in amplitude caused by wind turbine disconnection from the grid. The higher order rates of change of the amplitude / phase / damping ratio coefficient are used to identify nonlinear dynamic characteristics, such as the nonlinear amplitude fluctuation caused by the limiting of photovoltaic inverters.

[0066] In this embodiment, the first The wideband oscillation data of the frame Feature vectors of each modality According to the Frames The mode can obtain the th . Multimodal feature matrix of a frame .

[0067] ;

[0068] Among them, feature vector In the multimodal feature matrix The Data points, .

[0069] Step S102b: Use the feature vector to determine the respective clusters and modes contained in the broadband oscillation data of the current frame and the previous frame, and determine the cluster center based on the feature vector of the modes contained in the cluster.

[0070] In selecting clustering methods, this invention considers the following: DBSCAN (density-based spatial clustering) can be used for modality classification, but its reliance on Euclidean distance may lead to decreased recognition accuracy in high-dimensional feature spaces, and its parameters (such as neighborhood radius) are difficult to adapt to time-varying characteristics. Some studies use Support Vector Machines (SVM), deep learning (such as LSTM, CNN), etc., for modality classification, but these methods usually require a large amount of training data, are difficult to handle newly emerging oscillating modes, and have high computational complexity, making it difficult to achieve low-latency real-time processing in actual power grid monitoring. Therefore, specifically, in this embodiment and some embodiments of this invention, step S102b includes: using an improved k-means clustering algorithm to classify the first... The eigenvectors of each mode of the frame wideband oscillation data Clustering is performed to obtain Each of the aforementioned clusters and its respective cluster center ,in, , For the first Each cluster contains a set of all the aforementioned modalities. .

[0071] Understandably, the improved k-means clustering algorithm finds the nearest cluster center for each modality data point and then assigns the data point to the corresponding cluster. Further, in this embodiment, step S102b includes steps S102b1 to S102b4:

[0072] Step S102b1: Determine the set of cluster centers, specifically including:

[0073] Step S102b11: Randomly select the first cluster center as the selected cluster center.

[0074] In this embodiment and some embodiments of the present invention, from the multimodal feature matrix Randomly select a data point As the first cluster center, to distinguish between the selected cluster centers and the non-selected cluster centers, let be... ,in, For the first The first cluster center of the frame, For the first Frame number The feature vectors of each modality.

[0075] Step S102b12: Calculate the weighted distance from the data points to the selected cluster centers.

[0076] In this embodiment and some embodiments of the present invention, the minimum distance from all data points to the selected cluster centers is calculated: for time frames... Each data point that is not a selected cluster center Calculate data points Minimum weighted distance to the nearest selected cluster center:

[0077] ,

[0078] ,

[0079] ;

[0080] in, The representative of the selected number Cluster centers , recorded as , , Representative data points The minimum weighted distance to the nearest selected cluster center. , , , and Frequency and the number of The weighting coefficients for amplitude, phase, and damping ratio represent the influence of different features on intra-frame modal distance. The weights are related to the application scenario of modal continuity recognition; typically, the weights in the pre-training stage and the prediction stage can use the same values. In some embodiments of this invention, , , and .

[0081] Step S102b13: Calculate the probability that each data point will become the next cluster center based on the minimum weighted distance calculation result, and obtain the probability distribution.

[0082] In this embodiment and some embodiments of the present invention, the minimum distance from each data point to the selected cluster center is used. Calculate and select this point The probability of being the next cluster center:

[0083] , , ;

[0084] in, For data points The probability of being selected as the next cluster center. For the first Frame number The feature vectors of each modality make it more likely that data points that are far from the existing center point will be selected as the new center point, thereby ensuring a uniform distribution among different center points.

[0085] Step S102b14: Select new cluster centers as new selected cluster centers based on the probability distribution.

[0086] In this embodiment and some embodiments of the present invention, the probability distribution calculated in the previous step (step S102b13) is used to select a new cluster center. .

[0087] ;

[0088] in, For the newly selected cluster centers, To use the cumulative distribution function according to probability Data points selected through sampling.

[0089] Step S102b15: Return to step S102b12 to calculate the new selected cluster centers until a set of selected cluster centers is obtained. Specifically, repeat steps S102b12 to S102b14 until a cluster center is selected. Cluster centers , The number of cluster centers is preset. Generally, the number of cluster centers is less than the number of modes. This is used to reduce the dimensionality of adaptive threshold prediction and improve the speed of continuity judgment.

[0090] Step S102b2: Assign data points to the nearest cluster center.

[0091] In this embodiment and some embodiments of the present invention, data points are allocated. The formula for finding the nearest cluster center is: ;

[0092] in, for The label of the cluster center This indicates finding the minimum weighted distance. ,Right now closest cluster center , Indicates the first Data points and selected cluster centers The distance.

[0093] Step S102b3: After each data point allocation, update the new centroid of each cluster. The calculation method is as follows:

[0094] ;

[0095] in, The updated feature mean, For the first The set of all data points contained in the cluster to which each cluster center belongs. This represents the number of modes in the set.

[0096] .

[0097] Step S102b4: Determine whether the cluster assignment labels of all data points will change. If the cluster assignment labels of all data points no longer change, stop the iteration and output the final cluster centers. If it is the cluster center, otherwise continue executing steps S102b2 and S102b3 until the condition is met:

[0098] ;

[0099] in, Represents the latest cluster label number for all modalities within the frame. It represents the previous cluster label number for all modalities within the frame.

[0100] Step S102c: Determine the cluster center change characteristics of the current frame based on the cluster centers of the current frame and the previous frame.

[0101] Furthermore, the cluster center variation characteristics include the first... The first frame of wideband oscillation data Weighted shift of cluster centers , ,

[0102] ;

[0103] in, , , , , The first Frame wideband oscillation data The frequency of the cluster center and the first Amplitude, phase, and damping ratio coefficient , , , The first Frame wideband oscillation data The frequency of the cluster center and the first Amplitude, phase, and damping ratio coefficient , , and Frequency and the number of Weighting coefficients for amplitude, phase, and damping ratio.

[0104] Specifically, in this embodiment and in the embodiments of the present invention, the first... Frame clustering and the first The number of clusters in each frame is equal.

[0105] Specifically, in this embodiment and some embodiments of the present invention, step S103 includes steps S103a and S103b.

[0106] Step S103a: Construct a prediction input vector based on the cluster center change features of each frame within a preset time window. Specifically, construct the prediction input vector based on the cluster center change features of each frame within a preset time window. The predicted input vector ,in, Indicates the preset time window size. .

[0107] Step S103b involves inputting the predicted input vector into a pre-trained threshold prediction model, where the threshold prediction model outputs the modality matching threshold for the cluster. Specifically, the predicted input vector... The pre-trained LSTM model is used for adaptive threshold prediction. The modality matching threshold of the cluster is output by the LSTM model, i.e., the threshold of the first cluster. The first frame of wideband oscillation data Adaptive prediction threshold for each cluster .

[0108] In this embodiment, the threshold optimization model of the LSTM model is adopted because the LSTM model has strong time series modeling ability and nonlinear mapping ability, and can automatically learn modal evolution characteristics, effectively improving the accuracy and stability of modal continuity recognition. It is particularly suitable for the intelligent modal recognition scenario in the complex dynamic power system in this embodiment of the invention.

[0109] Specifically, in this embodiment and some embodiments of the present invention, the pre-training of the LSTM model uses the LSTM model as the threshold optimization model, and performs hyperparameter optimization and training on the LSTM model, including steps S103b10 to S103b14.

[0110] Step S103b10: Define the hyperparameter search space of the LSTM model and initialize the hyperparameter samples.

[0111] In this embodiment and some embodiments of the present invention, the prediction performance of the LSTM model depends on multiple hyperparameters, requiring the parameter decomposition of the LSTM model in the parameter space. Optimization was carried out, including, The learning rate is used to influence the speed of gradient descent. This refers to the size of the LSTM hidden layers, used to control the model capacity. This is the Dropout rate, used to prevent overfitting. Batch size, used to influence training stability. The number of training epochs affects the convergence rate.

[0112] Step S103b12: Use hyperparameter samples to train the LSTM model and calculate the loss.

[0113] In this embodiment and some embodiments of the present invention, random sampling is performed in the search space. Group hyperparameters , represented as:

[0114] ;

[0115] in, For the first The learning rate of the hyperparameters. For the first Hidden layer size of the LSTM model with set of hyperparameters For the first Dropout rate of the hyperparameter group. For the first Batch size of group hyperparameters, For the first The number of training epochs for the hyperparameters.

[0116] In the stage of determining the hyperparameters of the LSTM model, for each hyperparameter... Employing modal continuity recognition The input data for each broadband oscillation data sample is a feature vector constructed using one broadband oscillation data sample. Specifically, it is the first... The first broadband vibration data sample Clusters within a time window The eigenvectors formed by the cluster center weighted shift sequences of each time frame within the time frame , represented as:

[0117] ;

[0118] in, For the first The first broadband vibration data sample The cluster center is at the Cluster center weighted shift of the frame, For the first The first broadband vibration data sample The cluster in the th order of ... Frame and the The weighted shift of the cluster center of the frame. The time window size represents the number of historical frames used for prediction. .

[0119] Furthermore,

[0120] ;

[0121] ;

[0122] in, , , , The first The first broadband vibration data sample Frame number The frequency of the cluster center and the first Amplitude, phase, and damping ratio coefficient , , , The first The first broadband vibration data sample Frame number The frequency of the first cluster center and the second Amplitude, phase, and damping ratio coefficient , , and Frequency and the number of Weighting coefficients for amplitude, phase, and damping ratio.

[0123] In this embodiment and some embodiments of the present invention, the optimal adaptive threshold is predicted based on an LSTM model. Specifically, the LSTM model uses gating units for prediction, including input gates, forget gates, output gates, and memory units with the same shape as the hidden state.

[0124] (1) Input gate for:

[0125] ;

[0126] in, The activation value of the input gate determines the degree of information update. Here is the weight matrix of the input gate. For the first The hidden state of a frame, The predicted input vector is used during training. For the bias term of the input gate, It is a sigmoid activation function that controls the amount of information flowing in.

[0127] (2) Forgotten Gate for:

[0128] ;

[0129] in, The activation value of the forget gate determines the information that needs to be forgotten. Here is the weight matrix for the forget gate. The bias term representing the forget gate.

[0130] (3) Memory unit update for:

[0131] ;

[0132] in, To store long-term information for the current state of the memory cell. For the first The memory cell state of a frame. This indicates element-wise multiplication. To control the updating of the memory cells, the weight matrix is ​​used. The bias term that controls the updating of memory cells. This is the hyperbolic tangent function, used to control information updates.

[0133] (4) Output gate for:

[0134] ;

[0135] in, The activation value of the output gate determines the proportion of the current information output. This is the weight matrix of the output gate. This is the bias term for the output gate.

[0136] (5) The LSTM model calculates the predicted value of the adaptive threshold according to the following formula:

[0137] ;

[0138] in, For use of the Group hyperparameters The predicted first The first sample The first cluster of the cluster Frame modality matching threshold, The output layer weight matrix of the LSTM model. For the bias term of the output layer of the LSTM model, The current hidden state. .

[0139] Specifically, in this embodiment and some embodiments of the present invention, the first... Group hyperparameters Train the LSTM model and calculate the modality matching threshold loss function according to the following formula. ,

[0140] ;

[0141] in, For the first Group hyperparameters Modality matching threshold loss function, The number of broadband oscillation data samples for modal continuity identification. For use of the Group hyperparameters The predicted first The first broadband oscillation data sample The first cluster Frame modality matching threshold, To pass the first The first broadband oscillation data sample obtained The first cluster The true threshold of the frame.

[0142] Understandably, the true threshold for clustering is based on physical modeling and simulation methods to generate synthetic data with known modal evolution processes under controllable conditions. In this process, the system simulates the dynamic response behavior of oscillators with known frequency, amplitude, and phase characteristics across continuous time frames by injecting these oscillation sources. Because the modal parameters set during data generation have clear uniqueness and temporal continuity, the correspondence between the same mode in different time frames can be automatically labeled, constructing true continuous modal pairs.

[0143] Specifically, the first step is to adopt the... In the known broadband oscillation data sample, the first... For feature vectors of the same modality in different frames of a cluster, calculate the weighted distance between feature vectors of adjacent time frames of the same modality:

[0144] ;

[0145] in, For the first In the first broadband oscillation data sample The first cluster The modality of the first Frame and the The true weighted distance of the frames; and The first In the first broadband oscillation data sample The first cluster The mode in the th ... Frame and the Frame frequency parameters; and The first In the first broadband oscillation data sample The first cluster The mode in the th ... Frame and the The b-th order amplitude parameter of the frame; and The first In the first broadband oscillation data sample The first cluster The mode in the th ... Frame and the The b-th phase parameter of the frame; and The first In the broadband oscillation data sample, the first The first cluster The mode in the th ... Frame and the The b-th order damping ratio parameter of the frame; , , and Frequency and the number of Weighting coefficients for amplitude, phase, and damping ratio.

[0146] The previous step counted a total of the first... In the broadband oscillation data sample, the first In each cluster The modality is then used, and the weighted distances between these true matching pairs are averaged as the first modality. In the first broadband oscillation data sample The true matching threshold for each cluster:

[0147] .

[0148] Step S103b13: Determine the better and worse hyperparameter distributions and select new hyperparameters.

[0149] In this embodiment and some embodiments of the present invention, hyperparameters are obtained by using wideband oscillation data samples (historical experimental data). Divided into two categories:

[0150] The first type is a relatively optimal hyperparameter distribution, which has better performance, i.e., the... Modality matching threshold loss function for clusters Less than :

[0151] .

[0152] The second type is the poor hyperparameter distribution, which has poor performance, i.e., the... Modality matching threshold loss function for clusters Greater than :

[0153] .

[0154] New hyperparameters are selected by minimizing the ratio of the poor distribution to the good distribution:

[0155] ;

[0156] in, This represents the hyperparameter search space, i.e., the range of all possible hyperparameter values. For dynamically adjusted Ranked among the top in historical experiments for modal clustering centers. %The maximum loss function value in the results The modality matching threshold loss function value after LSTM training. These are the optimal hyperparameters.

[0157] In this embodiment, γ is the threshold for taking the top 10% of results in historical experiments. The set of all hyperparameters in the search space that have a loss function value less than this threshold after LSTM training is a better hyperparameter distribution, and vice versa.

[0158] Step S103b14: Optimal hyperparameters Substitute into the LSTM model for training.

[0159] In this embodiment and some embodiments of the present invention, a new loss function is described. :

[0160] ;

[0161] like Less than the set threshold End the optimization and let the final value of the hyperparameters be... And obtain the corresponding optimal hyperparameters. The weight matrices and bias terms of the LSTM model are used to obtain the corresponding optimal hyperparameters. If the LSTM model is satisfied, then proceed to step S103b12 to continue sampling new hyperparameters for optimization until the iterative convergence condition is met. .

[0162] Specifically, in this embodiment and some embodiments of the present invention, in step S103b, the optimal hyperparameters obtained by the pre-trained LSTM model are used. The predicted input vector The input corresponds to the optimal hyperparameters The LSTM model is used to obtain the predicted mode matching threshold. Furthermore, in this embodiment and some embodiments of the present invention, the optimal hyperparameters are employed. The LSTM model predicts the adaptive modality matching threshold using gated units, including:

[0163] (1) The 0-input gate is:

[0164] ;

[0165] in, The activation value of the input gate determines the degree of information update. Here is the weight matrix of the input gate. For the first The hidden state of a frame, For the first Frame number The predicted input vector for each cluster, For the bias term of the input gate, It is a sigmoid activation function that controls the amount of information flowing in.

[0166] (2) The forgetting gate is:

[0167] ;

[0168] in, The activation value of the forget gate determines the information that needs to be forgotten. Here is the weight matrix for the forget gate. The bias term representing the forget gate.

[0169] (3) The memory unit update formula is:

[0170] ;

[0171] in, To store long-term information for the current state of the memory cell. For the first The memory cell state of a frame. To control the updating of the memory cells, the weight matrix is ​​used. The bias term that controls the updating of memory cells. This is the hyperbolic tangent function, used to control information updates.

[0172] (4) The output gate is:

[0173] ;

[0174] in, The activation value of the output gate determines the proportion of the current information output. This is the weight matrix of the output gate. This is the bias term for the output gate.

[0175] (5) Calculate the adaptive modal matching threshold:

[0176] ;

[0177] in, For the predicted frame t+1, the... An adaptive modality matching threshold for each modality cluster center. This is the weight matrix of the output layer. For the bias term of the output layer, The current hidden state. .

[0178] Specifically, in this embodiment and some embodiments of the present invention, step S104 includes steps S104a to S104c:

[0179] Step S104a, calculate the first The wideband oscillation data of the frame Modality and the The wideband oscillation data of the frame Modality Weighted distance ;

[0180] , ;

[0181] Step S104b: Compare the weighted distances The modality matching threshold is Size;

[0182] Step S104c, in response to the weighted distance Less than the modal matching threshold Then the mode With the mode Continuous; otherwise, the mode This could be noise or a new mode.

[0183] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0184] Example 2

[0185] According to another aspect of the embodiments of the present invention, such as Figure 2 As shown, a modal continuity identification device for multimodal oscillation data is provided, comprising:

[0186] The feature extraction module is used to extract features from multiple frames of broadband oscillation data to obtain the modal features of each mode of the broadband oscillation data in each frame.

[0187] The clustering processing module is used to cluster multiple modes according to the modal features to obtain multiple clusters of broadband oscillation data in each frame, and to determine the cluster center variation features of broadband oscillation data in each frame.

[0188] The threshold prediction module is used to perform adaptive threshold prediction based on the cluster center change characteristics to obtain the modality matching threshold of the cluster in each frame.

[0189] The continuity judgment module is used to perform modal continuity matching using the cluster and the corresponding modal matching threshold to obtain the modal continuity recognition result.

[0190] Example 3

[0191] like Figure 3 As shown, Embodiment 3 of the present invention provides a terminal device, including at least one processor and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps in the modal continuity identification method for multimodal oscillation data as described in the first aspect of the present invention.

[0192] The memory and processor are connected via a bus, which can include any number of interconnecting buses and bridges, connecting various circuits of one or more processors and memories. The bus can also connect various other circuits, such as peripherals, voltage regulators, and power management circuits, via interfaces, as is well known in the art. Interfaces provide a connection between the bus and the transceiver, such as communication interfaces or user interfaces. The transceiver can be a single component or multiple components, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by the processor is transmitted over the wireless medium via an antenna, which further receives data and transmits it to the processor.

[0193] The processor manages the bus and general processing, and also provides various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory is used to store data used by the processor during operation.

[0194] Example 4

[0195] Embodiment 4 of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps in the modal continuity identification method for multimodal oscillation data as described in the first aspect of the present invention.

[0196] Those skilled in the art will understand from the foregoing description that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes, but is not limited to, various media capable of storing program code, such as USB flash drives, portable hard drives, magnetic storage devices, and optical storage devices.

[0197] Example 5

[0198] Embodiment 5 of the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the steps in the modal continuity identification method for multimodal oscillation data as described in the first aspect of the present invention.

[0199] This invention addresses certain limitations of current modality recognition and matching methods by performing intra-frame modality clustering on a time scale for each modality and combining it with a threshold prediction model based on preset hyperparameters for adaptive threshold prediction. Compared to methods that use a fixed frequency deviation threshold to determine whether two modalities are continuous, this invention achieves real-time and accurate modality continuity recognition for broadband oscillation data with frequency-dense, highly time-varying, and multimodal characteristics.

[0200] The embodiments of the present invention cluster each frame of broadband oscillation data and obtain the typical cluster change characteristics of the clusters for threshold prediction, which reduces the computational complexity and computational load of the threshold prediction model and can meet the real-time and accuracy requirements of modal continuity identification of broadband oscillation data with dense frequency, strong time variation and multimodal characteristics.

[0201] The embodiments of the present invention use cluster center change features to construct a prediction input vector for adaptive threshold prediction, which solves the problem of continuity misjudgment caused by using a single fixed threshold that is too strict or too lenient due to the dynamic change characteristics of power grid oscillation.

[0202] The improved k-means clustering algorithm used in this embodiment of the invention can adapt to the frequency-dense, time-varying, and multimodal characteristics of broadband oscillation signals in power grids. It has good recognition accuracy in high-dimensional space, and can process newly emerging oscillation modes with less training data. It has low computational complexity and can be applied to low-latency real-time processing scenarios in actual power grid monitoring.

[0203] This invention addresses the common problems faced by traditional modal identification methods when dealing with frequency-dense, highly time-varying, and multimodal oscillatory data, such as insufficient time correlation modeling capabilities, difficulty in identifying fuzzy modal transitions, and weak robustness to disturbances. Based on the strong time series modeling and nonlinear mapping capabilities of the pre-trained LSTM model, the method automatically learns modal evolution characteristics through the LSTM model, effectively improving the accuracy and stability of modal continuity identification. Therefore, the method and apparatus of this invention can be applied to modal identification in complex dynamic power systems.

[0204] Therefore, the embodiments of the present invention realize intelligent matching and continuity identification of oscillation modes in different time frames, which is of great significance for power grid safety monitoring, stability control and oscillation source tracing analysis.

[0205] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for identifying the modal continuity of multimodal oscillation data, characterized in that, include: Feature extraction is performed on multiple frames of wideband oscillation data to obtain the modal features of multiple modes in each frame of wideband oscillation data. Based on the modal features, multiple modes are clustered to obtain multiple clusters of broadband oscillation data for each frame, and the cluster center variation characteristics of broadband oscillation data for each frame are determined. Adaptive threshold prediction is performed based on the cluster center change characteristics to obtain the modality matching threshold of the cluster in each frame; Modal continuity matching is performed using the clusters and the corresponding modal matching thresholds to obtain modal continuity recognition results.

2. The method according to claim 1, characterized in that, The modal characteristics of the mode include at least the frequency, 0th-order amplitude, 1st-order amplitude, 0th-order phase, 1st-order phase, 0th-order damping ratio coefficient, and 1st-order damping ratio coefficient corresponding to the mode. Clustering is performed on multiple modes based on the modal features to obtain multiple clusters of broadband oscillation data for each frame, and the cluster center variation features of broadband oscillation data for each frame are determined, including: Obtain the modal features of the broadband oscillation data of the current frame and the previous frame, and construct the corresponding feature vector of the mode based on the modal features; The feature vector is used to determine the respective clusters and modes contained in the broadband oscillation data of the current frame and the previous frame, and the cluster center is determined according to the feature vector of the modes contained in the cluster. The cluster center change characteristics of the current frame are determined based on the cluster centers of the current frame and the previous frame, respectively. And / or, Adaptive threshold prediction is performed based on the cluster center change characteristics to obtain the modality matching threshold of the clusters in each frame, including: A prediction input vector is constructed based on the cluster center change characteristics of each frame within a preset time window. The predicted input vector is input into a pre-trained threshold prediction model, and the threshold prediction model outputs the modality matching threshold of the cluster.

3. The method according to claim 2, characterized in that, No. The wideband oscillation data of the frame Feature vectors of each modality for ， in, , , , The first Frame wideband oscillation data The frequency of the first mode, the first Amplitude value, first Phase order and the first Damping ratio coefficient, , , For the first The number of modes in the frame-wideband oscillation data; The clustering of broadband oscillation data in the current frame and the previous frame, and the modes they contain, are determined using the feature vectors. The cluster center is then determined based on the feature vectors of the modes contained in the clusters, including: An improved k-means clustering algorithm is used to perform clustering on the first... The eigenvectors of each mode of the frame wideband oscillation data Clustering is performed to obtain Each of the aforementioned clusters and its respective cluster center ,in, , For the first Each cluster contains a set of all the aforementioned modalities. .

4. The method according to claim 3, characterized in that, The cluster center variation characteristics include the first The first frame of wideband oscillation data Weighted shift of cluster centers of each cluster , , ； in, , , , , The first Frame wideband oscillation data The frequency of the cluster center and the first Amplitude, phase, and damping ratio coefficient , , , The first Frame wideband oscillation data The frequency of the cluster center and the first Amplitude, phase, and damping ratio coefficient , , and Frequency and the number of Weighting coefficients for amplitude, phase, and damping ratio.

5. The method according to claim 4, characterized in that, Adaptive threshold prediction is performed based on the cluster center change characteristics to obtain the modality matching threshold of the clusters in each frame, including: Construct a prediction input vector based on the cluster center change characteristics of each frame within a preset time window. The predicted input vector ,in, Indicates the preset time window size. ; The predicted input vector is input into a pre-trained LSTM model for adaptive threshold prediction, and the modality matching threshold of the cluster is output by the LSTM model.

6. The method according to any one of claims 1-5, characterized in that, The predicted first The wideband oscillation data of the frame Modality The modality matching threshold of the c-th cluster is ; Modal continuity matching of adjacent frames is performed using the clusters and the corresponding modal matching thresholds to obtain modal continuity recognition results, including: Calculate the first The wideband oscillation data of the frame Modality and the The wideband oscillation data of the frame Modality Weighted distance ; Compare the weighted distances The modality matching threshold is Size; In response to the weighted distance Less than the modal matching threshold Then the mode With the mode continuous.

7. A modal continuity identification device for multimodal oscillation data, characterized in that, include: The feature extraction module is used to extract features from multiple frames of broadband oscillation data to obtain the modal features of each mode of the broadband oscillation data in each frame. The clustering processing module is used to cluster multiple modes according to the modal features to obtain multiple clusters of broadband oscillation data in each frame, and to determine the cluster center variation features of broadband oscillation data in each frame. The threshold prediction module is used to perform adaptive threshold prediction based on the cluster center change characteristics to obtain the modality matching threshold of the cluster in each frame. The continuity judgment module is used to perform modal continuity matching using the cluster and the corresponding modal matching threshold to obtain the modal continuity recognition result.

8. A computer-readable storage medium, characterized in that, It stores a program that, when executed by a processor, implements the modal continuity identification method for multimodal oscillation data as described in any one of claims 1-6.

9. A terminal device, characterized in that, The method includes a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the modal continuity identification method for multimodal oscillation data as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the modal continuity identification method for multimodal oscillation data as described in any one of claims 1-6.

Images