Power grid broadband oscillation traceability method and system based on modal dynamic modulation graph network

Abstract

The present invention discloses a method and system for tracing the source of broadband power grid oscillations based on a modal dynamic modulation graph network. The method of the present invention includes inputting the oscillation data sequence of the power grid and the power system graph into the modal dynamic modulation graph network to obtain the oscillation source localization result based on the power system graph. The processing of each level of the modal dynamic modulation graph module in the modal dynamic modulation graph network includes: modulating the oscillation data sequence through a modal modulator and then performing eigenmode decomposition; extracting key features through convolutional layers and pooling layers; performing dynamic modulation based on a Prony dynamic modulator and segmenting feature blocks; extracting key features through convolutional layers and pooling layers; extracting node features through a graph network; and using a classifier for the node features to achieve the tracing of the broadband power grid oscillations. The purpose of the present invention is to reduce the interference of other information on the broadband oscillation information, dynamically analyze the data from both the time and space dimensions, and quickly and accurately achieve the localization of the broadband oscillation source.

Description

Technical Field

[0001] The present invention relates to a broadband oscillation source localization technology in the field of power systems, and specifically relates to a method and system for tracing the origin of power grid broadband oscillations based on a modal dynamic modulation graph network. Background Art

[0002] With the penetration of a high proportion of power electronic devices and distributed energy, the low-frequency oscillations observed in traditional power systems have gradually shifted to broadband oscillations. Accurately locating these oscillation sources is of great significance for ensuring the stability and security of power grid operation. However, broadband oscillations often have regional, highly random, and multi-source characteristics, and the power grid acquisition information also contains a large amount of redundant information, which greatly increases the difficulty of accurately locating broadband oscillation sources. Existing signal processing methods such as continuous wavelet transform and fast Fourier transform are not applicable to non-stationary signals, and often have problems such as mode mixing and slow calculation speed, and cannot timely and accurately extract oscillation components from the acquisition information; in terms of oscillation source localization methods, the current mainstream localization methods involve energy flow analysis. However, this method faces challenges in many aspects such as system structure, measurement resolution, and noise interference; the high-resolution data collection ability of waveform measurement units provides potential for broadband oscillation source localization. However, the introduction of high-frequency data analysis plus the dynamic characteristics of broadband oscillation sources has made the application of this method more and more complex; inspired by graph neural networks, methods based on graph neural networks can be effectively used for the prediction of fault locations. Nevertheless, the above methods mainly focus on low-frequency oscillations or major fault monitoring, and still ignore the time-varying and broadband characteristics of oscillations. Summary of the Invention

[0003] The technical problem to be solved by the present invention: Aiming at the above problems of the prior art, a method and system for tracing the origin of power grid broadband oscillations based on a modal dynamic modulation graph network are provided. The present invention aims to reduce the interference of other information on broadband oscillation information, dynamically analyze the data from both time and space dimensions, and quickly and accurately locate the broadband oscillation source.

[0004] In order to solve the above technical problems, the technical solutions adopted by the present invention are as follows:

[0005] A method for tracing the source of wide - frequency oscillations in a power grid based on a modal dynamic modulation graph network includes the following steps: obtaining an oscillation data sequence of the power grid and a power system graph, where the nodes in the power system graph are generators or loads in the power grid, and the edges are lines; inputting the oscillation data sequence of the power grid and the power system graph into the modal dynamic modulation graph network to obtain an oscillation source localization result based on the power system graph. The modal dynamic modulation graph network includes multiple - level modal dynamic modulation graph modules. The processing of the modal dynamic modulation graph module for the input oscillation data sequence of the power grid and the power system graph includes: S1, modulating the oscillation data sequence through a modal modulator and then performing characteristic modal decomposition to remove redundant and mixed - mode information; S2, extracting key features from the oscillation data after removing redundant and mixed - mode information through a convolutional layer and a pooling layer; S3, dynamically modulating the key features of the wide - frequency oscillation information based on a Prony dynamic modulator to calculate the wide - frequency oscillation frequency and determine the frequency window size, and dividing the wide - frequency oscillation information into feature blocks according to the frequency window size; S4, extracting key features from the dynamically modulated feature blocks through a convolutional layer and a pooling layer; S5, using the key features of the feature blocks and the power system graph as inputs to the graph network, and performing information transmission through the graph network to obtain node features of each node; S6, using a classifier based on the node features of each node finally output by the graph network to achieve oscillation source tracing.

[0006] Optionally, the oscillation data in the oscillation data sequence includes voltage , frequency and rate of change of frequency RoCoF. The voltage , frequency are normalized and scaled within the interval . The calculation function expression of the rate of change of frequency RoCoF is:

[0007] ,

[0008] In the above formula, is the rate of change of frequency RoCoF at time t, and are the normalized frequencies at time t + 1 and time t respectively , is the time interval; the function expression of the power system graph is , where is the node set, is the edge set, is the node feature, and the node feature includes the voltage , frequency and rate of change of frequency RoCoF of the node. The voltage , frequency are normalized and scaled within the interval inside, the edge set is encoded as an adjacency matrix , where is the number of nodes, and for any th row and th column element of the adjacency matrix and , if there is an edge between nodes which is an element of the edge set , then .

[0009] Optionally, when the oscillatory data sequence is modulated by a modal modulator and then subjected to eigenmode decomposition to remove redundant and mixed modal information in step S1, the functional expression of the input signal adopted by the modal modulator is:

[0010] ,

[0011] In the above formula, is the input signal adopted by the modal modulator, is the number of components of the internal periodic oscillation control signal, is the amplitude of the th component of the internal periodic oscillation control signal, is the time, is the index of the th component of the internal periodic oscillation control signal, is the th offset parameter of the internal periodic oscillation control signal component, used to simulate the irregular characteristics of the oscillation component; is the number of components of the external random perturbation signal, is the amplitude of the th component of the external random perturbation signal, is the th index of the external random perturbation signal component, is the adjacent perturbation time interval of the external random perturbation, is the number of components of the harmonic interference signal, is the amplitude of the th component of the harmonic interference signal, is the phase of the th harmonic interference signal component, is noise; the composite signal is a signal composed of the composite of oscillation and redundant information, and the functional expression of the composite signal is:

[0012] ,

[0013] ,

[0014] ,

[0015] ,

[0016] ,

[0017] In the above formula, is the output signal of the modal modulator, is the number of modes, is the th-order modal amplitude, is the th-order modal damping factor, is the th-order modal angular frequency, is the imaginary unit, is time, is the th-order modal eigenvalue, and are real numbers; and when the oscillation data sequence is modulated by the modal modulator and then the characteristic mode decomposition is performed to remove redundant and mixed modal information in step S1, performing the characteristic mode decomposition to remove redundant and mixed modal information means inputting the input signal of the modal modulator into an adaptive finite impulse response filter FIR for filtering to achieve removing redundant and mixed modal information.

[0018] Optionally, when extracting key features through the convolutional layer and the pooling layer, the functional expression of the convolutional operation performed by the convolutional layer is:

[0019] ,

[0020] In the above formula, is the output of the convolutional operation, is the number of types of oscillation data, is the input signal of the th type of oscillation data at time is the offset time at which the convolutional kernel acts, aligned with the time axis of , is the The convolution kernel corresponding to the oscillation data at moments.

[0021] Optionally, when the key features of the broadband oscillation information are dynamically modulated based on the Prony dynamic modulator in step S3 to calculate the broadband oscillation frequency and determine the frequency window size, the functional expression for determining the frequency window size is:

[0022] ,

[0023] In the above formula, is the frequency window size, is the total time step of the output timing signal of the modulator, corresponding to the data points in the discrete time series, and the number is the angular frequency corresponding to the output timing signal of the modulator, is the frequency of the output timing signal of the modulator.

[0024] Optionally, in step S5, the key features of the feature block and the power system diagram are used as the input of the graph network, and information is transmitted through the graph network to obtain the node features of each node, including:

[0025] S5.1, using the key features of the feature block and the power system diagram as the input of the graph network, transmitting information through the graph network to obtain the node features of each node, and the functional expression for the graph network to transmit information is:

[0026] ,

[0027] In the above formula, and are the outputs of the and layers of the graph network respectively, is the activation function, the matrix , the matrix , where is the adjacency matrix encoded from the edge set , is the identity matrix, is the upper diagonal matrix of the matrix ;

[0028] S5.2, adding the original input of the graph network to the output of the last layer of the graph network through the residual block according to the following formula to generate the final output of the graph network:

[0029] ,

[0030] In the above formula, is the final output of the graph network, is the output of the last layer of the graph network, The information transfer operation for the last layer of the graph network;

[0031] S5.3. Use the gated recurrent unit (GRU) for the final output of the graph network to obtain the node features of each node, and the functional expression of the gated recurrent unit (GRU) is:

[0032] ,

[0033] In the above formula, and are respectively the hidden states at the th step and the th step, is the update gate at the th step, is the candidate hidden state at the th step, represents element-wise multiplication.

[0034] Optionally, when implementing oscillation source tracing according to the node features of each node in the final output of the graph network in step S6, the classifier used is a multi-layer perceptron, and the multi-layer perceptron introduces a class weight parameter shown in the following formula to make the weight of positive samples inversely proportional to their quantity to balance the training loss:

[0035] ,

[0036] In the above formula, is the class weight parameter, is the quantity of positive samples.

[0037] In addition, the present invention also provides a power grid broadband oscillation source tracing system based on a modal dynamic modulation graph network, including a microprocessor and a memory connected to each other, and the microprocessor is programmed or configured to execute the power grid broadband oscillation source tracing method based on the modal dynamic modulation graph network.

[0038] In addition, the present invention also provides a computer-readable storage medium, in which a computer program or instruction is stored, and the computer program or instruction is programmed or configured to execute the power grid broadband oscillation source tracing method based on the modal dynamic modulation graph network through a processor.

[0039] In addition, the present invention also provides a computer program product, including a computer program or instruction, and the computer program or instruction is programmed or configured to execute the power grid broadband oscillation source tracing method based on the modal dynamic modulation graph network through a processor.

[0040] Compared with the prior art, the present invention mainly has the following advantages: The method of the present invention can fully capture the time and space information of the electrical parameters of the power grid nodes, dynamically analyze and segment the data in different wide-frequency oscillation ranges, and through the constructed modal dynamic modulation graph network, it can dynamically, quickly and accurately locate the wide-frequency oscillation source with finer granularity, thereby effectively reducing the interference of other information on the wide-frequency oscillation information, quickly and accurately realizing the location of the wide-frequency oscillation source, having important engineering practical significance, and solving the problem of wide-frequency oscillation location in the power grid under the new power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a schematic diagram of the basic process of the method of the embodiment of the present invention.

[0042] Figure 2 It is a schematic diagram of the network structure of the modal dynamic modulation graph network in the embodiment of the present invention.

[0043] Figure 3 It is the original frequency data of the generator based on the IEEE 9-node system in the embodiment of the present invention.

[0044] Figure 4 It is the frequency data of the generator after normalization processing based on the IEEE 9-node system in the embodiment of the present invention.

[0045] Figure 5 It is a schematic diagram of the oscillation location result in the embodiment of the present invention.

[0046] Figure 6 It is the accuracy evaluation result of each power system model component in the embodiment of the present invention.

[0047] Figure 7 It is a comparison of the running times of 5 different algorithms based on the IEEE 9-node system in the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] The method for tracing the wide-frequency oscillation of the power grid based on the modal dynamic modulation graph network of the present invention aims to make full use of the window dynamic modulation of the characteristic mode decomposition and the Prony modulator, effectively reduce the interference of other information on the wide-frequency oscillation information, dynamically analyze the data from two dimensions of time and space, and quickly and accurately locate the wide-frequency oscillation source. In order to enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention will be further described in detail below with reference to the accompanying drawings in the embodiments of the present invention.

[0049] As Figure 1 and Figure 2As shown in the figure, this embodiment provides a method for tracing the wide - frequency oscillation source of a power grid based on a modal dynamic modulation graph network, including the following steps: Obtain the oscillation data sequence of the power grid and the power system graph, where the nodes in the power system graph are generators or loads in the power grid, and the edges are lines; Input the oscillation data sequence of the power grid and the power system graph into the modal dynamic modulation graph network to obtain the oscillation source localization result based on the power system graph. The modal dynamic modulation graph network includes multiple - level modal dynamic modulation graph modules. The processing of the modal dynamic modulation graph module for the input oscillation data sequence of the power grid and the power system graph includes: S1, Modulate the oscillation data sequence through a modal modulator and then perform eigen - mode decomposition to remove redundant and mixed - mode information; S2, Extract key features from the oscillation data after removing redundant and mixed - mode information through a convolutional layer and a pooling layer; S3, Dynamically modulate the key features of the wide - frequency oscillation information based on a Prony dynamic modulator to calculate the wide - frequency oscillation frequency and determine the frequency window size, and segment the wide - frequency oscillation information into feature blocks according to the frequency window size; S4, Extract key features from the dynamically modulated feature blocks through a convolutional layer and a pooling layer; S5, Use the key features of the feature blocks and the power system graph as the input of the graph network, and perform information transfer through the graph network to obtain the node features of each node; S6, Use a classifier based on the node features of each node finally output by the graph network to achieve oscillation source tracing.

[0050] When obtaining the oscillation data sequence of the power grid, it includes collecting the voltage of the power grid and recording it as a time series . The collected voltage data contains wide - frequency oscillation information and a large amount of redundant information. Combine the dataset information to perform DC filtering on the original data to extract the frequency , and normalize the oscillation characteristics (including voltage and frequency ). As an alternative implementation, in order to remove the long - term trend in the original data, a high - pass Butterworth filter is used to perform DC filtering on the original data, extract and highlight the frequency information of the wide - frequency oscillation, and normalize the oscillation characteristics (including voltage and frequency ). Figure 3 This is the original frequency data of the generator based on the IEEE 9 - bus system in this embodiment. Figure 4 This is the frequency data of the generator based on the IEEE 9 - bus system in this embodiment after normalization processing. Figure 3 The frequencies of each generator in Figure 4 are almost the same, and it is difficult to distinguish subtle changes; however, as shown in

[0051] The rate of change of frequency (RoCoF) is introduced as an additional feature to enhance the feature set and capture the frequency variations of voltage data. Specifically, in this embodiment, the oscillation data in the oscillation data sequence includes voltage , frequency , and the rate of change of frequency RoCoF. The voltage , frequency are normalized and scaled within the interval . The calculation function expression of the rate of change of frequency RoCoF is:

[0052] ,

[0053] In the above formula, is the rate of change of frequency RoCoF at time t, and are the normalized frequencies at time t + 1 and time t respectively , is the time interval; the function expression of the power system diagram is , where is the set of nodes, and these nodes are generators or loads in the predefined power system topology, is the set of edges, corresponding to the buses connecting these nodes, is the node feature, and the node feature includes the voltage , frequency , and the rate of change of frequency RoCoF of the node. The voltage , frequency are normalized and scaled within the interval . The set of edges is encoded as an adjacency matrix , is the number of nodes. For any element in the th row and th column of the adjacency matrix , if there is an edge between nodes and is an element of the set of edges , then , that is: it is an edge of the power system topology; otherwise , that is: it is not an edge of the power system topology.

[0054] In this embodiment, the oscillation data sequence of the power grid and the power system diagram are input into the modal dynamic modulation graph network to obtain the oscillation source localization result based on the power system diagram. For the rich and complex information in the collected data, by constructing an embeddable modal dynamic modulation graph network, automatic feature learning is achieved through end-to-end training, multi-modal adaptive decomposition, and dynamic adjustment of different frequency band window lengths. With stronger adaptability and scalability, the multi-modal information in the collected data is decomposed, the interference of redundant information is reduced, the key features of the broadband oscillation information are extracted, and the dynamic scaling of the frequency window size is realized, so that the broadband oscillation source localization task can be completed more accurately and efficiently.

[0055] In step S1 of this embodiment, when the oscillation data sequence is modulated by the modal modulator and then subjected to feature modal decomposition to remove redundant and mixed modal information, the functional expression of the input signal adopted by the modal modulator is:

[0056] ,

[0057] In the above formula, is the input signal adopted by the modal modulator, is the number of internal periodic oscillation control signal components, is the amplitude of the i-th internal periodic oscillation control signal component, is the -th composite signal of the internal periodic oscillation control signal components, is time, is the -th index of the internal periodic oscillation control signal component, is the adjacent perturbation time interval of the internal periodic oscillation; is the -th offset parameter of the internal periodic oscillation control signal component, used to simulate the irregular characteristics of the oscillation component; is the number of external random perturbation signal components, is the amplitude of the -th external random perturbation signal component, is the -th composite signal of the external random perturbation signal components, is the -th index of the external random perturbation signal component, is the adjacent perturbation time interval of the external random perturbation, is the number of harmonic interference signal components, is the -th amplitude of the harmonic interference signal component, is the -th frequency of the harmonic interference signal component, is the The phase of each harmonic interference signal component is noise; the right side of the function expression of the input signal used by the above modal modulator consists of four parts. The first part represents the periodic oscillation within the power system, such as the periodic fluctuations of the system frequency and voltage caused by overreaction of the control system, changes in unit load, or mechanical failures. The offset parameter is set to 1% - 2% of the fluctuation period; the second part represents the random disturbances from outside the power system, such as the influence of external electromagnetic interference on measuring instruments or control systems; the third part represents the harmonic interference caused by nonlinear elements in the power system, such as transformers, rectifiers, etc.; the fourth part is noise. Among them, the composite signal is a signal composed of the composite of oscillation and redundant information, and the function expression is:

[0058] ,

[0059] ,

[0060] ,

[0061] ,

[0062] ,

[0063] In the above formula, is the output signal of the modal modulator, is the number of modes, is the amplitude of the nth-order mode, is the damping factor of the nth-order mode, is the angular frequency of the nth-order mode, is the imaginary unit, is time, is the eigenvalue of the nth-order mode, and are real numbers.

[0064] The characteristic mode decomposition can extract the decomposed modes through an adaptive finite impulse response (FIR) filter. Since the adaptive finite impulse response (FIR) filter is not restricted by the filter shape, bandwidth, and center frequency, redundant and mixed modes can be removed during the mode selection process. After iteration, the set target can be continuously approximated to obtain the required filtered signal, making the mode decomposition more thorough. Therefore, when the oscillating data sequence is modulated by a mode modulator and then undergoes characteristic mode decomposition to remove redundant and mixed mode information in step S1 of this embodiment, performing characteristic mode decomposition to remove redundant and mixed mode information means inputting the input signal of the mode modulator into the adaptive finite impulse response filter FIR for filtering to achieve the removal of redundant and mixed mode information. The output signal after characteristic mode decomposition is still time-series data containing voltage, frequency, and rate of change of frequency, but the output at this time only contains specific modes, realizing the decomposition of mixed modes and the elimination of redundant modes in the collected data.

[0065] When extracting key features through the convolutional layer and the pooling layer, the functional expression of the convolutional operation in the convolutional layer is:

[0066] ,

[0067] where, is the output signal at time after the convolutional operation, , , respectively time's convolution kernels of voltage, frequency, and rate of change of frequency RoCoF, is time's voltage, is time's rate of change of frequency RoCoF; time index, is the index of the convolution kernel, that is, the process of shifting the input signal. Extended to the convolution of multiple channels, the functional expression of the convolutional operation in the convolutional layer is:

[0068] ,

[0069] In the above formula, is the output of the convolutional operation, is the number of types of oscillating data, is time's input signal of the th type of oscillating data, is the offset time at which the convolution kernel acts, aligned with the time axis of , is the th type of oscillating data at The convolution kernel corresponding to the moment. The convolution pooling layer extracts key features from the input voltage, frequency, and frequency change rate, and performs weighted summation and dimensionality reduction to improve the computational efficiency of the model. At this time, the output is no longer the signal with the original granularity, but still the time-series data containing the key features of voltage, frequency, and frequency change rate, with sufficient oscillation information to ensure the processing of oscillation information by the subsequent Prony dynamic modulation.

[0070] Input the results of convolution and pooling operations into the Prony-based dynamic modulator. The Prony algorithm is suitable for identifying oscillation modes, decay rates, and frequency components, and is widely used in signal processing of power systems. The oscillation modes in the wide-frequency oscillation information after modal modulation are clearer. Analyze and calculate the frequencies of these oscillation modes using the output results after the aforementioned modal modulation. Divide different-sized frequency windows for different-frequency oscillation modes, and analyze each oscillation mode with an appropriate granularity to better adapt to the characteristics of different modes in wide-frequency oscillations, improve the frequency resolution, and thus more effectively identify and locate the wide-frequency oscillation source. When calculating the wide-frequency oscillation frequency based on the Prony dynamic modulator and determining the size of the frequency window in step S3 of this embodiment, the functional expression for determining the size of the frequency window is:

[0071] ,

[0072] In the above formula, is the size of the frequency window, is the total time step of the output time-series signal of the modulator, corresponding to the data points in the discrete time series, is the angular frequency corresponding to the output time-series signal of the modulator, is the frequency of the output time-series signal of the modulator. Then, the dynamic modulator uses the window size to divide the input time series into dynamic blocks , where each block contains time steps. The dynamic modulator calculates the wide-frequency oscillation frequency, determines the size of the frequency window, and divides the data into manageable segments. This segmentation enables the graph convolutional neural network (GCN) and gated recurrent unit (GRU) to adjust the data analysis granularity of wide-frequency oscillations in the time dimension, enabling the model to effectively solve the wide-frequency oscillation source localization task.

[0073] The modal dynamic modulation graph network includes multiple levels of modal dynamic modulation graph modules. Through the multiple levels of modal dynamic modulation graph modules, the fusion of modal decomposition and the modal dynamic modulator are stacked in multiple layers, and the graph topology information is fused to form the modal dynamic modulation graph network, aggregating the spatial dependencies and interactions within the power grid. In step S5 of this embodiment, the key features of the feature block and the power system graph are used as the input of the graph network, and the information transmission through the graph network to obtain the node features of each node includes:

[0074] S5.1, use the key features of the feature block and the power system graph as the input of the graph network, and perform information transmission through the graph network to obtain the node features of each node. The function expression for the information transmission of the graph network is:

[0075] ,

[0076] In the above formula, and are the outputs of the and layers of the graph network respectively, is the activation function, matrix , matrix , where is the adjacency matrix obtained by encoding the edge set , is the identity matrix, is the upper diagonal matrix of matrix ;

[0077] S5.2, superimpose the original input of the graph network on the output of the last layer of the graph network according to the following formula through the residual block to generate the final output of the graph network:

[0078] ,

[0079] In the above formula, is the final output of the graph network, is the output of the last layer of the graph network, is the information transmission operation of the last layer of the graph network;

[0080] S5.3, use the gated recurrent unit GRU for the final output of the graph network to obtain the node features of each node. Take the final output of the graph network as the input. The dynamic modulator dynamically adjusts the frequency window size w, allowing the gated recurrent unit GRU to divide the data into smaller blocks and process the data with finer granularity. In this way, the gated recurrent unit GRU can effectively capture the temporal dependencies and modalities in each block, learn from the sequential characteristics of the time series to improve its performance in processing broadband oscillations. In this embodiment, the function expression of the gated recurrent unit GRU is:

[0081] ,

[0082] In the above formula, and are the hidden states of the -th and -th steps respectively, is the update gate of the -th step, is the candidate hidden state of the -th step, represents element-wise multiplication. This process enables the GRU to utilize the spatial relationships identified by the GNN and the temporal dynamics inherent in the time series data, thereby improving the overall modeling performance. After finally obtaining the node features of each node, overlaying them on the power system diagram, the result can be obtained as shown in Figure 5 .

[0083] When using a classifier to achieve oscillation source tracing based on the node features of each node finally output by the graph network in step S6, the classifier used can be selected according to actual needs. For example, as an optional implementation, the classifier used in this embodiment is a multi-layer perceptron (MLP), and the multi-layer perceptron introduces a class weight parameter shown in the following formula to make the weight of positive samples inversely proportional to their quantity to balance the training loss:

[0084] ,

[0085] In the above formula, is the class weight parameter, is the number of positive samples. By setting the class weight parameter , the weight of positive samples can be made inversely proportional to their quantity. When the number of positive samples is small (i.e., is small), the value of the class weight parameter is large, giving higher weight to positive samples. When the number of positive samples is large, the weight value is small, making the influence of positive and negative samples more balanced. In a multi-layer perceptron (MLP), the data imbalance problem may have a significant impact on the training and performance of the model. Introducing the class weight parameter It can effectively alleviate these problems. This setting can reduce the impact of the data imbalance problem on the model, effectively utilize the spatial information from the graph network GCN and the temporal insights captured by the gated recurrent unit GRU, and produce more robust wideband oscillation source localization results. In this embodiment, the graph network automatically learns based on the voltage differences and frequency characteristics between power system nodes. The input temporal voltage data is segmented into feature blocks with different-sized frequency windows after modal dynamic modulation, which can improve the localization accuracy of the graph network. These data also correspond to their respective nodes, and then are classified by a multi-layer perceptron (MLP) to determine whether nodes 1, 2, 3, …, n are oscillation sources. The final output is the node numbers corresponding to the oscillation sources.

[0086] To better understand the contributions of each component in the power grid wideband oscillation source tracing method based on the modal dynamic modulation graph network in this embodiment, ablation experiments are conducted on this model, and these components are incrementally added to illustrate their respective contributions to the model performance. The accuracy evaluation results of 4 power system model components are as Figure 6 shown, where "FMD" represents the modal modulator, "Prony" represents the Prony dynamic modulator, "GCN" represents the link using the graph network for information transmission in step S5.1, "Residual" represents the residual block used in step S5.2, and "GRU" represents the gated recurrent unit GRU used in step S5.3. According to Figure 6 the ablation experiment results, the complete solution ("FMD + Prony + GCN + Residual + GRU") (the method of this embodiment) achieves an accuracy of 91.3% in the IEEE 9-node system, and each functional component has played a certain role.

[0087] To verify the performance superiority of the power grid wideband oscillation source tracing method based on the modal dynamic modulation graph network in this embodiment, 8 latest existing methods (including 4 traditional models and 4 machine learning models) are selected based on the IEEE 9-node system for comparative experiments with the method proposed in this embodiment. The data results are shown in Table 1. For the machine learning models, a comparison is made again with the single-task running time as the measurement index, and the comparison results are as Figure 7 shown.

[0088] Table 1 Comparison of Different Existing Methods Based on the IEEE 9-Node System

[0089]

[0090] As shown in Table 1 and Figure 7As shown in the figure, the accuracy, area under the curve (AUC), and F1 value of the modal dynamic modulation graph network proposed by the method of this embodiment are all better than those of the models adopted by other methods. In the IEEE 9-node system, the positioning accuracy and precision reach [25%, 38%] and [30%, 90%] respectively; and in the single task, the average running time is 10 ms, which is better than the models adopted by other methods, demonstrating its real-time positioning ability. In addition, based on the IEEE 9-node system, the modal dynamic modulation graph network proposed by the method of this embodiment is further experimented, and the method is used to perform positioning analysis on the oscillation conditions of different frequencies. Table 2 shows the experimental results of the positioning of oscillation sources with different frequencies based on the IEEE 9-node system.

[0091] Table 2 Experimental Results of the Positioning of Oscillation Sources with Different Frequencies Based on the IEEE 9-Node System

[0092]

[0093] Referring to Table 2, it can be seen that in the broadband oscillation positioning task, the modal dynamic modulation graph network proposed by the method of this embodiment exceeds the models of other methods in terms of effectiveness and superiority, achieving the state-of-the-art performance.

[0094] The power grid broadband oscillation source tracing method based on the modal dynamic modulation graph network in this embodiment can fully capture the time and space information of the electrical parameters of the power grid nodes, construct an adaptive scaling mechanism based on eigenmode decomposition and Prony modulator, and realize the dynamic scaling of the window size of the key features of the oscillation information, which can effectively cope with oscillations in different frequency bands and multiple modes, and dynamically achieve the precise positioning of the broadband oscillation source with a finer granularity; the modal dynamic modulator is an embeddable neural network layer, which is convenient for calling, plugging and stacking, and has strong portability. The method of this embodiment can effectively reduce the interference of other information on the broadband oscillation information, effectively solve the problems of wide broadband range and difficult window selection, effectively solve the problem of multi-modal broadband oscillation positioning, realize the fast and accurate positioning of the broadband oscillation source, and has important engineering practice significance.

[0095] In addition, this embodiment also provides a power grid broadband oscillation source tracing system based on the modal dynamic modulation graph network, including a microprocessor and a memory connected to each other, and the microprocessor is programmed or configured to execute the power grid broadband oscillation source tracing method based on the modal dynamic modulation graph network.

[0096] In addition, this embodiment also provides a computer-readable storage medium, in which a computer program or instruction is stored, and the computer program or instruction is programmed or configured to execute the power grid broadband oscillation source tracing method based on the modal dynamic modulation graph network through a processor.

[0097] In addition, this embodiment also provides a computer program product, including a computer program or instructions, which are programmed or configured to execute the power grid broadband oscillation source tracing method based on the modal dynamic modulation graph network through a processor.

[0098] Those skilled in the art should understand that the technical solutions provided by the embodiments of the present application can be in the form of a method, a system, or a computer program product. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes. The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of processes and / or blocks in the flowchart and / or block diagram can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for realizing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks. These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device realizes the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks. These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Therefore, the instructions executed on the computer or other programmable device provide steps for realizing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0099] The above is only the preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the idea of the present invention belong to the protection scope of the present invention. It should be noted that for those of ordinary skill in the art in this technical field, several improvements and refinements made without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A method for tracing the source of broadband oscillations in a power grid based on a modal dynamic modulation graph network, characterized in that: The method comprises the following steps: obtaining an oscillation data sequence of a power grid and a power system diagram, wherein the nodes in the power system diagram are generators or loads in the power grid and the edges are lines; The oscillation data sequence of the power grid and the power system diagram are input into the modal dynamic modulation diagram network to obtain the oscillation source positioning result based on the power system diagram. The modal dynamic modulation diagram network includes a multi-level modal dynamic modulation diagram module. The modal dynamic modulation diagram module processes the input oscillation data sequence of the power grid and the power system diagram, including: S1, modulating the oscillation data sequence through a modal modulator and then performing characteristic mode decomposition to remove redundant and mixed mode information; S2, extract key features from the oscillation data after removing redundant and mixed modal information through convolution layers and pooling layers; S3, dynamically modulate the key features of the broadband oscillation information based on the Prony dynamic modulator to calculate the broadband oscillation frequency and determine the frequency window size, and divide the broadband oscillation information into feature blocks according to the frequency window size; S4, extract key features from the dynamically modulated feature blocks through convolution layers and pooling layers; S5, take the key features of the feature blocks and the power system diagram as inputs of the graph network, and transfer information through the graph network to obtain the node features of each node; S6, use the classifier to realize oscillation tracing according to the node features of each node finally output by the graph network.

2. The method for tracing the source of broadband oscillation of a power grid based on a modal dynamic modulation graph network according to claim 1 is characterized in that: The oscillation data in the oscillation data sequence includes voltage ,frequency and the frequency change rate RoCoF, the voltage ,frequency The interval scaled by normalization The calculation function expression of the frequency change rate RoCoF is: ， In the above formula, is the frequency change rate RoCoF at time t, and are the normalized frequencies at time t+1 and time t respectively , is the time interval; the functional expression of the power system diagram is ,in is a node set, is the edge set, is the node characteristic, and the node characteristic includes the voltage of the node ,frequency and the frequency change rate RoCoF, the voltage ,frequency The interval scaled by normalization In the edge set is encoded as an adjacency matrix , is the number of nodes, the adjacency matrix Any of Line Elements of a column , if the node and Between is an edge set The elements in ,otherwise .

3. The method for tracing the source of broadband oscillation of a power grid based on a modal dynamic modulation graph network according to claim 1 is characterized in that: In step S1, when the oscillation data sequence is modulated by a modal modulator and then subjected to characteristic modal decomposition to remove redundant and mixed modal information, the function expression of the input signal used by the modal modulator is: ， In the above formula, is the input signal used by the modal modulator, is the number of internal periodic oscillation control signal components, is the amplitude of the i-th internal periodic oscillation control signal component, For the A composite signal of the internal periodic oscillation control signal components, For time, For the The index of the internal periodic oscillation control signal component, is the time interval between adjacent disturbances of the internal periodic oscillation; For the An offset parameter of an internal periodic oscillation control signal component, used to simulate the irregular characteristics of the oscillation component; is the number of external random disturbance signal components, For the The amplitude of the external random disturbance signal component, For the A composite signal of external random disturbance signal components, For the The index of the external random disturbance signal component, is the time interval between adjacent disturbances of external random disturbances, is the number of harmonic interference signal components, For the The amplitude of the harmonic interference signal component, For the The frequency of the harmonic interference signal component, For the The phase of the harmonic interference signal component, is noise; the composite signal is a signal composed of oscillation and redundant information, and the function expression of the composite signal is: ， ， ， ， ， In the above formula, is the output signal of the modal modulator, is the number of modes, For the The amplitude of the first mode, For the The damping factor of the first mode, For the The angular frequency of the first mode, is an imaginary unit, For time, For the The eigenvalues ​​of the order modes, and is a real number; and in step S1, when the oscillation data sequence is modulated by the modal modulator and then the characteristic modal decomposition is performed to remove redundant and mixed modal information, the characteristic modal decomposition to remove redundant and mixed modal information refers to inputting the input signal of the modal modulator into the adaptive finite impulse response filter FIR for filtering to achieve the removal of redundant and mixed modal information.

4. The method for tracing the source of broadband oscillation of a power grid based on a modal dynamic modulation graph network according to claim 1 is characterized in that: When extracting key features through the convolution layer and the pooling layer, the function expression of the convolution operation performed by the convolution layer is: ， In the above formula, is the output of the convolution operation, is the number of types of oscillatory data, for The moment The input signal of the oscillating data is is the convolution kernel The offset time is Align the time axis, For the Oscillation data in The convolution kernel corresponding to the moment.

5. The method for tracing the source of broadband oscillation of a power grid based on a modal dynamic modulation graph network according to claim 1 is characterized in that: In step S3, when the key features of the broadband oscillation information are dynamically modulated based on the Prony dynamic modulator to calculate the broadband oscillation frequency and determine the frequency window size, the function expression for determining the frequency window size is: ， In the above formula, is the frequency window size, is the total time step of the modulator output timing signal, corresponding to the data points in the discrete time series, is the angular frequency corresponding to the modulator output timing signal, The frequency of the modulator output timing signal.

6. The method for tracing the source of broadband oscillation of a power grid based on a modal dynamic modulation graph network according to claim 1, characterized in that: In step S5, the key features of the feature block and the power system diagram are used as inputs of the graph network, and information is transmitted through the graph network to obtain the node features of each node, including: S5.1, the key features of the feature block and the power system diagram are used as the input of the graph network, and information is transmitted through the graph network to obtain the node features of each node, and the function expression of the graph network for information transmission is: ， In the above formula, and They are the graph networks and The output of the layer, is the activation function, the matrix ,matrix ,in Edge Set The encoded adjacency matrix is is the identity matrix, For the matrix The upper diagonal matrix of ; S5.2, the original input of the graph network is superimposed on the output of the last layer of the graph network through the residual block according to the following formula to generate the final output of the graph network: ， In the above formula, is the final output of the graph network, is the output of the last layer of the graph network, It is the information passing operation of the last layer of the graph network; S5.3, the final output of the graph network is used to obtain the node features of each node using a gated recurrent unit GRU, and the function expression of the gated recurrent unit GRU is: ， In the above formula, and Respectively Step and The hidden state of the step, For the Step update door, For the The candidate hidden state of the step, Represents element-wise multiplication.

7. The method for tracing the source of broadband oscillation of a power grid based on a modal dynamic modulation graph network according to claim 1, characterized in that: In step S6, when a classifier is used to realize oscillation tracing according to the node features of each node finally output by the graph network, the classifier used is a multi-layer perceptron, and the multi-layer perceptron introduces a class weight parameter shown in the following formula so that the weight of the positive sample is inversely proportional to the number of the positive sample to balance the training loss: ， In the above formula, is the class weight parameter, is the number of positive samples.

8. A power grid broadband oscillation source tracing system based on a modal dynamic modulation graph network, comprising a microprocessor and a memory connected to each other, characterized in that: The microprocessor is programmed or configured to execute the method for tracing the source of broadband oscillation of a power grid based on a modal dynamic modulation diagram network as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program or instruction stored therein, characterized in that: The computer program or instruction is programmed or configured to execute the method for tracing the source of broadband oscillation of a power grid based on a modal dynamic modulation diagram network as described in any one of claims 1 to 7 through a processor.

10. A computer program product comprising a computer program or instructions, characterized in that The computer program or instruction is programmed or configured to execute the method for tracing the source of broadband oscillation of a power grid based on a modal dynamic modulation diagram network as described in any one of claims 1 to 7 through a processor.

Images