Wide-band oscillation parameter detection method and system for power system
Patent Information
- Application Number
- PCT/CN2025/086093
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-02-19
- Filing Date
- 2025-03-31
- Publication Date
- 2026-08-27
Smart Images
Figure CN2025086093_27082026_PF_FP_ABST
Abstract
Description
A method and system for detecting broadband oscillation parameters in power systems Technical Field
[0001] This invention relates to the field of power system technology, and in particular to a method and system for detecting broadband oscillation parameters in power systems. Background Technology
[0002] With the widespread integration of high-proportion renewable energy sources and power electronic devices into the power system, many devices and components in the power system, such as transformers, motors, rectifiers, and inverters, exhibit significant nonlinear characteristics. This causes the broadband oscillation signal of the power system to exhibit strong time-varying, strong nonlinear, and multimodal characteristics. Because multiple oscillation modes, such as low-frequency oscillations, subsynchronous oscillations, supersynchronous oscillations, medium-frequency oscillations, and high-frequency oscillations, may coexist and superimpose in the broadband oscillation signal, existing oscillation parameter detection methods cannot overcome the influence of mode aliasing. This leads to problems such as missing modes or false modes during oscillation parameter identification, thus preventing existing oscillation parameter detection methods from accurately and effectively detecting the broadband oscillation parameters of the power system. Summary of the Invention
[0003] This invention provides a method and system for detecting broadband oscillation parameters in power systems, thereby solving the technical problem that existing oscillation parameter detection methods cannot accurately and effectively detect broadband oscillation parameters in power systems.
[0004] To address the aforementioned technical problems, a first aspect of this invention provides a method for detecting broadband oscillation parameters in a power system, comprising:
[0005] Acquire the broadband oscillation signal to be detected;
[0006] A preset deep residual network is used to perform grouped convolution on the broadband oscillation signal to be detected, so as to obtain the oscillation mode features corresponding to the oscillation signals of different frequency bands in the broadband oscillation signal to be detected;
[0007] A preset temporal generative adversarial network is used to identify the oscillation parameters of the oscillation mode features, thereby obtaining the oscillation parameters corresponding to the oscillation signals in different frequency bands.
[0008] As a preferred embodiment, the step of using a preset deep residual network to perform grouped convolution on the broadband oscillation signal to be detected, and obtaining the oscillation mode features corresponding to the oscillation signals of different frequency bands in the broadband oscillation signal to be detected, specifically includes:
[0009] Based on the deep residual network, feature extraction is performed on the broadband oscillation signal to be detected to obtain initial signal features;
[0010] The initial signal features are grouped and convolved to obtain local signal features;
[0011] The local features of the signal are subjected to average pooling to obtain the oscillation mode features corresponding to the oscillation signal in different frequency bands.
[0012] As a preferred embodiment, the step of extracting features from the broadband oscillation signal to be detected based on the deep residual network to obtain initial signal features specifically includes:
[0013] Convolutional layers and max pooling layers are used to extract features and reduce the dimensionality of the broadband oscillation signal to be detected, respectively, to obtain the initial signal features.
[0014] As a preferred embodiment, the step of performing grouped convolution on the initial signal features to obtain local signal features specifically includes:
[0015] The initial signal features are grouped to obtain several grouped feature maps;
[0016] Based on several parallel residual convolution paths in the deep residual network, residual convolution is performed on several grouped feature maps to obtain the local features corresponding to each grouped feature map; wherein, the residual convolution path is formed by stacking several residual modules;
[0017] The local features are aggregated, and the aggregated local features are residually connected with the initial signal features to obtain the signal local features.
[0018] As a preferred embodiment, the residual convolution path includes several first residual modules and several second residual modules. The size of the convolutional layer and the number of output channels in the first residual module are 3×3 and 64, respectively, and the size of the convolutional layer and the number of output channels in the second residual module are 3×3 and 128, respectively. The number of residual convolution paths is equal to the number of grouped feature maps.
[0019] As a preferred embodiment, the step of using a preset temporal generative adversarial network to identify oscillation parameters of the oscillation mode features and obtain oscillation parameters corresponding to the oscillation signals in different frequency bands specifically includes:
[0020] Based on the temporal generative adversarial network, temporal features are extracted from the oscillation mode features to obtain oscillation parameter feature representations;
[0021] The oscillation parameter feature representation is used to identify the oscillation parameters to obtain the oscillation parameters corresponding to the oscillation signals in different frequency bands; wherein, the oscillation parameters include the oscillation frequency and the attenuation factor.
[0022] As a preferred embodiment, the step of extracting temporal features from the oscillation mode features based on the temporal generative adversarial network to obtain an oscillation parameter feature representation specifically includes:
[0023] Using several GRU layers in the temporal generative adversarial network, temporal features are extracted from the oscillation mode features to obtain temporal features of oscillation parameters;
[0024] By utilizing the fully connected layer in the temporal generative adversarial network, the temporal characteristics of the oscillation parameters are nonlinearly transformed to obtain the feature representation of the oscillation parameters.
[0025] As a preferred embodiment, the step of identifying oscillation parameters from the oscillation parameter feature representation to obtain the oscillation parameters corresponding to the oscillation signals in different frequency bands specifically includes:
[0026] Based on the output layer of the temporal generative adversarial network, the probability distribution of oscillation frequency and the probability distribution of attenuation factor corresponding to the oscillation parameter feature representation are calculated through activation functions;
[0027] Based on the oscillation frequency probability distribution and the attenuation factor probability distribution, the oscillation frequency and attenuation factor corresponding to the oscillation signal in different frequency bands are obtained respectively.
[0028] As a preferred embodiment, the method specifically trains the deep residual network and the temporal generative adversarial network through the following steps:
[0029] The ResNeXt network is trained using a broadband oscillation simulation signal to obtain the deep residual network; wherein the oscillation mode of the broadband oscillation simulation signal includes at least two of the following: low-frequency oscillation, subsynchronous oscillation, supersynchronous oscillation, mid-frequency oscillation, and high-frequency oscillation.
[0030] Using the oscillation mode features corresponding to the broadband oscillation simulation signal output by the ResNeXt network during training, the temporal generative adversarial network to be trained is trained to iteratively optimize the parameters of the generator and the discriminator in the temporal generative adversarial network to be trained, thereby obtaining the temporal generative adversarial network: wherein, the generator includes an input layer, a hidden layer and an output layer, and the hidden layer consists of two GRU layers and one fully connected layer.
[0031] A second aspect of the present invention provides a power system broadband oscillation parameter detection system, comprising:
[0032] The signal acquisition module is used to acquire the broadband oscillation signal to be detected.
[0033] The oscillation mode feature extraction module is used to perform grouped convolution on the broadband oscillation signal to be detected using a preset deep residual network to obtain the oscillation mode features corresponding to the oscillation signals of different frequency bands in the broadband oscillation signal to be detected.
[0034] The oscillation parameter identification module is used to identify the oscillation parameters of the oscillation mode features using a preset time-series generative adversarial network, and to obtain the oscillation parameters corresponding to the oscillation signals in different frequency bands.
[0035] Compared with the prior art, the beneficial effects of the embodiments of the present invention are that by using a deep residual network to perform grouped convolution on the broadband oscillation signal to be detected, it can effectively adapt to the multimodal characteristics of the broadband oscillation signal, thereby extracting the oscillation mode features of the oscillation signal in different frequency bands; considering the strong time-varying characteristics of the broadband oscillation signal, a temporal generative adversarial network can be used to further extract the temporal features of the oscillation parameters based on the oscillation mode features, thereby accurately obtaining the oscillation parameters of the oscillation signal in different frequency bands and effectively avoiding the problems of mode loss and false modes. Attached Figure Description
[0036] Figure 1 is a flowchart illustrating the power system broadband oscillation parameter detection method in an embodiment of the present invention;
[0037] Figure 2 is a schematic diagram showing the fitting relationship between the oscillation frequency obtained by using a deep residual network and a temporal generative adversarial network in an embodiment of the present invention and the actual set oscillation frequency.
[0038] Figure 3 is a schematic diagram showing the fitting relationship between the attenuation factor obtained by using a deep residual network and a temporal generative adversarial network in an embodiment of the present invention and the actual set attenuation factor.
[0039] Figure 4 is a schematic diagram of the power system broadband oscillation parameter detection system in an embodiment of the present invention. Detailed Implementation
[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0041] Please refer to Figure 1. A first aspect of this invention provides a method for detecting broadband oscillation parameters in a power system, comprising the following steps S1 to S3:
[0042] Step S1: Obtain the broadband oscillation signal to be detected;
[0043] Step S2: The broadband oscillation signal to be detected is grouped and convolved using a preset deep residual network to obtain the oscillation mode features corresponding to the oscillation signals of different frequency bands in the broadband oscillation signal to be detected.
[0044] Step S3: Use a preset temporal generative adversarial network to identify the oscillation parameters of the oscillation mode features and obtain the oscillation parameters corresponding to the oscillation signals in different frequency bands.
[0045] Specifically, since broadband oscillation signals have multimodal characteristics, meaning they may contain multiple oscillation modes such as low-frequency oscillation, subsynchronous oscillation, supersynchronous oscillation, mid-frequency oscillation, and high-frequency oscillation, for example, the frequency range of low-frequency oscillation is 0.1Hz to 2.5Hz, the frequency range of subsynchronous / supersynchronous oscillation is 2.5Hz to 100Hz, the frequency range of mid-frequency oscillation is 100Hz to 1000Hz, and the frequency range of high-frequency oscillation is above 1000Hz, this embodiment uses a deep residual network to perform grouped convolution on the acquired broadband oscillation signal to be detected, in order to extract the features of the broadband oscillation signal to be detected in groups, thereby realizing the extraction of complex features under multiple frequency components and avoiding the problems of mode loss and false modes.
[0046] Furthermore, since broadband oscillation signals have strong time-varying characteristics, this embodiment inputs the acquired oscillation mode features into a time-series generative adversarial network for further time-series feature extraction, thereby effectively capturing the dynamic changes of the broadband oscillation signal to be detected in the time dimension, determining the time-series features of the oscillation parameters, and using these time-series features of the oscillation parameters, the identification of the oscillation parameters can be achieved, thereby obtaining the oscillation parameters corresponding to oscillation signals in different frequency bands.
[0047] The power system broadband oscillation parameter detection method provided in this invention utilizes a deep residual network to perform grouped convolution on the broadband oscillation signal to be detected, which can effectively adapt to the multimodal characteristics of the broadband oscillation signal, thereby extracting the oscillation mode features of the oscillation signal in different frequency bands. Considering the strong time-varying characteristics of the broadband oscillation signal, a time-series generative adversarial network can be used to further extract the time-series features of the oscillation parameters based on the oscillation mode features, thereby accurately obtaining the oscillation parameters of the oscillation signal in different frequency bands and effectively avoiding the problems of mode loss and false modes.
[0048] As a preferred embodiment, the step of using a preset deep residual network to perform grouped convolution on the broadband oscillation signal to be detected, and obtaining the oscillation mode features corresponding to the oscillation signals of different frequency bands in the broadband oscillation signal to be detected, specifically includes:
[0049] Based on the deep residual network, feature extraction is performed on the broadband oscillation signal to be detected to obtain initial signal features;
[0050] The initial signal features are grouped and convolved to obtain local signal features;
[0051] The local features of the signal are subjected to average pooling to obtain the oscillation mode features corresponding to the oscillation signal in different frequency bands.
[0052] Specifically, this embodiment first extracts the initial signal features of the broadband oscillation signal to be detected. Then, it performs grouped convolution on these initial signal features to transform the features of each group into local features. Since the features of each group are processed separately, it can effectively extract important features of the oscillation signals in different frequency bands of the broadband oscillation signal to be detected, increasing the expressive power of the model while reducing the computational load. Furthermore, the signal local features obtained by grouped convolution are subjected to average pooling. It can be understood that average pooling retains the averaged overall features by calculating the average value of local regions, while also helping to remove noise, ultimately obtaining the oscillation mode features corresponding to the oscillation signals in different frequency bands.
[0053] As a preferred embodiment, the step of extracting features from the broadband oscillation signal to be detected based on the deep residual network to obtain initial signal features specifically includes:
[0054] Convolutional layers and max pooling layers are used to extract features and reduce the dimensionality of the broadband oscillation signal to be detected, respectively, to obtain the initial signal features.
[0055] Specifically, this embodiment first employs a 7×7 convolutional layer with 64 output channels and a stride of 2 to perform preliminary feature extraction on the broadband oscillation signal to be detected. Then, a 3×3 max-pooling layer with a stride of 2 is used to reduce the dimensionality of the initially extracted features, thereby shrinking the size of the feature map. It can be understood that the max-pooling layer can highlight important features in the data, enhancing the ability to identify the main features of local regions by preserving the strongest feature responses, ultimately obtaining the initial signal features. This embodiment achieves the goal of emphasizing different types of features through max pooling and average pooling, capturing more comprehensive feature information, strengthening the balance between feature enhancement and feature stabilization, enabling the overall network to adapt to different types of input data, thereby improving network performance.
[0056] As a preferred embodiment, the step of performing grouped convolution on the initial signal features to obtain local signal features specifically includes:
[0057] The initial signal features are grouped to obtain several grouped feature maps;
[0058] Based on several parallel residual convolution paths in the deep residual network, residual convolution is performed on several grouped feature maps to obtain the local features corresponding to each grouped feature map; wherein, the residual convolution path is formed by stacking several residual modules;
[0059] The local features are aggregated, and the aggregated local features are residually connected with the initial signal features to obtain the signal local features.
[0060] Specifically, in order to extract oscillation mode features such as low-frequency oscillations, subsynchronous oscillations, hypersynchronous oscillations, mid-frequency oscillations, and high-frequency oscillations, this embodiment first uses a 1×1 convolutional layer to group the initial signal features, obtaining several grouped feature maps. For example, based on experience, the number of groups is set to 32 groups. In actual applications, this number can be adjusted according to the complexity of the task, and this embodiment does not impose a specific limitation. Further, several parallel residual convolutional paths are used to perform parallel residual convolutions on several of the grouped feature maps. That is, the processing of each grouped feature map is independent of each other, and the topology of each residual convolutional path is the same. Finally, the local features of each grouped feature map are extracted. It is worth noting that the residual convolutional path is formed by stacking several residual modules, which can increase the depth of the deep residual network, enabling the model to simultaneously capture both shallow and deep complex features of broadband oscillation signals, achieving hierarchical extraction of all important features. Each residual module contains a convolutional layer, a batch normalization layer, and a ReLU activation function. The ReLU activation function is a non-linear activation function that outputs 0 when the input value is less than 0, and outputs the input value when the input value is greater than or equal to 0. This piecewise characteristic allows the ReLU function to introduce non-linearity because it is not linear throughout its entire domain. Moreover, the activation mode of the ReLU function is non-linear; different input values will cause different neurons to be activated or deactivated. This combination of activation modes can produce non-linear outputs, thus adapting to the non-linear characteristics of broadband oscillating signals.
[0061] Furthermore, the various local features are aggregated, and the aggregated local features are residually connected with the initial signal features, thereby preserving the information of the initial signal features and helping to avoid the gradient vanishing problem.
[0062] As a preferred embodiment, the residual convolution path includes several first residual modules and several second residual modules. The size of the convolutional layer and the number of output channels in the first residual module are 3×3 and 64, respectively, and the size of the convolutional layer and the number of output channels in the second residual module are 3×3 and 128, respectively. The number of residual convolution paths is equal to the number of grouped feature maps.
[0063] As a preferred embodiment, the step of using a preset temporal generative adversarial network to identify oscillation parameters of the oscillation mode features and obtain oscillation parameters corresponding to the oscillation signals in different frequency bands specifically includes:
[0064] Based on the temporal generative adversarial network, temporal features are extracted from the oscillation mode features to obtain oscillation parameter feature representations;
[0065] The oscillation parameter feature representation is used to identify the oscillation parameters to obtain the oscillation parameters corresponding to the oscillation signals in different frequency bands; wherein, the oscillation parameters include the oscillation frequency and the attenuation factor.
[0066] Specifically, the temporal generative adversarial network in this embodiment is an improved generative adversarial network. It maps high-dimensional time series data to a low-dimensional latent space, reducing the dimensionality of the generator's learning space while retaining the important features of the time series data, thus improving the accuracy of parameter identification. Using the temporal generative adversarial network, temporal feature extraction can be performed based on oscillation mode features. It can be understood that the oscillation frequency is the number of periodic changes that the oscillation completes per unit time, which determines the speed of signal change, while the attenuation factor refers to the proportion of signal attenuation in each period. That is, as the broadband oscillation signal dynamically changes in the time dimension, the oscillation parameters also change. Therefore, through temporal feature extraction, a characteristic representation of the oscillation parameters can be obtained.
[0067] As a preferred embodiment, the step of extracting temporal features from the oscillation mode features based on the temporal generative adversarial network to obtain an oscillation parameter feature representation specifically includes:
[0068] Using several GRU layers in the temporal generative adversarial network, temporal features are extracted from the oscillation mode features to obtain temporal features of oscillation parameters;
[0069] By utilizing the fully connected layer in the temporal generative adversarial network, the temporal characteristics of the oscillation parameters are nonlinearly transformed to obtain the feature representation of the oscillation parameters.
[0070] Specifically, the temporal generative adversarial network in this embodiment consists of a generator and a discriminator. The generator includes an input layer, a hidden layer, and an output layer. The hidden layer consists of several GRU (Gated Recurrent Unit) layers and one fully connected layer. The input layer is used to transmit oscillation mode features to the hidden layer. The GRU layer in the hidden layer is responsible for extracting features in the time dimension. Through the GRU gating mechanism, namely the update gate and reset gate, it can effectively capture the dynamic changes of broadband oscillation signals in the time dimension, remember information over long time spans, adapt to the dynamic changes of broadband oscillation signals in time, and provide real-time oscillation parameter identification capabilities. The update gate and reset gate allow the model to simultaneously handle the long-short-term dependencies in the signal, capturing short-term changes while retaining long-term dependency information, thereby obtaining the temporal features of the broadband oscillation signal. Then, the fully connected layer is used to integrate the temporal features extracted by the GRU layer and realize flexible transformation from the input feature dimension to the output feature dimension, so as to map the high-dimensional temporal feature space to the low-dimensional oscillation parameter identification space.
[0071] As a preferred embodiment, the step of identifying oscillation parameters from the oscillation parameter feature representation to obtain the oscillation parameters corresponding to the oscillation signals in different frequency bands specifically includes:
[0072] Based on the output layer of the temporal generative adversarial network, the probability distribution of oscillation frequency and the probability distribution of attenuation factor corresponding to the oscillation parameter feature representation are calculated through activation functions;
[0073] Based on the oscillation frequency probability distribution and the attenuation factor probability distribution, the oscillation frequency and attenuation factor corresponding to the oscillation signal in different frequency bands are obtained respectively.
[0074] Specifically, based on the output layer of the generator in the temporal generative adversarial network, the activation function can be used to calculate the probability of different oscillation frequencies and different attenuation factors corresponding to the current oscillation parameter feature representation, thereby determining the probability distribution of oscillation frequency and the probability distribution of attenuation factor. Based on the oscillation frequency and attenuation factor with the highest probability in the probability distribution of oscillation frequency and attenuation factor, the oscillation frequency and attenuation factor corresponding to the oscillation signal of different frequency bands can be output.
[0075] As a preferred embodiment, the method specifically trains the deep residual network and the temporal generative adversarial network through the following steps:
[0076] The ResNeXt network is trained using a broadband oscillation simulation signal to obtain the deep residual network; wherein the oscillation mode of the broadband oscillation simulation signal includes at least two of the following: low-frequency oscillation, subsynchronous oscillation, supersynchronous oscillation, mid-frequency oscillation, and high-frequency oscillation.
[0077] Using the oscillation mode features corresponding to the broadband oscillation simulation signal output by the ResNeXt network during training, the temporal generative adversarial network to be trained is trained to iteratively optimize the parameters of the generator and the discriminator in the temporal generative adversarial network to be trained, thereby obtaining the temporal generative adversarial network: wherein, the generator includes an input layer, a hidden layer and an output layer, and the hidden layer consists of two GRU layers and one fully connected layer.
[0078] Specifically, the deep residual network in this embodiment is the ResNeXt network. The ResNeXt network borrows the grouping idea from the Inception structure, proposing a grouped convolutional module, which, combined with the residual module, results in a simpler network structure with fewer parameters, and outperforms the ResNet network in performance. It not only retains the advantage of the ResNet network in avoiding the vanishing gradient problem caused by network stacking, but also incorporates the idea of multi-scale processing, dividing the input data into multiple datasets and inputting them separately into different paths of the model to extract features. Each feature extraction channel in the model uses the same stacked convolutional neural network structure, reducing the number of hyperparameters compared to other multi-scale networks. To enable the ResNeXt network to learn the ability to extract oscillation mode features of different oscillation modes, this embodiment first simulates a wideband oscillation simulation signal with oscillation modes including at least two of low-frequency oscillation, subsynchronous oscillation, supersynchronous oscillation, mid-frequency oscillation, and high-frequency oscillation. Preferably, the oscillation modes of the wideband oscillation simulation signal simultaneously include low-frequency oscillation, subsynchronous / supersynchronous oscillation, mid-frequency oscillation, and high-frequency oscillation. Then, the original ResNeXt network is trained using this wideband oscillation simulation signal. For example, the wideband oscillation simulation signal is divided into a training set and a test set in a 7:3 ratio. The training set is used for training, and the test set is used to verify the network performance after each training iteration. The parameters of the ResNeXt network are optimized through iterative training, and finally, the required deep residual network is obtained.
[0079] Furthermore, during the training of the ResNeXt network, the extracted oscillatory mode features are selected and dimensionality reduced to filter out unnecessary features and decrease data dimensionality. Then, these pre-trained oscillatory mode features are used to train the temporal generative adversarial network (GAN) to be trained, iteratively optimizing the parameters of the generator and discriminator in the GAN. In each training round, the generator parameters are first fixed, and the discriminator is trained. Real temporal samples and the temporal samples generated by the generator in the previous round are input into the discriminator for scoring, and the loss function (usually binary cross-entropy loss) is calculated based on the scores to update the discriminator parameters. Then, the discriminator parameters are fixed again, and the generator is trained. The generator generates corresponding temporal features based on the input oscillatory mode features, which are input into the discriminator for scoring. The difference between the score and the true label (usually 1) is used as the loss function, and the generator parameters are updated through backpropagation to minimize the loss function. It is worth noting that the hidden layer of the generator in this embodiment includes two GRU layers, which ensures both fast extraction of temporal features and accurate detection of oscillation parameters.
[0080] After training the ResNeXt network and the temporal generative adversarial network, the oscillation parameter detection performance of the trained deep residual network and the temporal generative adversarial network was evaluated using independent validation sets and the aforementioned test sets. The evaluation metrics were mean absolute error and root mean square error. The test set was input into the trained deep residual network and tested in conjunction with the temporal generative adversarial network. The oscillation frequency and attenuation factor in the test results were evaluated. Table 1 below shows the average analysis results of 300 tests.
[0081] Table 1. Detection and evaluation results of oscillation frequency and attenuation factor
[0082] Figures 2 and 3 show the fitting results between the oscillation frequency and attenuation factor detected by the trained deep residual network and temporal generative adversarial network and the actual set oscillation frequency and attenuation factor, respectively. As can be seen from Table 1, Figure 2 and Figure 3, the deep residual network and temporal generative adversarial network in this embodiment have good detection performance for oscillation parameters, and the fitting effect between the detected oscillation frequency and attenuation factor and the actual oscillation frequency and attenuation factor is good, with good detection accuracy.
[0083] Please refer to Figure 4. A second aspect of this invention provides a power system broadband oscillation parameter detection system, comprising:
[0084] Signal acquisition module 11 is used to acquire the broadband oscillation signal to be detected;
[0085] The oscillation mode feature extraction module 12 is used to perform group convolution on the broadband oscillation signal to be detected using a preset deep residual network to obtain the oscillation mode features corresponding to the oscillation signals of different frequency bands in the broadband oscillation signal to be detected.
[0086] The oscillation parameter identification module 13 is used to identify the oscillation parameters of the oscillation mode features using a preset time-series generative adversarial network, and obtain the oscillation parameters corresponding to the oscillation signals in different frequency bands.
[0087] As a preferred embodiment, the oscillation mode feature extraction module 12 is used to perform grouped convolution on the broadband oscillation signal to be detected using a preset deep residual network to obtain the oscillation mode features corresponding to the oscillation signals of different frequency bands in the broadband oscillation signal to be detected, specifically including:
[0088] Based on the deep residual network, feature extraction is performed on the broadband oscillation signal to be detected to obtain initial signal features;
[0089] The initial signal features are grouped and convolved to obtain local signal features;
[0090] The local features of the signal are subjected to average pooling to obtain the oscillation mode features corresponding to the oscillation signal in different frequency bands.
[0091] As a preferred embodiment, the oscillation mode feature extraction module 12 is used to extract features from the broadband oscillation signal to be detected based on the deep residual network to obtain initial signal features, specifically including:
[0092] Convolutional layers and max pooling layers are used to extract features and reduce the dimensionality of the broadband oscillation signal to be detected, respectively, to obtain the initial signal features.
[0093] As a preferred embodiment, the oscillation mode feature extraction module 12 is used to perform grouped convolution on the initial signal features to obtain local signal features, specifically including:
[0094] The initial signal features are grouped to obtain several grouped feature maps;
[0095] Based on several parallel residual convolution paths in the deep residual network, residual convolution is performed on several grouped feature maps to obtain the local features corresponding to each grouped feature map; wherein, the residual convolution path is formed by stacking several residual modules;
[0096] The local features are aggregated, and the aggregated local features are residually connected with the initial signal features to obtain the signal local features.
[0097] As a preferred embodiment, the residual convolution path includes several first residual modules and several second residual modules. The size of the convolutional layer and the number of output channels in the first residual module are 3×3 and 64, respectively, and the size of the convolutional layer and the number of output channels in the second residual module are 3×3 and 128, respectively. The number of residual convolution paths is equal to the number of grouped feature maps.
[0098] As a preferred embodiment, the oscillation parameter identification module 13 is used to identify the oscillation parameters of the oscillation mode features using a preset time-series generative adversarial network, and to obtain the oscillation parameters corresponding to the oscillation signals in different frequency bands, specifically including:
[0099] Based on the temporal generative adversarial network, temporal features are extracted from the oscillation mode features to obtain oscillation parameter feature representations;
[0100] The oscillation parameter feature representation is used to identify the oscillation parameters to obtain the oscillation parameters corresponding to the oscillation signals in different frequency bands; wherein, the oscillation parameters include the oscillation frequency and the attenuation factor.
[0101] As a preferred embodiment, the oscillation parameter identification module 13 is used to extract temporal features from the oscillation mode features based on the temporal generative adversarial network to obtain an oscillation parameter feature representation, specifically including:
[0102] Using several GRU layers in the temporal generative adversarial network, temporal features are extracted from the oscillation mode features to obtain temporal features of oscillation parameters;
[0103] By utilizing the fully connected layer in the temporal generative adversarial network, the temporal characteristics of the oscillation parameters are nonlinearly transformed to obtain the feature representation of the oscillation parameters.
[0104] As a preferred embodiment, the oscillation parameter identification module 13 is used to identify the oscillation parameters from the oscillation parameter feature representation to obtain the oscillation parameters corresponding to the oscillation signals in different frequency bands, specifically including:
[0105] Based on the output layer of the temporal generative adversarial network, the probability distribution of oscillation frequency and the probability distribution of attenuation factor corresponding to the oscillation parameter feature representation are calculated through activation functions;
[0106] Based on the oscillation frequency probability distribution and the attenuation factor probability distribution, the oscillation frequency and attenuation factor corresponding to the oscillation signal in different frequency bands are obtained respectively.
[0107] As a preferred embodiment, the system further includes a model training module, used for:
[0108] The ResNeXt network is trained using a broadband oscillation simulation signal to obtain the deep residual network; wherein the oscillation mode of the broadband oscillation simulation signal includes at least two of the following: low-frequency oscillation, subsynchronous oscillation, supersynchronous oscillation, mid-frequency oscillation, and high-frequency oscillation.
[0109] Using the oscillation mode features corresponding to the broadband oscillation simulation signal output by the ResNeXt network during training, the temporal generative adversarial network to be trained is trained to iteratively optimize the parameters of the generator and the discriminator in the temporal generative adversarial network to be trained, thereby obtaining the temporal generative adversarial network: wherein, the generator includes an input layer, a hidden layer and an output layer, and the hidden layer consists of two GRU layers and one fully connected layer.
[0110] The power system broadband oscillation parameter detection system provided in this invention utilizes a deep residual network to perform grouped convolution on the broadband oscillation signal to be detected, which can effectively adapt to the multimodal characteristics of the broadband oscillation signal, thereby extracting the oscillation mode features of the oscillation signal in different frequency bands. Considering the strong time-varying characteristics of the broadband oscillation signal, a time-series generative adversarial network can be used to further extract the time-series features of the oscillation parameters based on the oscillation mode features, thereby accurately obtaining the oscillation parameters of the oscillation signal in different frequency bands and effectively avoiding the problems of mode loss and false modes.
[0111] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.
Claims
1. A method for detecting broadband oscillation parameters in a power system, characterized in that, include: Acquire the broadband oscillation signal to be detected; A preset deep residual network is used to perform grouped convolution on the broadband oscillation signal to be detected, so as to obtain the oscillation mode features corresponding to the oscillation signals of different frequency bands in the broadband oscillation signal to be detected; A preset temporal generative adversarial network is used to identify the oscillation parameters of the oscillation mode features, thereby obtaining the oscillation parameters corresponding to the oscillation signals in different frequency bands.
2. The power system broadband oscillation parameter detection method as described in claim 1, characterized in that, The step involves using a pre-defined deep residual network to perform grouped convolution on the broadband oscillation signal to be detected, thereby obtaining the oscillation mode features corresponding to oscillation signals in different frequency bands within the broadband oscillation signal to be detected. Specifically, this includes: Based on the deep residual network, feature extraction is performed on the broadband oscillation signal to be detected to obtain initial signal features; The initial signal features are grouped and convolved to obtain local signal features; The local features of the signal are subjected to average pooling to obtain the oscillation mode features corresponding to the oscillation signal in different frequency bands.
3. The method for detecting broadband oscillation parameters in a power system as described in claim 2, characterized in that, The step of extracting features from the broadband oscillation signal to be detected based on the deep residual network to obtain initial signal features specifically includes: Convolutional layers and max pooling layers are used to extract features and reduce the dimensionality of the broadband oscillation signal to be detected, respectively, to obtain the initial signal features.
4. The method for detecting broadband oscillation parameters in a power system as described in claim 2, characterized in that, The step of performing grouped convolution on the initial signal features to obtain local signal features specifically includes: The initial signal features are grouped to obtain several grouped feature maps; Based on several parallel residual convolution paths in the deep residual network, residual convolution is performed on several grouped feature maps to obtain the local features corresponding to each grouped feature map; wherein, the residual convolution path is formed by stacking several residual modules; The local features are aggregated, and the aggregated local features are residually connected with the initial signal features to obtain the signal local features.
5. The power system broadband oscillation parameter detection method as described in claim 4, characterized in that, The residual convolution path includes several first residual modules and several second residual modules. The size of the convolutional layer and the number of output channels in the first residual module are 3×3 and 64, respectively. The size of the convolutional layer and the number of output channels in the second residual module are 3×3 and 128, respectively. The number of residual convolution paths is equal to the number of grouped feature maps.
6. The method for detecting broadband oscillation parameters in a power system as described in claim 1, characterized in that, The step of using a preset temporal generative adversarial network to identify oscillation parameters of the oscillation mode features, and obtaining the oscillation parameters corresponding to the oscillation signals in different frequency bands, specifically includes: Based on the temporal generative adversarial network, temporal features are extracted from the oscillation mode features to obtain oscillation parameter feature representations; The oscillation parameter feature representation is used to identify the oscillation parameters to obtain the oscillation parameters corresponding to the oscillation signals in different frequency bands; wherein, the oscillation parameters include the oscillation frequency and the attenuation factor.
7. The power system broadband oscillation parameter detection method as described in claim 6, characterized in that, The step of extracting temporal features from the oscillation mode features based on the temporal generative adversarial network to obtain oscillation parameter feature representations specifically includes: Using several GRU layers in the temporal generative adversarial network, temporal features are extracted from the oscillation mode features to obtain temporal features of oscillation parameters; By utilizing the fully connected layer in the temporal generative adversarial network, the temporal characteristics of the oscillation parameters are nonlinearly transformed to obtain the feature representation of the oscillation parameters.
8. The method for detecting broadband oscillation parameters in a power system as described in claim 6, characterized in that, The step of identifying oscillation parameters from the oscillation parameter feature representation to obtain the oscillation parameters corresponding to the oscillation signals in different frequency bands specifically includes: Based on the output layer of the temporal generative adversarial network, the probability distribution of oscillation frequency and the probability distribution of attenuation factor corresponding to the oscillation parameter feature representation are calculated through activation functions; Based on the oscillation frequency probability distribution and the attenuation factor probability distribution, the oscillation frequency and attenuation factor corresponding to the oscillation signal in different frequency bands are obtained respectively.
9. The method for detecting broadband oscillation parameters in a power system as described in claim 1, characterized in that, The method specifically trains the deep residual network and the temporal generative adversarial network through the following steps: The ResNeXt network is trained using a broadband oscillation simulation signal to obtain the deep residual network; wherein the oscillation mode of the broadband oscillation simulation signal includes at least two of the following: low-frequency oscillation, subsynchronous oscillation, supersynchronous oscillation, mid-frequency oscillation, and high-frequency oscillation. Using the oscillation mode features corresponding to the broadband oscillation simulation signal output by the ResNeXt network during training, the temporal generative adversarial network to be trained is trained to iteratively optimize the parameters of the generator and the discriminator in the temporal generative adversarial network to be trained, thereby obtaining the temporal generative adversarial network: wherein, the generator includes an input layer, a hidden layer and an output layer, and the hidden layer consists of two GRU layers and one fully connected layer.
10. A power system wide area oscillation parameter detection system, characterized by, include: The signal acquisition module is used to acquire the broadband oscillation signal to be detected. The oscillation mode feature extraction module is used to perform grouped convolution on the broadband oscillation signal to be detected using a preset deep residual network to obtain the oscillation mode features corresponding to the oscillation signals of different frequency bands in the broadband oscillation signal to be detected. The oscillation parameter identification module is used to identify the oscillation parameters of the oscillation mode features using a preset time-series generative adversarial network, and to obtain the oscillation parameters corresponding to the oscillation signals in different frequency bands.