Power grid subsynchronous oscillation suppression method based on neural network prediction model

Abstract

The invention discloses a power grid subsynchronous oscillation suppression method based on a neural network prediction model, and particularly relates to the technical field of power grid control. The method comprises the following steps: extracting a high-dimensional feature vector of power grid measurement data and performing frequency domain decomposition to obtain feature distribution information of different frequency components; utilizing a sparse coding algorithm to realize mode separation, and identifying a running state abnormal region set; constructing a dynamic coupling model to evaluate the interaction relationship between the oscillation mode and the operation state in the abnormal region and the time evolution characteristic; analyzing modal distribution characteristics through adaptive filtering, and calculating a local modal separation degree; based on oscillation evolution information and a modal structure evaluation result, power grid operation parameters and suppression device configuration are dynamically adjusted, and the recognition and response capability of a neural network to a potential dangerous subsynchronous oscillation mode is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid control, and more specifically, to a method for suppressing subsynchronous oscillation of a power grid based on a neural network prediction model. Background Art

[0002] With the continuous expansion of the scale of modern power systems, a large number of various types of power electronic devices and new energy power generation units are connected, making the problem of subsynchronous oscillation increasingly serious, and showing complex characteristics of multi-modal and multi-frequency interweaving.

[0003] In order to effectively identify and suppress subsynchronous oscillation phenomena, neural network prediction technologies based on high-dimensional power grid measurement data are widely used. However, in practical applications, it is found that traditional neural network models often have problems of aliasing and masking in the modal feature space during the feature extraction process, that is, the feature patterns of subsynchronous oscillation modes with different frequency sources or different intensities are easily confused or ignored, resulting in weak modes or emerging modes with higher potential risks that cannot be identified and predicted in a timely and accurate manner, thus restricting the reliability and effectiveness of power grid subsynchronous oscillation prediction and suppression technologies.

[0004] To solve the above problems, a technical solution is provided now. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide a method for suppressing subsynchronous oscillation of a power grid based on a neural network prediction model to solve the problems raised in the above background art.

[0006] To achieve the above object, the present invention provides the following technical solutions: A method for suppressing subsynchronous oscillation of a power grid based on a neural network prediction model includes the following steps: Extract high-dimensional feature vectors from power grid measurement data, perform frequency-domain decomposition processing on the high-dimensional feature vectors to obtain feature distribution information of different frequency components; Use a sparse coding algorithm to perform pattern separation processing on the feature distribution information, analyze the influence of changes in the power grid operation state on the subsynchronous oscillation mode, and identify a set of abnormal operation state regions; Analyze the interaction relationship between the subsynchronous oscillation mode and the power grid operation state within the set of abnormal operation state regions by constructing a dynamic coupling model, and evaluate the time evolution characteristics of the subsynchronous oscillation mode; Analyze the distribution characteristics of the subsynchronous oscillation mode within the set of abnormal operation state regions by an adaptive filtering algorithm, and evaluate the modal separation degree of the local region; Based on the time evolution characteristics of the subsynchronous oscillation mode and the modal separation degree of the local region, adjust the power grid operation parameters and the configuration of suppression devices through a feedback adjustment mechanism.

[0007] In a preferred embodiment, a high-dimensional feature vector is extracted from the power grid measurement data, and the high-dimensional feature vector is subjected to frequency-domain decomposition processing to obtain the feature distribution information of different frequency components. Specifically: Perform a fast Fourier transform on the power grid measurement data to convert the power grid measurement data from a time-domain signal to a frequency-domain signal, and obtain initial spectrum data; According to the amplitude distribution characteristics of the initial spectrum data, determine the interval thresholds of different frequency components, and divide the initial spectrum data into multiple frequency sub-intervals; In each frequency sub-interval, use the principal component analysis method to extract the principal feature components and obtain the feature distribution information characterizing the operation state of the power grid; Through the dimensionality expansion of the feature vector of the feature distribution information, a high-dimensional feature vector set is formed.

[0008] In a preferred embodiment, use the sparse coding algorithm to perform pattern separation processing on the feature distribution information, analyze the influence of the change of the power grid operation state on the subsynchronous oscillation mode, and identify the set of abnormal operation state regions. Specifically: Input the high-dimensional feature vector set into the sparse dictionary learning model to construct a sparse feature representation dictionary; Based on the sparse feature representation dictionary, perform orthogonal matching pursuit processing on the high-dimensional feature vector set to obtain sparse coding coefficients; Perform clustering analysis on the sparse coding coefficients in the time dimension and the frequency dimension, and extract the dynamic mode features of the subsynchronous oscillation mode caused by the change of the power grid operation state; According to the distribution density and frequency aggregation degree of the dynamic mode features, identify the time section and spatial location where the power grid operation state changes abnormally, and form a set of abnormal operation state regions.

[0009] In a preferred embodiment, analyze the interaction relationship between the subsynchronous oscillation mode and the power grid operation state in the set of abnormal operation state regions by constructing a dynamic coupling model, and evaluate the time evolution characteristics of the subsynchronous oscillation mode. Specifically: Based on the set of abnormal operation state regions, extract the high-dimensional feature vector sequence and sparse coding coefficient sequence in the corresponding time period, and construct a joint feature sequence reflecting the feature change trajectory; Construct a dynamic coupling model according to the joint feature sequence; Use the recursive least squares algorithm to perform dynamic estimation on the dynamic coupling model, and obtain a set of coupling relationship parameters between the change of the power grid operation state and the response of the subsynchronous oscillation mode; Based on the set of coupling relationship parameters, establish a time series evolution expression of the subsynchronous oscillation mode, and combine the time series stability judgment function of the dynamic coupling model to extract a sequence of time evolution characteristic indicators.

[0010] In a preferred embodiment, the distribution characteristics of the subsynchronous oscillation modes in the set of abnormal operation state regions are analyzed through an adaptive filtering algorithm to evaluate the modal separation degree of the local regions, specifically as follows: Based on the sparse coding coefficient sequence of the subsynchronous oscillation modes in the set of abnormal operation state regions, construct a set of time series observation vectors; Select the Kalman filter structure as the adaptive filtering algorithm framework, set the state transition matrix and the observation matrix, and initialize the state covariance and the process noise covariance; Perform recursive filtering operations on each set of time series observation vectors, and extract the dynamic amplitude sequence and frequency sequence of the corresponding modal components after filtering; According to the amplitude change range and frequency difference degree of the modal components in different local regions, calculate the modal overlap index; According to the modal overlap index, evaluate the modal separation degree between the subsynchronous oscillation modes in each local region.

[0011] In a preferred embodiment, based on the time evolution characteristics of the subsynchronous oscillation modes and the modal separation degree of the local regions, the grid operation parameters and the suppression device configuration are quantitatively adjusted through feedback control, specifically as follows: Extract the key evolution parameters of each subsynchronous oscillation mode based on the time evolution characteristic index sequence; Combine the modal overlap index and the key evolution parameters to construct a modal interference evaluation function, determine the feedback control adjustment priority, and construct a feedback control mapping table; According to the feedback control mapping table, combine the historical scheduling strategy and the current grid operation state, and use the rolling optimization algorithm to generate the grid operation parameter adjustment amount and the suppression device configuration instruction.

[0012] Technical effects and advantages of the method for suppressing grid subsynchronous oscillation based on the neural network prediction model of the present invention: By introducing methods such as frequency domain decomposition, sparse coding, modal separation, adaptive filtering, and dynamic coupling modeling into the original grid measurement data, a full-link processing mechanism from low-level feature extraction to high-level feedback regulation is established, realizing the separation and identification of oscillation modes with different frequencies and different sources and the evolutionary monitoring, and avoiding the covering effect of the dominant mode on the potential mode. Furthermore, through the comprehensive evaluation of the modal separation degree and the evolutionary characteristics, the operation parameters and the suppression device configuration are dynamically adjusted according to the region, improving the directivity and pertinence of the suppression strategy. The neural network's recognition robustness and response accuracy for complex grid multimodal subsynchronous oscillation are enhanced, and it is applicable to the grid stability operation guarantee scenario under a multi-source disturbance environment. Description of the Drawings

[0013] Figure 1 It is a schematic diagram of the method for suppressing grid subsynchronous oscillation based on the neural network prediction model of the present invention. Detailed implementation manners

[0014] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part rather than all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0015] Embodiment 1 Figure 1 A method for suppressing power grid subsynchronous oscillation based on a neural network prediction model of the present invention is given, which includes the following steps: Extract high-dimensional feature vectors from power grid measurement data, perform frequency-domain decomposition processing on the high-dimensional feature vectors, and obtain the feature distribution information of different frequency components; Use the sparse coding algorithm to perform pattern separation processing on the feature distribution information, analyze the influence of changes in the power grid operation state on the subsynchronous oscillation mode, and identify the set of abnormal operation state regions; Analyze the interaction relationship between the subsynchronous oscillation mode and the power grid operation state within the set of abnormal operation state regions by constructing a dynamic coupling model, and evaluate the time evolution characteristics of the subsynchronous oscillation mode; Analyze the distribution characteristics of the subsynchronous oscillation mode within the set of abnormal operation state regions through an adaptive filtering algorithm, and evaluate the modal separation degree of the local region; Based on the time evolution characteristics of the subsynchronous oscillation mode and the modal separation degree of the local region, adjust the power grid operation parameters and the configuration of the suppression device through a feedback regulation mechanism.

[0016] Specifically, extracting high-dimensional feature vectors from power grid measurement data and performing frequency-domain decomposition processing on the high-dimensional feature vectors to obtain the feature distribution information of different frequency components includes: Perform a fast Fourier transform on the power grid measurement data to convert the power grid measurement data from a time-domain signal into a frequency-domain signal, and obtain initial spectrum data; Specifically, collect power grid measurement data through a power grid measurement device (such as a synchronized phasor measurement unit), including multi-dimensional information such as voltage amplitude, phase, current amplitude, and frequency. Perform a frequency-domain transformation operation on the power grid measurement data. Use the fast Fourier transform method to transform the continuously sampled power grid measurement data into a spectrum representation, and obtain spectrum data including multiple frequency components and their corresponding amplitudes. The spectrum data contains energy concentration regions in different frequency intervals, which is an important basis for identifying oscillation characteristics.

[0017] For example, assume that the input signal is a sequence of voltage fluctuations obtained at a fixed sampling interval. After applying the fast Fourier transform, the power spectral density distribution in different frequency ranges can be obtained. Each frequency point corresponds to a complex representation, where the real and imaginary parts represent the amplitude information of the sine and cosine components respectively.

[0018] According to the amplitude distribution characteristics of the initial spectral data, determine the interval thresholds for different frequency components, and divide the initial spectral data into multiple frequency sub-intervals; Specifically, after the spectral data is generated, in order to clarify the characterization ability of different frequency bands for subsynchronous oscillation, it is necessary to perform segmentation processing based on the amplitude distribution characteristics. By calculating the average amplitude and standard amplitude range of the spectral data in different frequency intervals, the frequency axis is segmented to form multiple non-overlapping frequency sub-intervals. Each sub-interval contains a series of adjacent frequency points, and the corresponding amplitudes fall within a relatively close statistical range.

[0019] For example, if the amplitude within a certain frequency band is continuously higher than three times the average amplitude of the full spectrum, it can be regarded as a frequency sub-interval where oscillation aggregation may occur. By comparing the mean value, deviation, and change trend of the spectral amplitudes in each sub-interval, the sensitivity and representativeness of each sub-interval to the oscillation characteristics can be initially determined.

[0020] The set of frequency sub-intervals will be used as the input data range for principal feature extraction and dimension expansion.

[0021] In each frequency sub-interval, use the principal component analysis method to extract the principal feature components and obtain the feature distribution information characterizing the power grid operation state; Specifically, input the data of each frequency sub-interval into the principal component analysis process respectively. The principal component analysis method is used to extract the main change direction from multi-dimensional spectral amplitude data and compress redundant information to improve the model processing efficiency. For each frequency sub-interval, construct a covariance matrix, and obtain the eigenvectors that can best reflect the main direction of the data through eigenvalue decomposition. Select several eigenvectors with a cumulative contribution rate exceeding a preset ratio to form the set of principal feature components of this sub-interval.

[0022] For the spectral data matrix composed of multiple sampling points within a frequency sub-interval, use the matrix diagonalization method to determine the main axis direction, and transform the spectral data through linear combination to form a set of principal feature components. The set of principal feature components is used to characterize the change structure of the spectral data in the main axis direction within the frequency sub-interval, thereby reflecting the response characteristics of the frequency sub-interval to the change of the power grid operation state.

[0023] The output result is a set of principal feature component sets corresponding to each frequency sub-interval, which are summarized to form the feature distribution information in the frequency dimension and are the middle-level representation of the power grid state.

[0024] Form a high-dimensional feature vector set through the dimensionality expansion of the feature distribution information; Specifically, after extracting the main feature components of multiple frequency sub-intervals, the main feature components of each sub-interval are combined by feature splicing. Perform dimensionality normalization processing on the feature components output by all frequency sub-intervals to ensure that the feature values in different frequency bands are expressed on the same scale. Taking time as the axis, splice the feature components of each frequency band at the same moment in a fixed order to construct a multi-dimensional vector.

[0025] Through the above dimensionality expansion method, the main feature components in different frequency bands of the original time series are combined into a single high-dimensional feature vector, so that each time point corresponds to a high-dimensional state representation with a unified structure. The high-dimensional feature vector set can cover the comprehensive information of oscillation, steady state and disturbance behaviors at different frequencies in the power grid, and is the core input in the sparse coding and pattern recognition processes.

[0026] Specifically, use the sparse coding algorithm to perform pattern separation processing on the feature distribution information, analyze the influence of the change of the power grid operation state on the subsynchronous oscillation mode, and identify the set of abnormal operation state regions, including: Input the high-dimensional feature vector set into the sparse dictionary learning model to construct a sparse feature representation dictionary; Specifically, the high-dimensional feature vector set comes from the frequency segmentation processing and main feature component extraction of the spectrum data. In order to extract the subsynchronous oscillation mode features in the power grid operation state, the sparse dictionary learning method is used to train the high-dimensional feature vector set. The sparse dictionary learning model aims at the sparse representation of sample features in the low-dimensional space, constructs a set of complete linear bases, so that any input feature vector can be linearly represented by a finite number of dictionary basis vectors.

[0027] First, initialize a set of dictionary basis vector sets with a unified dimension, and adopt an iterative optimization strategy to update the dictionary content, so that in each round of iteration, the combined loss of the reconstruction error and the sparsity constraint function reaches the minimum. The output sparse feature representation dictionary is composed of several normalized atomic vectors, and each atomic vector is regarded as a typical subsynchronous oscillation structure unit in physical meaning, which can correspond to the local mode in the power grid operation.

[0028] The sparse feature representation dictionary is used for the sparse coding calculation of high-dimensional feature vectors.

[0029] Based on the sparse feature representation dictionary, perform orthogonal matching pursuit processing on the high-dimensional feature vector set to obtain sparse coding coefficients; Specifically, after obtaining the sparse feature representation dictionary, each high-dimensional feature vector is represented as a linear combination of several atomic vectors in the sparse feature representation dictionary. The orthogonal matching pursuit method is used to perform sparse coding on each input vector. Orthogonal matching pursuit is a greedy search algorithm that gradually approximates. In each step, the dictionary atom most relevant to the current residual is selected, and the residual is iteratively updated until a preset representation accuracy or sparsity limit is reached.

[0030] For each high-dimensional feature vector, first calculate the projection magnitudes between it and all dictionary atoms, select the atom with the largest projection amplitude as the current coding basis; then calculate the fitting contribution of this atom to the original vector, and subtract this contribution from the original vector to form a residual vector; repeat this process until the residual amplitude is lower than the preset threshold, or the number of selected atoms reaches the sparsity limit. Finally, obtain the sparse coding coefficients of the high-dimensional feature vector under the sparse dictionary, representing the linear component intensities on the atoms of each typical oscillation mode.

[0031] Perform clustering analysis on the sparse coding coefficients in the time dimension and frequency dimension to extract the dynamic mode features of the subsynchronous oscillation modes caused by changes in the power grid operation state; Specifically, extract the potential subsynchronous oscillation modes and perform two-dimensional clustering processing on the sparse coding coefficient set. Classify the sparse coding coefficient sequences at different times along the time dimension to identify the typical oscillation responses of the power grid operation state at different stages; classify the response intensities of different dictionary atoms in the coding coefficients along the frequency dimension to extract the mode activation distributions in different frequency ranges.

[0032] In the clustering analysis, select a density-based non-parametric clustering algorithm to construct the distribution map of the sparse coefficients. By setting the distance threshold and the minimum clustering scale parameter, several oscillation mode clusters are formed. Each mode cluster contains multiple data points with similar coding structures and similar time series characteristics. Each oscillation mode cluster represents a potential subsynchronous oscillation dynamic response structure, which includes the start and end times of mode activation, the response frequency range, and the set of participating dictionary atoms.

[0033] According to the distribution density and frequency aggregation degree of the dynamic mode features, identify the time segments and spatial positions where abnormal changes occur in the power grid operation state, and form a set of abnormal operation state regions; Specifically, by analyzing the distribution density of the dynamic mode features on the time axis and frequency axis, judge which regions show atypical oscillation aggregation behavior. Specifically, count the number of coding events and the frequency range span included in each dynamic mode cluster; when a certain mode cluster is frequently activated in a short time and its frequency aggregation degree exceeds the average density of the entire frequency range, it can be determined that this mode belongs to an abnormal subsynchronous oscillation response.

[0034] According to the measurement positions corresponding to the power grid nodes, map the measurement points corresponding to each coding coefficient in the dynamic mode clusters, and determine whether there is a high-density response concentrated in certain specific regions. When high-amplitude coding activities occur simultaneously at multiple measurement points within a certain region, and the response frequencies of the dynamic modes within the region are concentrated in a narrow frequency band, it can be determined as an abnormal operation state region.

[0035] The final result is a set of abnormal operation state regions, including complete indicators such as time segments, spatial positions, frequency ranges, etc., providing input for the target regions for dynamic coupling analysis and oscillation suppression control strategies.

[0036] Specifically, analyze the interaction relationship between the subsynchronous oscillation modes and the power grid operation state within the set of abnormal operation state regions by constructing a dynamic coupling model, and evaluate the time evolution characteristics of the subsynchronous oscillation modes, including: Based on the set of abnormal operation state regions, extract the high-dimensional feature vector sequence and sparse coding coefficient sequence within the corresponding time period, and construct a joint feature sequence reflecting the feature change trajectory; Specifically, according to the start time and end time marked in the set, as well as the corresponding geographical node positions of the power grid, retrospectively extract the original feature data corresponding to this time segment and spatial region, including: extracting the high-dimensional feature vector sequence at each sampling moment within this time period, which is spliced by the main feature component dimensions of the frequency sub-intervals and reflects the multi-band combined structure of the power grid state; Extract the sparse coding coefficient sequence consistent with the time axis of the high-dimensional feature vectors, and this sequence records the response intensities of each time point on each atom of the sparse dictionary.

[0037] Align the high-dimensional feature vector sequence and the sparse coding coefficient sequence at the time points, and splice them into a single joint feature vector at each time point, and finally form a joint feature sequence. The joint feature sequence completely retains the structural information between the changes in the power grid operation state and the activation behavior of the oscillation modes, and is the input basis for dynamic coupling modeling.

[0038] Construct a dynamic coupling model according to the joint feature sequence; Specifically, perform a structural decomposition on the joint feature sequence. Extract the physical quantities (such as voltage amplitude, current phase angle, frequency fluctuation, etc.) contained in the high-dimensional feature vector part as the set of power grid operation parameters and regard them as the input variables of the dynamic coupling model; regard the sparse coding coefficient part as the output variable of the dynamic coupling model, which is used to represent the activation situation of the subsynchronous oscillation modes in the sparse dictionary.

[0039] The dynamic coupling model structure adopts a multi-input single-output non-linear dynamic regression form, which can capture the causal relationship in time between the power grid operation state and the subsynchronous oscillation response. Multiple lag time terms are introduced in the dynamic coupling model to represent that the current oscillation response is not only related to the current operation parameters, but also jointly affected by the operation states at the previous or several previous moments.

[0040] For example, if the response coefficient of the oscillation mode at the current moment is jointly driven by the voltage change rate and the amplitude of frequency fluctuation at the previous two moments, a corresponding historical input variable window will be constructed in the dynamic coupling model to ensure the dynamic integrity of the causal structure.

[0041] The recursive least squares algorithm is used to dynamically estimate the dynamic coupling model to obtain a set of coupling relationship parameters between the change of the power grid operation state and the subsynchronous oscillation mode response; Specifically, the recursive least squares algorithm is introduced to perform online estimation of the parameters of the dynamic coupling model. The recursive least squares algorithm can, when updating at each sampling point, recursively adjust the parameters of the dynamic coupling model according to the previous moment's estimated value, the current input variable and output variable, so as to achieve dynamic adaptive fitting.

[0042] The parameter update process mainly includes: determining the incremental influence of the current input variable on the historical data covariance matrix, and updating the current parameter estimated value based on the observation error and the incremental weight.

[0043] The parameter estimation result forms a set of coupling relationship parameters, which clearly identifies the contribution intensity of different power grid operation parameters to the subsynchronous oscillation mode response. For example, if the coupling coefficient of a certain voltage-related variable remains high in the model for a long time, it indicates that this variable is the dominant driving factor for the formation of the current oscillation mode.

[0044] Based on the set of coupling relationship parameters, a time-series evolution expression of the subsynchronous oscillation mode is established, and combined with the time-series stability determination function of the dynamic coupling model, a sequence of time evolution characteristic indicators is extracted; Specifically, a time-series evolution expression of the oscillation mode is established using the set of coupling relationship parameters. The time-series evolution expression is based on the dynamic coupling model and expresses the changing trend of the predicted value of the oscillation response over time through a regression structure.

[0045] When constructing the time-series evolution expression, the functional relationship of the output response variable to multiple input variables is clearly expressed, and the coefficients between all variables are obtained from the recursive least squares estimation. The time-series evolution expression also contains a lag structure of time steps to reflect the dynamic lag of the response.

[0046] To determine whether the oscillation evolution process tends to be stable, a time-series stability determination function is introduced to analyze the convergence and change trends of the parameters in the evolution expression. For example, if some coupling coefficients vary violently or show a divergent trend in multiple time windows, it is determined to be in an unstable state.

[0047] Combining the structure of the time-series evolution expression and the output results of the time-series stability determination function, a sequence of time-evolution characteristic indicators is extracted, including the oscillation amplitude growth rate, response duration window, coupling strength fluctuation degree, etc. These indicators are used to identify the adjustment priorities and intervention targets in subsequent control strategies.

[0048] Specifically, the distribution characteristics of the subsynchronous oscillation modes in the set of abnormal operation state regions are analyzed through an adaptive filtering algorithm, and the modal separation degree of the local region is evaluated, including: Based on the sparse coding coefficient sequence of the subsynchronous oscillation modes in the set of abnormal operation state regions, a set of time-series observation vectors is constructed; Specifically, from the set of abnormal operation state regions, the sparse coding coefficient sequence contained in each abnormal region is extracted. The sparse coding coefficient sequence is derived from the response intensities of multiple nodes in the power grid to the sparse dictionary atoms within a continuous time period, and its dimension includes multiple time points and multiple dictionary atom numbers. Arrange the sparse response intensities of each dictionary atom in chronological order to form the time-series response vector of the atom.

[0049] Taking a measurement node in each region as an example, its sparse coding coefficients can form a two-dimensional matrix driven by time. Each row of the matrix represents the response value of a dictionary atom at a certain time point, and each column represents the response change of the dictionary atom at different times. By combining the data of multiple nodes, a multi-dimensional time-series observation vector set is formed.

[0050] Select the Kalman filter structure as the framework of the adaptive filtering algorithm, set the state transition matrix and the observation matrix, and initialize the state covariance and the process noise covariance; Specifically, the Kalman filter structure is used for adaptive estimation. The filter structure includes: state variables, representing the dynamic values of the hidden modal components; observation variables, representing the original observation sequence composed of sparse coding coefficients.

[0051] Set the state transition matrix, which is used to describe the influence of the modal components at the previous moment on the current state, and the matrix dimension is the same as the number of modes. Set the observation matrix, which is used to represent the linear mapping relationship between the observation vector and the state vector. Initialize the state estimation covariance matrix and the process noise covariance matrix, which are used to evaluate the initial state uncertainty and the degree of internal system perturbation respectively. The initial parameters should be determined according to the actual power grid oscillation frequency fluctuation range and the measurement device accuracy to effectively capture the low-amplitude modal components.

[0052] Perform recursive filtering operations on each set of time series observation vectors, and extract the dynamic amplitude sequence and frequency sequence of the corresponding modal components after filtering; Specifically, input the set of observation vectors into the Kalman filter structure, and perform prediction and update operations in chronological order. At each moment, the filter predicts the current modal state based on the state estimate value and the state transition matrix at the previous moment, then calculates the observation error based on the observation vector and the observation matrix at the current moment, and corrects the current state estimate value using the weighted update rule.

[0053] After recursive filtering processing, the output result is the dynamic state sequences of multiple modal components, and each modal component corresponds to a time-varying oscillation mode that matches the dictionary atom number. For each modal component, by calculating its response envelope curve within a sliding time window, the dynamic amplitude sequence of this component within this time period can be obtained; further, by analyzing the periodic characteristics of the continuous envelope curve, the instantaneous frequency sequence of each modal component can be extracted.

[0054] The output result includes the dynamic amplitude sequences and instantaneous frequency sequences of multiple modal components, providing basic quantitative data for modal separation degree evaluation.

[0055] Calculate the modal overlap index according to the amplitude change range and frequency difference degree of the modal components in different local regions; Specifically, for amplitude overlap calculation, normalize the amplitudes of different modal components at the same moment, and calculate the integral value of their amplitude cross region. The more concentrated the amplitudes and the larger the overlapping area, the stronger the coupling trend between the modal components.

[0056] Calculate the frequency overlap. Statistically analyze the difference distribution between the instantaneous frequencies of each modal component. If the frequency difference between multiple modal components within a certain time period is less than the preset frequency resolution threshold, then frequency aggregation behavior is considered to exist within this time period.

[0057] Combine the amplitude overlap and the frequency overlap in a weighted summation manner to form the modal overlap index. The higher the value of the modal overlap index, the more difficult it is to distinguish different oscillation modes within the region, and there is a risk of coupling interference.

[0058] Evaluate the modal separation degree between the subsynchronous oscillation modes in each local region according to the modal overlap index; Specifically, the modal overlap index is processed by moving average on the time axis to obtain the trend change curve of the modal overlap index over the entire observation period. The modal overlap index is compared with a preset modal separation threshold: if the modal overlap index is less than or equal to the modal separation threshold, it indicates that there is good separation between the oscillation modes; otherwise, it indicates that there is a modal fusion phenomenon in the region, and it is difficult to effectively suppress it through a single control strategy.

[0059] The final result is the modal separation degree level corresponding to each abnormal operation state area, and the grading criteria may include completely separable, partially overlapping, and highly coupled. The evaluation result will be used as an important input condition for the oscillation suppression feedback regulation strategy.

[0060] Specifically, based on the time evolution characteristics of the subsynchronous oscillation mode and the modal separation degree of the local area, the grid operation parameters and the configuration of the suppression device are quantitatively adjusted through feedback control, including: Extract the key evolution parameters of each subsynchronous oscillation mode based on the time evolution characteristic index sequence; Specifically, the time evolution characteristic index sequence output by the dynamic coupling model is analyzed. The time evolution characteristic index sequence covers the response characteristics of each subsynchronous oscillation mode in the abnormal operation state area, and its original data comes from the time evolution of the sparse coding coefficients.

[0061] Calculate the moving average value of the response amplitude of each oscillation mode on the time axis to form a smooth curve reflecting the change of oscillation intensity over time; by calculating the derivative sequence of the smooth curve, extract the growth trend of the oscillation amplitude, that is, the average growth rate per unit time.

[0062] Extract the frequency change rate from the instantaneous frequency sequence to measure the fluctuation degree of the oscillation frequency. The larger the frequency change rate, the worse the frequency stability.

[0063] Statistically calculate the continuous response duration experienced by each oscillation mode from activation to recovery to a stable state as the duration window, indicating the retention of the grid disturbance.

[0064] The output result is a set of three-dimensional key evolution parameter sets for each oscillation mode.

[0065] Combine the modal overlap index and the key evolution parameters to construct a modal interference evaluation function, determine the priority of feedback control regulation, and construct a feedback control mapping table; Specifically, according to the modal overlap index, combined with the key evolution parameters, construct a set of function models for evaluating the degree of modal coupling interference in the region. The function model is constructed based on the following three weight factors: The weight of the modal overlap index, which measures the spatial coincidence degree of different oscillation modes on the time axis and the frequency axis; The weight of the oscillation frequency change rate is used to evaluate whether the modes with rapid changes are likely to cause control stability disorders; The weight of the response duration window is used to determine whether the oscillation event has sufficient interference persistence.

[0066] Normalize and sum the weights of the above three factors to form a unified modal interference intensity score, which takes values between zero and one. The higher the modal interference intensity score, the more serious the modal interference in the region and the higher the priority of the suppression strategy intervention.

[0067] Implement hierarchical management for different regions of the power grid. Sort according to the modal interference scores from high to low to determine the feedback control adjustment priorities for each abnormal region. The adjustment priorities directly determine the response control frequencies, adjustment amplitudes, and control response times of each region.

[0068] Construct a feedback control mapping table based on factors such as the historical operating status, available adjustment resources, voltage stability boundary, and frequency response ability of each region. The mapping table structure includes: Target adjustment parameters: Clearly define the types of operating variables to be adjusted, such as the capacity of shunt compensation equipment, active power distribution ratio, reactive power adjustment amplitude, etc.; Power grid operation constraint conditions: Include node voltage limits, maximum branch power flows, load limits of interconnected transformers, etc.; Adjustment step size: Set the adjustable amplitude and interval of each round of control strategy to avoid secondary oscillations in the power grid caused by large disturbances.

[0069] The feedback control mapping table serves as the core control dictionary called by the subsequent optimization algorithm to ensure that each adjustment target has physical boundaries and response limitations.

[0070] According to the feedback control mapping table, combine the historical dispatching strategy and the current power grid operating status, and use the rolling optimization algorithm to generate the adjustment amount of power grid operating parameters and the configuration instructions of suppression devices; Specifically, use the rolling optimization algorithm to execute the generation of dynamic control strategies. The rolling optimization algorithm jointly inputs the power grid operating status, historical dispatching scheme, and feedback control mapping table, establishes the objective function and constraint equations, and performs an optimization calculation once in each dispatching cycle.

[0071] The optimization objective is to minimize the sum of the system oscillation response amount and the adjustment resource cost while satisfying all operating constraint conditions. During the solution process, for the adjustment priorities of the oscillation modes in each region, incrementally adjust the target adjustment parameters in sequence to form a new adjustment amount of operating parameters. Synchronously generate the configuration instructions of suppression devices, including the startup timing of power electronic devices, compensation amount setting, and switching path selection, etc.

[0072] The optimized results are sent to the power grid operation control system in real time. The power grid operation control system performs specific adjustment operations and updates the operation status in the next scheduling cycle, repeating the rolling iteration to achieve continuous intervention and dynamic suppression of the oscillation process.

[0073] The above formulas are all dimensionless and take their numerical values for calculation. The formula is obtained by collecting a large amount of data for software simulation to get a formula closest to the actual situation. The preset parameters and threshold selection in the formula are set by those skilled in the art according to the actual situation.

[0074] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wire (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that the computer can access or a data storage device such as a server or data center that contains one or more collections of available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0075] Those of ordinary skill in the art can realize that the modules and algorithm steps of the examples described in combination with the embodiments disclosed herein can be implemented by electronic hardware or a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0076] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and modules described above can refer to the corresponding processes in the foregoing method embodiments and will not be repeated here.

[0077] In several embodiments provided in the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division. In actual implementation, there may be other division methods. For example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling, direct coupling, or communication connection between each other can be through some interfaces. The indirect coupling or communication connection of the devices or modules can be in electrical, mechanical, or other forms.

[0078] The modules described as separate components may or may not be physically separated. The components displayed as modules may or may not be physical modules. They can be located in one place or distributed to multiple network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0079] In addition, in each embodiment of the present application, the functional modules can be integrated in a processing module, or each module can exist physically alone, or two or more modules can be integrated in one module.

[0080] If the above functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0081] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present application, and all should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0082] Finally, the above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for suppressing subsynchronous oscillation of power grid based on neural network prediction model, characterized in that, It includes the following steps: Extract high-dimensional feature vectors from power grid measurement data, perform frequency-domain decomposition processing on the high-dimensional feature vectors, and obtain the feature distribution information of different frequency components; Use the sparse coding algorithm to perform pattern separation processing on the feature distribution information, analyze the impact of power grid operation state changes on the subsynchronous oscillation mode, and identify the set of abnormal operation state regions; Analyze the interaction relationship between the subsynchronous oscillation mode and the power grid operation state within the set of abnormal operation state regions by constructing a dynamic coupling model, and evaluate the time evolution characteristics of the subsynchronous oscillation mode; Analyze the distribution characteristics of the subsynchronous oscillation mode within the set of abnormal operation state regions through the adaptive filtering algorithm, and evaluate the modal separation degree of the local region; Based on the time evolution characteristics of the subsynchronous oscillation mode and the modal separation degree of the local region, adjust the power grid operation parameters and the configuration of suppression devices through the feedback regulation mechanism.

2. The method for suppressing subsynchronous oscillation of power grid based on neural network prediction model according to claim 1, characterized in that, Extract high-dimensional feature vectors from power grid measurement data, perform frequency-domain decomposition processing on the high-dimensional feature vectors, and obtain the feature distribution information of different frequency components. Specifically: Perform fast Fourier transform on the power grid measurement data, convert the power grid measurement data from a time-domain signal to a frequency-domain signal, and obtain the initial spectrum data; According to the amplitude distribution characteristics of the initial spectrum data, determine the interval thresholds of different frequency components, and divide the initial spectrum data into multiple frequency sub-intervals; Adopt the principal component analysis method to extract the principal feature components within each frequency sub-interval, and obtain the feature distribution information characterizing the power grid operation state; Form a high-dimensional feature vector set through the dimensionality expansion of the feature vectors of the feature distribution information.

3. The method for suppressing subsynchronous oscillation of power grid based on neural network prediction model according to claim 2, characterized in that, Use the sparse coding algorithm to perform pattern separation processing on the feature distribution information, analyze the impact of power grid operation state changes on the subsynchronous oscillation mode, and identify the set of abnormal operation state regions. Specifically: Input the high-dimensional feature vector set into the sparse dictionary learning model to construct a sparse feature representation dictionary; Based on the sparse feature representation dictionary, perform orthogonal matching pursuit processing on the high-dimensional feature vector set to obtain the sparse coding coefficients; Perform clustering analysis on the sparse coding coefficients in the time dimension and the frequency dimension, and extract the dynamic mode features of the subsynchronous oscillation mode caused by the power grid operation state changes; According to the distribution density and frequency aggregation degree of the dynamic mode features, identify the time segments and spatial positions where the power grid operation state undergoes abnormal changes, and form the set of abnormal operation state regions.

4. The method for suppressing subsynchronous oscillation of power grid based on neural network prediction model according to claim 3, characterized in that, Analyze the interaction relationship between the subsynchronous oscillation mode and the power grid operation state within the set of abnormal operation state regions by constructing a dynamic coupling model, and evaluate the time evolution characteristics of the subsynchronous oscillation mode. Specifically: Based on the set of abnormal operation state regions, extract the high-dimensional feature vector sequence and the sparse coding coefficient sequence within the corresponding time period, and construct a joint feature sequence reflecting the feature change trajectory; Construct a dynamic coupling model according to the joint feature sequence; Use the recursive least squares algorithm to perform dynamic estimation on the dynamic coupling model, and obtain the coupling relationship parameter set between the power grid operation state change and the subsynchronous oscillation mode response; Based on the coupling relationship parameter set, establish the time series evolution expression of the subsynchronous oscillation mode, and combine the time series stability judgment function of the dynamic coupling model to extract the time evolution characteristic index sequence.

5. The method for suppressing subsynchronous oscillation of power grid based on neural network prediction model according to claim 4, characterized in that, Analyze the distribution characteristics of subsynchronous oscillation modes in the set of abnormal operation state regions through an adaptive filtering algorithm, and evaluate the modal separation degree of local regions, specifically as follows: Based on the sparse coding coefficient sequence of the subsynchronous oscillation modes in the set of abnormal operation state regions, construct a set of time series observation vectors; Select the Kalman filter structure as the adaptive filtering algorithm framework, set the state transition matrix and observation matrix, and initialize the state covariance and process noise covariance; Perform recursive filtering operations on each set of time series observation vectors, and extract the dynamic amplitude sequence and frequency sequence of the corresponding modal components after filtering; Calculate the modal overlap index according to the amplitude change range and frequency difference degree of the modal components in different local regions; Evaluate the modal separation degree between subsynchronous oscillation modes in each local region according to the modal overlap index.

6. The method for suppressing subsynchronous oscillation of power grid based on neural network prediction model according to claim 5, characterized in that, Based on the time evolution characteristics of subsynchronous oscillation modes and the modal separation degree of local regions, quantitatively adjust the grid operation parameters and suppress the device configuration through feedback control, specifically as follows: Extract the key evolution parameters of each subsynchronous oscillation mode based on the time evolution characteristic index sequence; Combine the modal overlap index and the key evolution parameters to construct a modal interference evaluation function, determine the feedback control adjustment priority, and construct a feedback control mapping table; According to the feedback control mapping table, combine the historical dispatching strategy and the current grid operation state, and use the rolling optimization algorithm to generate the grid operation parameter adjustment amount and the suppression device configuration instruction.

Images