Power system low-frequency oscillation prediction method, device, equipment, medium and product

Abstract

The invention provides a power system low-frequency oscillation prediction method, device, equipment, medium and product, and the method comprises the steps: obtaining the current electrical parameters of a power system, the current electrical parameters comprising at least one of the following items: the amplitude, frequency, peak-to-peak value, active power and phase angle of a current power supply signal; performing feature extraction on the current electrical parameters through a feature extraction module to obtain input features; through a space and channel reconstruction convolution module, space channel redundancy and time channel redundancy of input features are suppressed, and data features are obtained; performing feature enhancement on the data features through a two-dimensional global average pooling module and an extrusion incentive attention mechanism module to obtain enhanced data features; and inputting the enhanced data features into a long short-term memory network layer to capture dynamic time sequence characteristics, and outputting a low-frequency oscillation prediction result through a full connection layer. According to the scheme, the accuracy of low-frequency oscillation prediction of the power system can be improved.

Description

Technical Field

[0001] This invention relates to the field of power control technology, and in particular to a method, apparatus, equipment, medium, and product for predicting low-frequency oscillations in power systems. Background Technology

[0002] Low-frequency oscillations in power systems refer to oscillations with frequencies below 2 Hz, primarily manifested as power angle oscillations and voltage oscillations. These low-frequency oscillations significantly impact the stability and reliability of power systems, making their prediction and analysis crucial.

[0003] In practice, low-frequency oscillations can lead to problems such as voltage instability, degraded power quality, equipment damage, and power system collapse. Therefore, by predicting low-frequency oscillations, early warnings can be issued and corresponding measures can be taken to ensure the reliable operation of the power system.

[0004] In some exemplary technologies, the prediction of low-frequency oscillations in a novel power system is achieved by forecasting a comprehensive index of low-frequency oscillations, thus enabling effective early warning of such oscillations. However, existing technologies suffer from three main problems: inability to adaptively adjust parameters, redundancy in temporal and spatial channels, and low feature correlation.

[0005] Therefore, current predictions of low-frequency oscillations in power systems suffer from low accuracy. Summary of the Invention

[0006] This invention provides a method, apparatus, equipment, medium, and product for predicting low-frequency oscillations in power systems, in order to solve the problem of low prediction accuracy in existing low-frequency oscillation prediction technologies.

[0007] This invention provides a method for predicting low-frequency oscillations in power systems, comprising the following steps.

[0008] Obtain the current electrical parameters of the power system, wherein the current electrical parameters include at least one of the following: the amplitude of the current power supply signal, the frequency of the current power supply signal, the peak-to-peak value of the current power supply signal, the active power of the current power supply signal, and the current phase angle; The current electrical parameters are input into the low-frequency oscillation prediction model so that the low-frequency oscillation prediction model processes the current electrical parameters and obtains the low-frequency oscillation prediction result output by the low-frequency oscillation prediction model. The low-frequency oscillation prediction model processes the current electrical parameters to obtain the low-frequency oscillation prediction result output by the low-frequency oscillation prediction model, including: The current electrical parameters are extracted using the feature extraction module to obtain the input features; By reconstructing the convolutional module using spatial and channel features, spatial and temporal channel redundancy of the input features is suppressed to obtain data features. The data features are enhanced by using a two-dimensional global average pooling module and a squeezed incentive attention mechanism module to obtain enhanced data features. The enhanced data features are input into a long short-term memory network layer to capture dynamic temporal characteristics, and the low-frequency oscillation prediction results are output through a fully connected layer.

[0009] According to a method for predicting low-frequency oscillations in a power system provided by the present invention, before obtaining the current electrical parameters of the power system, the method further includes: Construct an initial low-frequency oscillation prediction model; Multiple historical electrical parameters of the power system are acquired, and the low-frequency oscillation reference results corresponding to each historical electrical parameter are marked. Each historical electrical parameter includes the amplitude of the historical power supply signal, the frequency of the historical power supply signal, the peak-to-peak value of the historical power supply signal, the active power of the historical power supply signal, and the phase angle of the historical power supply signal. A training set is constructed based on the aforementioned historical electrical parameters and the corresponding low-frequency oscillation reference results for each historical electrical parameter; The initial low-frequency oscillation prediction model is iteratively trained based on the training set. During the iterative training process, the model parameters of the current low-frequency oscillation prediction model are adjusted using a parameter adaptive adjustment method until the current low-frequency oscillation prediction model meets the requirements, thereby obtaining the low-frequency oscillation prediction model.

[0010] According to the present invention, a method for predicting low-frequency oscillations in a power system includes adjusting the model parameters of the current low-frequency oscillation prediction model during iterative training using a parameter adaptive adjustment method, comprising: Set the initial model parameters and create the initial population; Calculate the objective function value for each individual in the current population, and find the optimal individual from the previous iteration based on the objective function value of each individual; each individual in the population represents a set of parameter configurations. Calculate the deviation between the current individual and the optimal individual of the previous iteration to obtain the current system deviation; Based on the current system deviation, the incremental output value is calculated using the proportional-integral-differential algorithm, and the zero output value is also calculated. Update each individual in the population based on the incremental output value and the zero output value; Determine whether the termination condition is met. If the termination condition is met, output the adaptively adjusted model parameters; otherwise, return to the step of calculating the objective function value of each individual in the current population. The termination condition includes: reaching the maximum number of iterations or the objective function value converging.

[0011] According to the present invention, a method for predicting low-frequency oscillations in power systems is provided, wherein the model parameters include at least one of the following: learning rate, regularization coefficient, channel splitting ratio, number of channels in grouped convolution, number of hidden layer nodes in long short-term memory network layers, and channel squeezing ratio of the squeeze-incentive attention mechanism.

[0012] According to the present invention, a method for predicting low-frequency oscillations in a power system includes, wherein marking the low-frequency oscillation reference result corresponding to each historical electrical parameter includes: A density-based spatial clustering algorithm with noise is applied to cluster historical active power data from multiple historical electrical data sets to obtain multiple historical electrical data sets that do not exhibit low-frequency oscillations and multiple historical electrical data sets that do exhibit low-frequency oscillations. The low-frequency oscillation reference result corresponding to each historical electrical data point that exhibits low-frequency oscillation is marked as having occurred, and the low-frequency oscillation reference result corresponding to each historical electrical data point that does not exhibit low-frequency oscillation is marked as not having occurred.

[0013] According to the present invention, a method for predicting low-frequency oscillations in a power system, the step of obtaining the current electrical parameters of the power system specifically includes: The data measured by the phasor measurement unit is acquired, and the current electrical parameters are calculated based on the data measured by the phasor measurement unit.

[0014] The present invention also provides a low-frequency oscillation prediction device for power systems, comprising the following modules: The acquisition module is used to acquire the current electrical parameters of the power system, wherein the current electrical parameters include at least one of the following: the amplitude of the current power supply signal, the frequency of the current power supply signal, the peak-to-peak value of the current power supply signal, the active power of the current power supply signal, and the phase angle of the current power supply signal; The prediction module is used to input the current electrical parameters into the low-frequency oscillation prediction model, so that the low-frequency oscillation prediction model processes the current electrical parameters and obtains the low-frequency oscillation prediction result output by the low-frequency oscillation prediction model; The low-frequency oscillation prediction model processes the current electrical parameters to obtain the low-frequency oscillation prediction result output by the low-frequency oscillation prediction model, including: The current electrical parameters are extracted using the feature extraction module to obtain the input features; By reconstructing the convolutional module using spatial and channel features, spatial and temporal channel redundancy of the input features is suppressed to obtain data features. The data features are enhanced by using a two-dimensional global average pooling module and a squeezed incentive attention mechanism module to obtain enhanced data features. The enhanced data features are input into a long short-term memory network layer to capture dynamic temporal characteristics, and the low-frequency oscillation prediction results are output through a fully connected layer.

[0015] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the power system low-frequency oscillation prediction method as described above.

[0016] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the power system low-frequency oscillation prediction method as described above.

[0017] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the power system low-frequency oscillation prediction method as described above.

[0018] The present invention provides a method, apparatus, equipment, medium, and product for predicting low-frequency oscillations in power systems. By acquiring the current electrical parameters of the power system, it can comprehensively reflect the current state of the power system. The acquisition of electrical parameters provides a rich data foundation for subsequent feature extraction and low-frequency oscillation prediction. Furthermore, the acquired current electrical parameters are input into a low-frequency oscillation prediction model for processing and prediction. Pre-trained models are used for prediction to ensure the accuracy and reliability of the prediction results. Specifically, a feature extraction module extracts features from the current electrical parameters to obtain input features, which reduces the amount of data processed subsequently, helps improve the model's prediction performance, and increases the model's processing efficiency. A spatial and channel reconstruction convolution module suppresses spatial and temporal channel redundancy of the input features, obtains data features, reduces redundant information, lowers the computational burden of the model, improves the model's operating efficiency, and, after reducing redundant information, the features have higher discriminative power, which helps the model learn and predict better. A two-dimensional global average pooling module can capture global information of the feature map, enhancing the global representativeness of the features. A squeeze-incentive attention mechanism module can highlight important features and suppress unimportant features, further enhancing the discriminative power of the features. The enhanced features better reflect the key information of the data, improving the model's predictive performance. Based on the above steps, the enhanced data features are input into a Long Short-Term Memory (LSTM) network layer to capture dynamic temporal characteristics, and then output accurate low-frequency oscillation prediction results through a fully connected layer. In summary, the solution of this invention improves the accuracy of low-frequency oscillation prediction in power systems, thereby enhancing the safety and reliability of the power system. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is one of the flowcharts of the low-frequency oscillation prediction method for power systems provided by the present invention.

[0021] Figure 2 This is a schematic diagram of the low-frequency oscillation prediction model provided by the present invention.

[0022] Figure 3 This is a schematic diagram of the SCNN structure provided by the present invention.

[0023] Figure 4 This is the second flowchart of the low-frequency oscillation prediction method for power systems provided by the present invention.

[0024] Figure 5 This is a flowchart illustrating the DPSCLA model provided by the present invention.

[0025] Figure 6 This is a schematic diagram of the PSA process provided by the present invention.

[0026] Figure 7 This is a schematic diagram of the structure of the low-frequency oscillation prediction device for power systems provided by the present invention.

[0027] Figure 8 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0029] It should be noted that the brief descriptions of terms in this application are only for the convenience of understanding the embodiments described below, and are not intended to limit the embodiments of this application. Unless otherwise stated, these terms should be understood in their ordinary and common meaning.

[0030] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar or related objects or entities and do not necessarily imply a specific order or sequence, unless otherwise indicated. It should be understood that such terms can be used interchangeably where appropriate, for example, in situations where implementation can proceed in an order other than those given in the embodiments illustrated or described in this application.

[0031] Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover but not exclude inclusion. For example, a product or device that includes a series of components is not necessarily limited to those explicitly listed, but may include other components not explicitly listed or inherent to such product or device. As used in this application, the term "module" means any known or subsequently developed hardware, software, firmware, artificial intelligence, fuzzy logic, or combination of hardware and / or software code capable of performing the functions associated with that element.

[0032] The technical solution of this application and how it solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The following is a combination of... Figures 1-6 The present invention describes a method for predicting low-frequency oscillations in power systems.

[0033] Figure 1 This is one of the flowcharts illustrating the low-frequency oscillation prediction method for power systems provided by this invention, such as... Figure 1 As shown, the method includes the following: Step 101: Obtain the current electrical parameters of the power system. The current electrical parameters include at least one of the following: the amplitude of the current power supply signal, the frequency of the current power supply signal, the peak-to-peak value of the current power supply signal, the active power of the current power supply signal, and the phase angle of the current power supply signal. Step 102: Input the current electrical parameters into the low-frequency oscillation prediction model; Step 103: Extract features from the current electrical parameters using the feature extraction module to obtain the input features; Step 104: Reconstruct the convolution module using spatial and channel methods to suppress spatial and temporal channel redundancy of the input features and obtain data features; Step 105: Enhance the data features by using the two-dimensional global average pooling module and the squeezed incentive attention mechanism module to obtain the enhanced data features; Step 106: Input the enhanced data features into the long short-term memory network layer to capture dynamic temporal characteristics, and output the low-frequency oscillation prediction results through the fully connected layer.

[0034] In practical applications, the main implementer of this power system low-frequency oscillation prediction method can be a power system low-frequency oscillation prediction device. There are various ways to implement such a device. For example, it can be implemented through computer programs, such as application software; or, for example, chips. It can also be implemented as a medium storing the relevant computer program, such as a USB flash drive or cloud storage; or, it can be implemented through a physical device that integrates or installs the relevant computer program, such as a server or smart device.

[0035] Specifically, step 101 includes: obtaining the current electrical parameters of the power system, which include at least one of the following: the amplitude of the current power supply signal, the frequency of the current power supply signal, the peak-to-peak value of the current power supply signal, the active power of the current power supply signal, and the phase angle of the current power supply signal.

[0036] Among them, the power supply signal refers to the signal output by the power system to the load. The power supply signal can be a voltage signal or a current signal.

[0037] Specifically, the amplitude of a power supply signal refers to its maximum or peak value. The amplitude of the power supply signal reflects the voltage or current intensity of the power system and is an important indicator for assessing the operating status of the power system.

[0038] Specifically, the frequency of a power supply signal refers to the number of times the power supply signal completes a periodic change per unit time, usually measured in Hertz (Hz). Frequency reflects the stability and synchronization of a power system; deviations in frequency may indicate system faults or instability.

[0039] Specifically, the peak-to-peak value of a power supply signal refers to the difference between the maximum and minimum values ​​of the power supply signal. The peak-to-peak value reflects the dynamic range of the signal and can be used to detect transient events in a power system (such as short circuits or lightning strikes).

[0040] Specifically, the active power of a power supply signal refers to the actual power transmitted by the power supply signal, usually measured in watts (W). Active power reflects the load condition and energy transmission efficiency of the power system and is an important indicator for assessing the supply and demand balance of the power system.

[0041] Specifically, the phase angle of a power supply signal refers to the phase angle of the power supply signal, usually measured in degrees (°) or radians (rad). The phase angle reflects the phase information of the power supply signal and is of great significance for the synchronization and stability analysis of power systems.

[0042] In this embodiment, the method for obtaining the current electrical parameters of the power system is not specifically limited. In one example, data acquisition devices, sensor networks, etc., can be set up in the power system to collect the current electrical parameters of the power system in real time, and transmit the collected current electrical parameters to the power system low-frequency oscillation prediction device.

[0043] Optionally, in one example, step 101 above specifically includes: The system acquires data measured by the phasor measurement unit and calculates the current electrical parameters based on the data measured by the phasor measurement unit. The current electrical parameters include at least one of the following: the amplitude of the current power supply signal, the frequency of the current power supply signal, the peak-to-peak value of the current power supply signal, the active power of the current power supply signal, and the phase angle of the current power supply signal.

[0044] A phasor measurement unit (PMU) is a high-precision measurement device capable of acquiring electrical parameters of a power system in real time. In one example, PMU data includes voltage phasors, current phasors, the frequency of the power supply signal, and the active and reactive power of the power supply signal. In practical applications, power system low-frequency oscillation prediction devices acquire PMU data and calculate the current electrical parameters based on this data.

[0045] It is understandable that by obtaining the current electrical parameters of the power system, the current state of the power system can be fully reflected, and the acquisition of electrical parameters provides a rich data foundation for subsequent feature extraction and low-frequency oscillation prediction.

[0046] Further, step 102 includes: inputting the current electrical parameters into the low-frequency oscillation prediction model.

[0047] The low-frequency oscillation prediction model is a pre-built and trained model. It includes a feature extraction module, a spatial and channel reconstruction convolution (ScConv) module, a two-dimensional global average pooling module, a squeeze-and-excitation attention (SE Attention) module, a long short-term memory (LSTM) network layer, and a fully connected layer.

[0048] It is understandable that the current electrical parameters are input into the low-frequency oscillation prediction model so that the model can process the current electrical parameters and obtain accurate low-frequency oscillation prediction results.

[0049] In practical applications, the low-frequency oscillation prediction model processes the current electrical parameters to obtain accurate low-frequency oscillation prediction results output by the low-frequency oscillation prediction model, including steps 103 to 106.

[0050] Figure 2 This is a schematic diagram of the low-frequency oscillation prediction model provided by the present invention, as shown below. Figure 2As shown, the low-frequency oscillation prediction model includes: a Spatial Convolutional Neural Network (SCNN) module, a two-dimensional global average pooling module, a squeeze-excited attention mechanism (SE Attention) module, a long short-term memory network layer (LSTM), and a fully connected layer. Among them, the Spatial Convolutional Neural Network (SCNN) includes a feature extraction module and a spatial and channel reconstruction convolution (ScConv) module.

[0051] Figure 3 This is a schematic diagram of the SCNN structure provided by the present invention, as shown below. Figure 3 As shown, the Spatial Convolutional Neural Network (SCNN) includes a feature extraction module and a spatial and channel reconstruction convolution (ScConv) module.

[0052] The following is combined Figure 2 and Figure 3 Steps 103 to 106 will be explained.

[0053] Specifically, step 103 includes: extracting features from the current electrical parameters using a feature extraction module to obtain input features.

[0054] Understandably, by using the feature extraction module to extract features from the current electrical parameters and obtain input features, the amount of data to be processed in the subsequent stages can be reduced, which helps to improve the model's predictive performance and processing efficiency.

[0055] In one example, such as Figure 2 As shown, the feature extraction module includes a sequence folding unit (seqfold), a first convolutional unit (Conv1), a corrected linear unit (ReLU), and a second convolutional unit (Conv2).

[0056] Specifically, the current electrical parameters are processed sequentially through a sequence folding unit, a first convolutional unit, a corrected linear unit, and a second convolutional unit to obtain the input feature X.

[0057] The sequence folding unit performs sequence folding processing on the current electrical parameters to ensure that the data channels are aligned for subsequent operations such as convolution.

[0058] For ReLU, which is the activation function, its formula is: The ReLU formula means that if the input x is less than 0, the output is 0; if the input x is greater than or equal to 0, the output is equal to the input value x. It is a commonly used activation function in neural networks, which can introduce nonlinearity, alleviate the gradient vanishing problem, and speed up training.

[0059] In this embodiment, the convolution operation uses discrete convolution, and its formula is as follows: When performing convolution operations on images or data matrices f It can be viewed as an input data matrix. It is a convolution kernel (small matrix). h This is the output after convolution. The data used in this paper's method is two-dimensional data. Assume the input data is a W×H matrix (W represents width, H represents height), and the convolution kernel is a k×k matrix (in this paper, the kernel is 3×1; this is to reduce overfitting through parameter sharing, capture local features, and integrate multivariate information, while also being computationally efficient, flexible, and scalable). Then, for each element in the output feature map... h ( i , j The calculation formula is as follows: in,( i , j ) are the coordinates in the output feature map, ( m , n ) are the coordinates in the convolution kernel.

[0060] Further, step 104 includes: suppressing spatial and temporal channel redundancy of the input features by reconstructing the convolution module through spatial and channel features to obtain data features.

[0061] like Figure 3 As shown, the Spatial and Channel Reconstruction Convolution (ScConv) module includes Channel Reconstruction Unit (CRU) units and Spatial Reconstruction Unit (SRU) units.

[0062] Combination Figure 3 Spatial reconstruction is performed using SRU units. Let N be the batch size, C be the number of channels, and H and W be the height and width of the spatial dimension, respectively. For the input features... Standardization is performed using a Group Normalization (GN) layer, as shown in the following formula: in, and The input features are the mean and standard deviation of X. Small positive numbers were added to ensure the stability of division. and These are the trainable affine transformation parameters.

[0063] Combination Figure 3 Furthermore, utilizing Obtain normalized correlation weights Normalized related weights The formula is as follows: Among them, the normalized relevant weights Reflecting the importance of different feature maps Represents vector weights The i One element, The trainable affine transformation parameters obtained after passing through the GN layer The first in The element, its reaction The scaling degree of each channel feature, and its value, also reflects the importance or variance of the channel feature to some extent. A larger variance means that the channel may contain more information. The larger the value.

[0064] Combination Figure 3 Furthermore, the sigmoid function is used to normalize the relevant weights. Mapping is performed, and gating is applied based on a preset threshold. If the value exceeds the preset threshold, the weight is set to 1 and recorded as the information weight. Conversely, a weight of 0 is set as a non-information weight. The formula is as follows: Furthermore, input features X respectively with and Multiplication separates information-rich features. and the characteristic of less information , This represents element-wise multiplication. and The formula is as follows: Furthermore, cross-reconstruction is performed on the separated features, where This represents element-wise addition, and the formula is as follows: Furthermore, regarding the cross-reconstruction features and To connect, use the following formula: The spatial refinement feature is obtained through the above formula. In the formula The above operations represent connection operations, which suppress spatial redundancy and enhance feature representation.

[0065] Combination Figure 3 Subsequently, a channel reconstruction operation is performed using a CRU unit. The spatial refinement features output by the SRU operation are then processed. proportionally (Example) Divide the sample into two parts (selected as 0.5), and then use 1. 1 convolution with compression ratio (Example) Select channel 2) for compression, and obtain and . Entering the up-conversion stage, for Using group-wise convolution (GWC) and point-wise convolution (PWC) operations reduces computation and extracts high-level features.

[0066] The GWC operation formula is as follows: in, For the first 1 input feature vector For the first Input features Weights in a globally weighted combination For the first Input features The relevant scores, where exp is an exponential function.

[0067] The PwC operation formula is as follows: in, Input features and Pair-weighted combination weights between To match the features The relevant scores, Used to measure the importance between features This is the difference vector between features.

[0068] Combination Figure 3 Furthermore, the input features after GWC and PWC operations are... The output feature map is obtained by adding them together. Y 1. The formula is as follows: in, , These are the learnable weight matrices for GWC and PWC, respectively.

[0069] Combination Figure 3 , Proceeding to the next transformation stage. (Simplified 1) 1 PWC The operation yields a feature map with shallow details, and it is reused. The output feature map is obtained by concatenating and reusing features. Y 2. The formula is as follows: in, The learnable weight matrix for PwC. The same as above indicates a join operation.

[0070] Furthermore, on Y 1 and Y 2. Perform global average pooling to obtain S 1 and S 2, of which m =1, 2, the global average optimization formula is as follows: Furthermore, stacking S 1 and S 2. The feature importance vector is obtained through channel soft attention operation. The expression for the feature importance vector is as follows: Combination Figure 3 ,merge Y 1 and Y 2. Through adaptive fusion to reduce channel redundancy, data feature Y is obtained. The expression for data feature Y is as follows: Further, step 105 includes: performing feature enhancement on the data features through a two-dimensional global average pooling module and a squeeze-incentive attention mechanism module to obtain enhanced data features.

[0071] Specifically, for two-dimensional input data features (Where H is the height, W is the width, and C is the number of channels), the output of two-dimensional global average pooling. (c = 1,…,C) The calculation formula is as follows: in, The input is a two-dimensional feature map, which is a three-dimensional tensor with dimensions corresponding to channels, height, and width, respectively.H The dimension of the input feature map in the height direction is the number of rows in the feature map. W The width dimension represents the size of the input feature map, i.e., the number of columns in the feature map. C represents the number of channels in the input feature map; for example, C=1. Represents the input feature map The element value at position c, height h, and width w. The output is a vector of length C, where each element... y c It is the average value of all elements in the corresponding channel, that is, the value obtained after performing average pooling operation on all elements of each channel of the input feature map.

[0072] In this embodiment, the Squeeze-Excitation Attention (SE Attention) mechanism is integrated into the CNN-LSTM hybrid model to enhance the correlation between features and further optimize prediction accuracy. Specifically, the Squeeze-Excitation Attention module consists of three parts: Squeeze, Excitation, and Scale. Squeeze refers to compressing the spatial information of the input feature map into a vector through global average pooling. This operation captures global information and reduces the spatial dimension of the feature map. Excitation involves using at least one fully connected layer to learn the dependencies between channels and generate weights for each channel. Scale refers to applying the generated weights to the original feature map to complete the feature recalibration.

[0073] Understandably, the two-dimensional global average pooling module can capture global information from the feature map, enhancing the global representativeness of the features. The squeeze-incentive attention mechanism module can highlight important features and suppress unimportant features, further enhancing the discriminative power of the features. The enhanced features can better reflect the key information of the data, improving the model's predictive performance. Therefore, performing two-dimensional global average pooling on the data features output by the SCNN module to capture the overall features and enhance their representativeness before feeding them into the SE Attention module can further enhance the correlation between features.

[0074] Further, step 106 includes: inputting the enhanced data features into a long short-term memory network layer to capture dynamic temporal characteristics, and outputting low-frequency oscillation prediction results through a fully connected layer.

[0075] Understandably, based on the above steps, the enhanced data features are input into the Long Short-Term Memory network layer to capture dynamic temporal characteristics, and the accurate low-frequency oscillation prediction results are output through the fully connected layer.

[0076] In practical applications, a low-frequency oscillation prediction model needs to be pre-built and trained. Specifically, in one possible implementation, Figure 4 This is the second flowchart of the low-frequency oscillation prediction method for power systems provided by the present invention, as shown below. Figure 2 As shown, based on the above embodiments, before step 101, the above power system low-frequency oscillation prediction method further includes: Step 401: Construct an initial low-frequency oscillation prediction model; Step 402: Obtain multiple historical electrical parameters of the power system and mark the low-frequency oscillation reference results corresponding to each historical electrical parameter; wherein, each historical electrical parameter includes the amplitude of the historical power supply signal, the frequency of the historical power supply signal, the peak-to-peak value of the historical power supply signal, the active power of the historical power supply signal, and the phase angle of the historical power supply signal; Step 403: Construct a training set based on multiple historical electrical parameters and the low-frequency oscillation reference results corresponding to each historical electrical parameter; Step 404: Iteratively train the initial low-frequency oscillation prediction model based on the training set, and adjust the model parameters of the current low-frequency oscillation prediction model through the parameter adaptive adjustment method during the iterative training process until the current low-frequency oscillation prediction model meets the requirements, thereby obtaining the low-frequency oscillation prediction model.

[0077] The initial low-frequency oscillation prediction model refers to the initially established, untrained model. In this invention, the initial low-frequency oscillation prediction model is constructed based on the DPSCLA (DBSCAN-PSA-SCNN-LSTM-SE Attention) model.

[0078] Figure 5 This is a flowchart illustrating the DPSCLA model provided by the present invention, as shown below. Figure 5 As shown, during the training phase, multiple historical electrical parameters of the power system are acquired, and the low-frequency oscillation reference results corresponding to each historical electrical parameter are labeled using the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm.

[0079] Optionally, in one possible implementation, the low-frequency oscillation reference result corresponding to each historical electrical parameter in step 402 above includes: A density-based spatial clustering algorithm with noise is applied to cluster historical active power data from multiple historical electrical data sets to obtain multiple historical electrical data sets that do not exhibit low-frequency oscillations and multiple historical electrical data sets that do exhibit low-frequency oscillations. The low-frequency oscillation reference result corresponding to each historical electrical data point that exhibits low-frequency oscillation is marked as having occurred, and the low-frequency oscillation reference result corresponding to each historical electrical data point that does not exhibit low-frequency oscillation is marked as not having occurred.

[0080] In practical applications, multiple historical electrical data points are acquired within a predetermined historical time period. For example, historical PMU data within the predetermined time period is acquired, and multiple historical electrical data points are calculated from the historical PMU data. Each historical electrical parameter includes the amplitude of the historical power supply signal, the frequency of the historical power supply signal, the peak-to-peak value of the historical power supply signal, the active power of the historical power supply signal, and the phase angle of the historical power supply signal.

[0081] A density-based spatial clustering algorithm with noise (DBSCAN) is applied to cluster the active power of historical power supply signals from multiple historical electrical parameters. Historical electrical parameters without low-frequency oscillations are marked as 1, and those with low-frequency oscillations are marked as 0. Based on the range of active power in the historical electrical parameters marked as 1, the power interval corresponding to the low-frequency oscillation event is determined. Conversely, based on the range of active power in the historical electrical parameters marked as 0, the power interval corresponding to the event without low-frequency oscillations is determined.

[0082] For each historical electrical parameter, if the active power of the historical power supply signal in that historical electrical parameter is within the power range corresponding to the low-frequency oscillation event, then the low-frequency oscillation reference result corresponding to that historical electrical parameter is marked as having experienced low-frequency oscillation; otherwise, the low-frequency oscillation reference result corresponding to that historical electrical parameter is marked as not having experienced low-frequency oscillation.

[0083] Furthermore, a training set is constructed based on multiple historical electrical parameters and the corresponding low-frequency oscillation reference results for each historical electrical parameter.

[0084] It is understandable that the low-frequency oscillation reference result corresponding to each historical electrical parameter refers to whether low-frequency oscillations occur in the power system under the conditions of that historical electrical parameter. Each training data point in the training set includes a historical electrical parameter and the corresponding low-frequency oscillation reference result.

[0085] Combination Figure 5 Furthermore, the initial low-frequency oscillation prediction model is iteratively trained based on the training set. During the iterative training process, the model parameters of the current low-frequency oscillation prediction model are adjusted through the parameter self-adjustment (PSA) module until the current low-frequency oscillation prediction model meets the requirements, thus obtaining the low-frequency oscillation prediction model.

[0086] Specifically, in one example, the model parameters include at least one of the following: learning rate, regularization coefficient, channel splitting ratio, number of channels in grouped convolutions, number of hidden nodes in the long short-term memory network layer, and channel squeezing ratio of the squeeze-incentive attention mechanism.

[0087] Specifically, in one implementation, step 404 above adjusts the model parameters of the current low-frequency oscillation prediction model during iterative training using a parameter adaptive adjustment method, including: Set the initial model parameters and create the initial population; Calculate the objective function value for each individual in the current population, and find the optimal individual from the previous iteration based on the objective function value of each individual; each individual in the population represents a set of parameter configurations. Calculate the deviation between the current individual and the best individual in the previous iteration to obtain the current system deviation; Based on the current system deviation, the incremental output value is calculated using the proportional-integral-differential algorithm, and the zero output value is also calculated. Update each individual in the population based on the incremental output value and the zero output value; Determine if the termination condition is met. If the termination condition is met, output the adaptively adjusted model parameters; otherwise, return to the step of calculating the objective function value of each individual in the current population. The termination condition includes: reaching the maximum number of iterations or the objective function value converging.

[0088] In practical applications, the PSA method determines the system deviation by calculating the objective function value between an individual and the historical best individual. Then, it comprehensively processes the deviation based on the proportional, integral, and derivative terms in the Proportional-Integral-Derivative (PID) algorithm, combined with a mechanism to prevent local optima, to determine the optimization direction. This method effectively solves the problem of parameters not being able to adaptively adjust; therefore, this invention uses the PSA algorithm to address the difficulty of parameters not being able to adaptively adjust. PSA employs a mechanism based on incremental PID algorithm to converge the solution of the optimization problem to the optimal state. The idea is to treat each individual in the population as a state in the system, guiding the population towards the optimal solution by continuously adjusting the deviation between each individual and the best individual. For a given optimization problem, including a set of decision variables, constraints, and an objective function, the PSA algorithm performs iterative optimization in the search space. Its core calculation process is defined as follows.

[0089] Figure 6 This is a schematic diagram of the PSA process provided by the present invention, as shown below. Figure 6 As shown, the initial model parameters are set, and the initial population is created.

[0090] In the initial stage, the initial population is generated randomly. The formula is as follows: Among them, the decision variables of the problem are The upper and lower bounds of the variable are and The maximum number of iterations for the control parameters of the PSA algorithm is T, and the population size is n. In the formula... It is a random number between 0 and 1.

[0091] Furthermore, the objective function value of each individual in the current population is calculated, and based on the objective function value of each individual, the optimal individual from the previous iteration is found. Each individual in the population represents a set of parameter configurations.

[0092] During the iteration process, the current individual is calculated. Compared with the best individual in the previous iteration The deviation between them is used to obtain the current system deviation. The formula is as follows: in, It is the best individual from the previous iteration. It refers to the current individual.

[0093] Furthermore, based on the current system bias The incremental output value is calculated using the proportional-integral-differential algorithm. And calculate the zero output value. .

[0094] Specifically, the proportional coefficient can be calculated based on the PID control formula. Integral coefficient and differential coefficients The PID control formulas for these three output adjustment values ​​are as follows: At the same time, a zero-output formula is introduced to prevent the algorithm from getting trapped in local optima. The zero-output formula is as follows: in, To adjust the coefficient, n rows Column random number matrix, Let Levy's flight function be defined in the Levy flight function formula. and For n rows that each follow a standard normal distribution Column random number matrix, The factor is set to 1.5 in this paper. This is the gamma function. The Levy flight function can accurately describe the characteristics of a stochastic process. Using the Levy flight function in the PSA algorithm allows the algorithm to occasionally make long jumps during the search process, which helps to escape local optima and enhances the algorithm's global search capability.

[0095] Finally, the algorithm updates the population size according to the following formula, repeating this process until the termination condition is met. The update formula is as follows: in, It is an n-row, 1-column matrix of random numbers between 0 and 1. .

[0096] The low-frequency oscillation prediction method for power systems provided by this invention comprehensively reflects the current state of the power system by acquiring the current electrical parameters. The acquisition of these electrical parameters provides a rich data foundation for subsequent feature extraction and low-frequency oscillation prediction. Furthermore, the acquired current electrical parameters are input into the low-frequency oscillation prediction model for processing and prediction. Pre-trained models are then used for prediction to ensure the accuracy and reliability of the prediction results. Specifically, a feature extraction module extracts features from the current electrical parameters to obtain input features, reducing the amount of data processed subsequently and improving the model's prediction performance and processing efficiency. A spatial and channel reconstruction convolution module suppresses spatial and temporal channel redundancy in the input features, obtaining data features, reducing redundant information, lowering the computational burden on the model, improving its operating efficiency, and increasing the discriminative power between features after reducing redundant information, thus aiding in better model learning and prediction. A two-dimensional global average pooling module captures global information from the feature map, enhancing the global representativeness of the features. A squeeze-incentive attention mechanism module highlights important features and suppresses unimportant features, further enhancing the discriminative power of the features. The enhanced features better reflect the key information of the data, improving the model's predictive performance. Based on the above steps, the enhanced data features are input into a Long Short-Term Memory (LSTM) network layer to capture dynamic temporal characteristics, and then output accurate low-frequency oscillation prediction results through a fully connected layer. In summary, the solution of this invention improves the accuracy of low-frequency oscillation prediction in power systems, thereby enhancing the safety and reliability of the power system.

[0097] The low-frequency oscillation prediction device for power systems provided by the present invention is described below. The low-frequency oscillation prediction device for power systems described below can be referred to in correspondence with the low-frequency oscillation prediction method for power systems described above.

[0098] Figure 7 This is a schematic diagram of the low-frequency oscillation prediction device for power systems provided by the present invention, as shown below. Figure 7 As shown, the power system low-frequency oscillation prediction device includes: The acquisition module 71 is used to acquire the current electrical parameters of the power system. The current electrical parameters include at least one of the following: the amplitude of the current power supply signal, the frequency of the current power supply signal, the peak-to-peak value of the current power supply signal, the active power of the current power supply signal, and the phase angle of the current power supply signal. Prediction module 72 is used to input the current electrical parameters into the low-frequency oscillation prediction model so that the low-frequency oscillation prediction model processes the current electrical parameters and obtains the low-frequency oscillation prediction result output by the low-frequency oscillation prediction model. The low-frequency oscillation prediction model processes the current electrical parameters to obtain the low-frequency oscillation prediction results output by the model, including: The feature extraction module extracts features from the current electrical parameters to obtain the input features. By reconstructing the convolutional module spatially and temporally, spatial and temporal redundancy of the input features is suppressed, thereby obtaining data features. The data features are enhanced by using a two-dimensional global average pooling module and a squeezed incentive attention mechanism module to obtain the enhanced data features. The enhanced data features are input into the Long Short-Term Memory (LSTM) network layer to capture dynamic temporal characteristics, and the low-frequency oscillation prediction results are output through the fully connected layer.

[0099] Optionally, in one possible implementation, the acquisition module 71 described above is specifically used for: The system acquires data measured by the phasor measurement unit and calculates the current electrical parameters based on the data measured by the phasor measurement unit. The current electrical parameters include at least one of the following: the amplitude of the current power supply signal, the frequency of the current power supply signal, the peak-to-peak value of the current power supply signal, the active power of the current power supply signal, and the phase angle of the current power supply signal.

[0100] Optionally, in one possible implementation, the above-mentioned power system low-frequency oscillation prediction device further includes: a training module, the training module being used for: Construct an initial low-frequency oscillation prediction model; Multiple historical electrical parameters of the power system are acquired, and the low-frequency oscillation reference results corresponding to each historical electrical parameter are marked. Each historical electrical parameter includes the amplitude of the historical power supply signal, the frequency of the historical power supply signal, the peak-to-peak value of the historical power supply signal, the active power of the historical power supply signal, and the phase angle of the historical power supply signal. A training set is constructed based on multiple historical electrical parameters and the corresponding low-frequency oscillation reference results for each historical electrical parameter; The initial low-frequency oscillation prediction model is iteratively trained based on the training set. During the iterative training process, the model parameters of the current low-frequency oscillation prediction model are adjusted through an adaptive parameter adjustment method until the current low-frequency oscillation prediction model meets the requirements, thus obtaining the low-frequency oscillation prediction model.

[0101] Optionally, in one possible implementation, when the training module is used to adjust the model parameters of the current low-frequency oscillation prediction model through a parameter adaptive adjustment method during iterative training, it is specifically used for: Set the initial model parameters and create the initial population; Calculate the objective function value for each individual in the current population, and find the optimal individual from the previous iteration based on the objective function value of each individual; each individual in the population represents a set of parameter configurations. Calculate the deviation between the current individual and the best individual in the previous iteration to obtain the current system deviation; Based on the current system deviation, the incremental output value is calculated using the proportional-integral-differential algorithm, and the zero output value is also calculated. Update each individual in the population based on the incremental output value and the zero output value; Determine if the termination condition is met. If the termination condition is met, output the adaptively adjusted model parameters; otherwise, return to the step of calculating the objective function value of each individual in the current population. The termination condition includes: reaching the maximum number of iterations or the objective function value converging.

[0102] Optionally, in one possible implementation, the above model parameters include at least one of the following: learning rate, regularization coefficient, channel splitting ratio, number of channels in grouped convolution, number of hidden layer nodes in long short-term memory network layers, and channel squeezing ratio of the squeeze-incentive attention mechanism.

[0103] Optionally, in one possible implementation, when the training module is used to label the low-frequency oscillation reference result corresponding to each historical electrical parameter, it is specifically used for: A density-based spatial clustering algorithm with noise is applied to cluster historical active power data from multiple historical electrical data sets to obtain multiple historical electrical data sets that do not exhibit low-frequency oscillations and multiple historical electrical data sets that do exhibit low-frequency oscillations. The low-frequency oscillation reference result corresponding to each historical electrical data point that exhibits low-frequency oscillation is marked as having occurred, and the low-frequency oscillation reference result corresponding to each historical electrical data point that does not exhibit low-frequency oscillation is marked as not having occurred.

[0104] In the low-frequency oscillation prediction device for power systems provided by this invention, the acquisition module obtains the current electrical parameters of the power system, which comprehensively reflects the current state of the power system. The acquisition of these electrical parameters provides a rich data foundation for subsequent feature extraction and low-frequency oscillation prediction. Furthermore, the prediction module inputs the acquired current electrical parameters into the low-frequency oscillation prediction model for processing and prediction. The prediction is performed using a pre-trained model, ensuring the accuracy and reliability of the prediction results. Specifically, the feature extraction module extracts features from the current electrical parameters to obtain input features, reducing the amount of data processed subsequently, thus improving the model's prediction performance and processing efficiency. The spatial and channel reconstruction convolution module suppresses spatial and temporal channel redundancy of the input features, obtaining data features, reducing redundant information, lowering the computational burden on the model, improving its operating efficiency, and increasing the discriminative power between features after reducing redundant information, which helps the model learn and predict better. The two-dimensional global average pooling module captures global information from the feature map, enhancing the global representativeness of the features. The squeeze-incentive attention mechanism module highlights important features and suppresses unimportant features, further enhancing the discriminative power of the features. The enhanced features better reflect the key information of the data, improving the model's predictive performance. Based on the above steps, the enhanced data features are input into a Long Short-Term Memory (LSTM) network layer to capture dynamic temporal characteristics, and then output accurate low-frequency oscillation prediction results through a fully connected layer. In summary, the solution of this invention improves the accuracy of low-frequency oscillation prediction in power systems, thereby enhancing the safety and reliability of the power system.

[0105] Figure 8 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 8As shown, the electronic device may include: a processor 810, a communications interface 820, a memory 830, and a communications bus 840, wherein the processor 810, the communications interface 820, and the memory 830 communicate with each other through the communications bus 840. The processor 810 can call logic instructions in the memory 830 to execute a low-frequency oscillation prediction method for a power system. This method includes: acquiring current electrical parameters of the power system, including at least one of the following: amplitude of the current power supply signal, frequency of the current power supply signal, peak-to-peak value of the current power supply signal, active power of the current power supply signal, and phase angle of the current power supply signal; inputting the current electrical parameters into a low-frequency oscillation prediction model, so that the low-frequency oscillation prediction model processes the current electrical parameters to obtain a low-frequency oscillation prediction result output by the low-frequency oscillation prediction model; wherein, the processing of the current electrical parameters by the low-frequency oscillation prediction model to obtain the low-frequency oscillation prediction result output by the low-frequency oscillation prediction model includes: extracting features from the current electrical parameters through a feature extraction module to obtain input features; suppressing spatial and temporal redundancy of the input features through a spatial and channel reconstruction convolution module to obtain data features; enhancing the data features through a two-dimensional global average pooling module and a squeeze-excited attention mechanism module to obtain enhanced data features; inputting the enhanced data features into a long short-term memory network layer to capture dynamic temporal characteristics, and outputting the low-frequency oscillation prediction result through a fully connected layer.

[0106] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0107] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the power system low-frequency oscillation prediction method provided by the above methods. The method includes: acquiring the current electrical parameters of the power system, the current electrical parameters including at least one of the following: the amplitude of the current power supply signal, the frequency of the current power supply signal, the peak-to-peak value of the current power supply signal, the active power of the current power supply signal, and the phase angle of the current power supply signal; inputting the current electrical parameters into a low-frequency oscillation prediction model so that the low-frequency oscillation prediction model processes the current electrical parameters. The low-frequency oscillation prediction model outputs the prediction results of the low-frequency oscillation. This model processes current electrical parameters to obtain the predicted results, including: extracting features from the current electrical parameters using a feature extraction module to obtain input features; suppressing spatial and temporal redundancy of the input features using a spatial and channel reconstruction convolution module to obtain data features; enhancing the data features using a two-dimensional global average pooling module and a squeeze-excitation attention mechanism module to obtain enhanced data features; inputting the enhanced data features into a long short-term memory network layer to capture dynamic temporal characteristics; and outputting the low-frequency oscillation prediction results through a fully connected layer.

[0108] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the power system low-frequency oscillation prediction method provided by the methods described above. The method includes: acquiring current electrical parameters of the power system, the current electrical parameters including at least one of the following: amplitude of the current power supply signal, frequency of the current power supply signal, peak-to-peak value of the current power supply signal, active power of the current power supply signal, and phase angle of the current power supply signal; inputting the current electrical parameters into a low-frequency oscillation prediction model, so that the low-frequency oscillation prediction model processes the current electrical parameters to obtain the low-frequency oscillation prediction model output by the low-frequency oscillation prediction model. The oscillation prediction results are as follows: The low-frequency oscillation prediction model processes the current electrical parameters to obtain the low-frequency oscillation prediction results output by the low-frequency oscillation prediction model. This includes: extracting features from the current electrical parameters through a feature extraction module to obtain input features; suppressing spatial and temporal redundancy of the input features through a spatial and channel reconstruction convolution module to obtain data features; enhancing the data features through a two-dimensional global average pooling module and a squeeze-excited attention mechanism module to obtain enhanced data features; inputting the enhanced data features into a long short-term memory network layer to capture dynamic temporal characteristics, and outputting the low-frequency oscillation prediction results through a fully connected layer.

[0109] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0110] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0111] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting low-frequency oscillations in power systems, characterized in that, The method includes: Obtain the current electrical parameters of the power system, wherein the current electrical parameters include at least one of the following: the amplitude of the current power supply signal, the frequency of the current power supply signal, the peak-to-peak value of the current power supply signal, the active power of the current power supply signal, and the current phase angle; The current electrical parameters are input into the low-frequency oscillation prediction model so that the low-frequency oscillation prediction model processes the current electrical parameters and obtains the low-frequency oscillation prediction result output by the low-frequency oscillation prediction model. The low-frequency oscillation prediction model processes the current electrical parameters to obtain the low-frequency oscillation prediction result output by the low-frequency oscillation prediction model, including: The current electrical parameters are extracted using the feature extraction module to obtain the input features; By reconstructing the convolutional module using spatial and channel features, spatial and temporal channel redundancy of the input features is suppressed to obtain data features. The data features are enhanced by using a two-dimensional global average pooling module and a squeezed incentive attention mechanism module to obtain enhanced data features. The enhanced data features are input into a long short-term memory network layer to capture dynamic temporal characteristics, and the low-frequency oscillation prediction results are output through a fully connected layer.

2. The method for predicting low-frequency oscillations in power systems according to claim 1, characterized in that, Before obtaining the current electrical parameters of the power system, the method further includes: Construct an initial low-frequency oscillation prediction model; Multiple historical electrical parameters of the power system are acquired, and the low-frequency oscillation reference results corresponding to each historical electrical parameter are marked. Each historical electrical parameter includes the amplitude of the historical power supply signal, the frequency of the historical power supply signal, the peak-to-peak value of the historical power supply signal, the active power of the historical power supply signal, and the phase angle of the historical power supply signal. A training set is constructed based on the aforementioned historical electrical parameters and the corresponding low-frequency oscillation reference results for each historical electrical parameter; The initial low-frequency oscillation prediction model is iteratively trained based on the training set. During the iterative training process, the model parameters of the current low-frequency oscillation prediction model are adjusted using a parameter adaptive adjustment method until the current low-frequency oscillation prediction model meets the requirements, thereby obtaining the low-frequency oscillation prediction model.

3. The method for predicting low-frequency oscillations in power systems according to claim 2, characterized in that, The method of adjusting the model parameters of the current low-frequency oscillation prediction model through adaptive parameter adjustment during iterative training includes: Set the initial model parameters and create the initial population; Calculate the objective function value for each individual in the current population, and find the optimal individual from the previous iteration based on the objective function value of each individual; each individual in the population represents a set of parameter configurations. Calculate the deviation between the current individual and the optimal individual of the previous iteration to obtain the current system deviation; Based on the current system deviation, the incremental output value is calculated using the proportional-integral-differential algorithm, and the zero output value is also calculated. Update each individual in the population based on the incremental output value and the zero output value; Determine whether the termination condition is met. If the termination condition is met, output the adaptively adjusted model parameters; otherwise, return to the step of calculating the objective function value of each individual in the current population. The termination condition includes: reaching the maximum number of iterations or the objective function value converging.

4. The method for predicting low-frequency oscillations in power systems according to claim 2, characterized in that, The model parameters include at least one of the following: learning rate, regularization coefficient, channel splitting ratio, number of channels in grouped convolutions, number of hidden nodes in the long short-term memory network layer, and channel squeezing ratio of the squeeze-incentive attention mechanism.

5. The method for predicting low-frequency oscillations in power systems according to claim 2, characterized in that, The low-frequency oscillation reference results corresponding to each historical electrical parameter are marked, including: A density-based spatial clustering algorithm with noise is applied to cluster historical active power data from multiple historical electrical data sets to obtain multiple historical electrical data sets that do not exhibit low-frequency oscillations and multiple historical electrical data sets that do exhibit low-frequency oscillations. The low-frequency oscillation reference result corresponding to each historical electrical data point that exhibits low-frequency oscillation is marked as having occurred, and the low-frequency oscillation reference result corresponding to each historical electrical data point that does not exhibit low-frequency oscillation is marked as not having occurred.

6. The method for predicting low-frequency oscillations in power systems according to any one of claims 1-5, characterized in that, The acquisition of the current electrical parameters of the power system specifically includes: The data measured by the phasor measurement unit is acquired, and the current electrical parameters are calculated based on the data measured by the phasor measurement unit.

7. A low-frequency oscillation prediction device for power systems, characterized in that, The device includes: The acquisition module is used to acquire the current electrical parameters of the power system, wherein the current electrical parameters include at least one of the following: the amplitude of the current power supply signal, the frequency of the current power supply signal, the peak-to-peak value of the current power supply signal, the active power of the current power supply signal, and the phase angle of the current power supply signal; The prediction module is used to input the current electrical parameters into the low-frequency oscillation prediction model, so that the low-frequency oscillation prediction model processes the current electrical parameters and obtains the low-frequency oscillation prediction result output by the low-frequency oscillation prediction model; The low-frequency oscillation prediction model processes the current electrical parameters to obtain the low-frequency oscillation prediction result output by the low-frequency oscillation prediction model, including: The current electrical parameters are extracted using the feature extraction module to obtain the input features; By reconstructing the convolutional module using spatial and channel features, spatial and temporal channel redundancy of the input features is suppressed to obtain data features. The data features are enhanced by using a two-dimensional global average pooling module and a squeezed incentive attention mechanism module to obtain enhanced data features. The enhanced data features are input into a long short-term memory network layer to capture dynamic temporal characteristics, and the low-frequency oscillation prediction results are output through a fully connected layer.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the power system low-frequency oscillation prediction method as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the low-frequency oscillation prediction method for power systems as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the low-frequency oscillation prediction method for power systems as described in any one of claims 1 to 6.

Images