FCNs-BiGRU-based power system transient stability evaluation method

Abstract

The invention belongs to the field of power system transient stability evaluation, and particularly relates to an FCNs-BiGRU-based power system transient stability evaluation method, which comprises the following steps of: generating an original data set reflecting different power flow operation modes and fault conditions, and extracting transient electrical quantity as an input characteristic; modals with different characteristics of a sample are extracted by improving empirical mode decomposition, and transient characteristics of an electrical signal are revealed; in consideration of respective short-term and long-term feature extraction capabilities of a full convolutional neural network and a bidirectional gating cycle unit, obtaining a mapping relation between original high-dimensional input and a sample stability label according to the processed electrical measurement data; and after a multi-dimensional feature driving integration model is constructed based on offline training, online transient stability evaluation is completed in an online application process by using the trained model. The method can consider the time-frequency domain information of the electric power system sample, construct a data set with moderate and abundant sample characteristics, and accurately judge the actual operation condition of the electric power system.

Description

Technical Field

[0001] The present invention belongs to the field of power system transient stability assessment, and specifically relates to a power system transient stability assessment method based on FCNs-BiGRU. Background Art

[0002] Power system transient stability refers to the ability of a power system to reach a new steady-state operation mode or restore to the original operation mode after a large disturbance occurs under a certain operation mode. With the continuous development of the smart grid construction, it is crucial to maintain the stability of the power system. When the power system is subjected to a large disturbance and corresponding countermeasures are not taken in a timely manner, the power system is very likely to experience transient instability, leading to the collapse of the entire system and causing large-scale regional power outages. Therefore, it is necessary to conduct transient stability assessment of the power system.

[0003] The power system transient stability assessment method based on deep learning has great potential for rapid and accurate judgment of power system transient stability. A deep learning model can be trained using a large amount of historical data in an end-to-end training mode to establish a mapping relationship with the stable state, and predict the transient stability state of the current power system to achieve accurate and rapid assessment. With the rapid development of the power system and the continuous access of a large number of renewable energy sources, the normal operation of the power system faces increasing risks and challenges. The power system is affected by various faults during normal operation, resulting in an increasing possibility of transient instability. Once transient instability occurs and no control measures are taken, it will directly lead to the out-of-step of the generator power angle, and in severe cases, the power system will be disconnected and collapsed, causing inestimable losses to the normal operations of large enterprises. The phenomenon of power system transient instability caused by faults has occurred repeatedly, which has had a great impact on the normal operation of the power grid. In the existing power system transient stability assessment (TSA) technology, there is a problem that the feature extraction is limited by a single model. Therefore, it is necessary to design a transient assessment method based on an improved fully convolutional neural network - bidirectional gated recurrent unit to solve the above problems. Summary of the Invention

[0004] In order to solve the problem that the feature extraction is limited by a single model in the existing transient stability assessment, the present invention provides a power system transient stability assessment method based on FCNs-BiGRU.

[0005] The technical solution of the present invention to solve the above problems is as follows: A power system transient stability assessment method based on FCNs-BiGRU includes the following steps:

[0006] Step 1: Through time-domain simulation, batch simulation generation of samples is performed to obtain an operation dataset of the power system;

[0007] Step 2: Process the operating dataset through the improved empirical mode decomposition method IEMD to establish the training dataset of the transient stability evaluation model;

[0008] Step 3: Based on the training dataset, train the transient stability evaluation model based on the improved fully convolutional neural network FCNs and bidirectional gated recurrent unit BiGRU. The trained transient stability evaluation model is used for online transient stability evaluation;

[0009] Step 4: Use the phasor measurement unit PMU to obtain the required measurement data online and input it into the trained transient stability evaluation model to output the transient stability evaluation result.

[0010] For the above-mentioned power system transient stability evaluation method based on FCNs-BiGRU, in Step 1, build an actual power system through time-domain simulation, set different fault times, fault locations, and new energy output powers, and extract the time-series data from before the fault to after the fault excision during each simulation process as feature quantities, and obtain the corresponding transient stability labels corresponding to the feature quantities after the simulation. Simulate different operating conditions to finally generate an operating dataset containing a large amount of operating state information of the power system.

[0011] For the above-mentioned power system transient stability evaluation method based on FCNs-BiGRU, in Step 1, specifically select the voltage amplitude and phase angle of the bus, and the active and reactive powers of the branch as feature quantities to form the operating dataset ; denotes the feature quantity, denotes the transient stability label determined by the system, corresponding to the feature quantity ; Before Step 2, the feature quantity is also normalized.

[0012] For the above-mentioned power system transient stability evaluation method based on FCNs-BiGRU, the specific process of Step 2 is as follows:

[0013] Step 2-1: First, superimpose an independent white noise on the time-series data in the normalized feature quantity , which is expressed as follows: . After the operation of the empirical mode decomposition method EMD, is decomposed into components and residuals . The number of decomposition times is K, and the decomposition result is as follows: ;

[0014] Step 2-2: Calculate the corresponding Mean and residual of components The mean value is calculated as follows: , , where represents the mean value of the p-th component, represents the mean value of the residual;

[0015] Based on the mean values of these components and the mean value of the residual, the processed time series data is obtained as follows: ; The processed time series data and the corresponding transient stability labels form a training dataset ;

[0016] Step 2 - 3: Compare the total energy of all components after calculating the mean value to dynamically adjust the decomposition times K to maximize the total energy and optimize the decomposition process; The energy ratio relationship between all components and the time series data is as follows: , represents the total energy of all components when the decomposition times is K, represents the initial total energy, which is a fixed value. When the energy ratio reaches the maximum value, the total energy of the corresponding all components also reaches the maximum value, indicating that the decomposition times K is optimal.

[0017] In the above transient stability assessment method for power systems based on FCNs - BiGRU, in Step 3, by improving the fully convolutional neural network FCNs and the bidirectional gated recurrent unit BiGRU, the spatial features and temporal relationship features in are extracted respectively.

[0018] In the above transient stability assessment method for power systems based on FCNs - BiGRU, the specific training process of the transient stability assessment model is as follows:

[0019] Step 3 - 1: To improve the fully convolutional neural network FCNs, two fully convolutional neural networks FCN are used in parallel to achieve dual - channel processing of samples. One channel processes the data to obtain time - domain spatial features, and the other channel first performs the fast Fourier transform FFT and then processes the data to obtain frequency - domain spatial features; The convolutional kernels in FCNs are adjusted to one - dimensional; A global average pooling GAP layer is applied after the last convolutional layer; The channel attention mechanism is used, and SENet is embedded after each convolutional operation;

[0020] Step 3-2: The transient stability assessment model includes FCN, BiGRU, and a feature merging module. Two FCNs extract spatial features in the time-frequency domain through convolution, channel attention mechanism, and pooling; the BiGRU module extracts temporal relationship features; then, after integrating the spatial features and temporal relationship features through a fully connected layer, the Sigmoid function is used to map the result to the interval [0, 1] to represent the probability of belonging to transient stability or transient instability.

[0021] Step 3-3: According to the training dataset obtained in Step 2, input it into the transient stability assessment model for training to establish a transient stability assessment model based on FCNs-BiGRU.

[0022] For the above-mentioned power system transient stability assessment method based on FCNs-BiGRU, in Step 4, based on the established transient stability assessment model, when applied online, the actual measurement data is input into the transient stability assessment model, and the transient stability assessment model outputs the assessment result, thereby obtaining the power system transient stability assessment result.

[0023] The present invention proposes a dual-channel transient stability assessment model FCNs-BiGRU that integrates a fully convolutional neural network and a bidirectional gated recurrent unit, realizing multi-dimensional extraction and high-precision classification of key features of the power system. The improved empirical mode decomposition method IEMD is used to enhance the features of the training dataset to construct a dataset with an appropriate and rich number of sample features; the synergistic effect of the parallel dual-channel architecture and the channel attention mechanism can consider the time-frequency domain information of power system samples and accurately judge the actual operating conditions of the power system, showing advantages in the transient stability assessment task. Description of the Drawings

[0024] In order to make the content of the present invention easier to be clearly understood, the following further details the present invention according to specific embodiments of the present invention in combination with the drawings, where:

[0025] Figure 1 Shows a flowchart of a power system transient stability assessment method based on FCNs-BiGRU according to an embodiment of the present application;

[0026] Figure 2 Shows the FCNs-BiGRU model structure diagram according to an embodiment of the present application. Detailed Embodiments

[0027] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the following will, in conjunction with the accompanying drawings in the embodiments of this application, clearly and completely describe the technical solutions in the embodiments of this application. It should be understood that the accompanying drawings in this application are only for the purposes of illustration and description, and are not used to limit the protection scope of this application. The flowcharts used in this application illustrate the operations implemented according to some embodiments of this application. It should be understood that the operations in the flowchart may not be implemented in sequence, and steps without a logical context relationship may be reversed in order or implemented simultaneously. In addition, those skilled in the art may add one or more other operations to the flowchart or remove one or more operations from the flowchart under the guidance of the content of this application.

[0028] In addition, the described embodiments are only a part of the embodiments of this application, rather than all of the embodiments. The components of the embodiments of this application usually described and illustrated in the accompanying drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of this application claimed, but only represents the selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative efforts fall within the protection scope of this application.

[0029] For ease of understanding of this application, the following will, in conjunction with specific embodiments, elaborate on the technical solutions provided by this application in detail.

[0030] Figure 1 It is a flowchart of a transient stability assessment method for a power system based on FCNs-BiGRU provided by an embodiment of this application. As Figure 1 shown, it includes the following steps:

[0031] Step 1: Through time-domain simulation, batch simulation generation of samples is performed to obtain an operating dataset of the power system.

[0032] In this embodiment, an actual power system is built through time-domain simulation, different fault times, fault locations, and new energy output powers are set, the sampling interval is set to 0.01 s, and the time-series data from before the fault occurs to after the fault is cleared during each simulation process is extracted as feature quantities, and the transient stability labels corresponding to the feature quantities are obtained after the simulation ends. Simulations are performed for different working conditions, and finally an operating dataset containing a large amount of operating state information of the power system is generated.

[0033] Specifically, time-series data such as the voltage amplitude and phase angle of the busbar, and the active power and reactive power of the branch are selected as feature quantities to form an operating dataset ; denotes the feature quantity, Indicates the transient stability label determined by the system, corresponding to the characteristic quantity Correspondingly. If the system is transiently stable, the transient stability label is 1; otherwise, if the system is transiently unstable, the transient stability label is 0.

[0034] Select the maximum power angle difference between two generator sets at any time after the system fault as the basis for transient stability discrimination. The transient stability index TSI is calculated as follows:

[0035]

[0036] Where is the maximum power angle difference between two generator sets at any time; when , the system is determined to be transiently stable; otherwise, the system is determined to be transiently unstable.

[0037] Before the model training in step 2, normalize the characteristic quantity .

[0038] Step 2: Process the operation dataset through an improved empirical mode decomposition method IEMD to establish a training dataset with an appropriate and rich number of features. The principle of IEMD is as follows:

[0039] Step 2-1: First, use the empirical mode decomposition EMD to extract multiple intrinsic mode functions from the time series data in the normalized characteristic quantity. These intrinsic mode functions help isolate various frequency components, thus providing detailed insights into the characteristic quantity. For the time series data

[0040] , first superimpose an independent white noise in the EMD decomposition. It can be expressed as follows: ; ;

[0041] After the EMD operation, is decomposed into components and the residual . The number of decomposition times is K, and the number of is also represented by K. The decomposition result is as follows:

[0042]

[0043] Step 2-2: Calculate the mean of the corresponding components and the mean of the residual . To eliminate the influence of white noise on the components, calculate the mean of the corresponding components and the residual The mean value is calculated as follows:

[0044]

[0045] In the formula, represents the mean value of the p-th component, represents the mean value of the residual, and M is the total number of components.

[0046] Based on the mean values of these components and the mean value of the residual, the processed time series data is obtained as follows:

[0047]

[0048] This step can enhance the noise reduction ability.

[0049] The processed time series data and the corresponding transient stability labels form a training data set .

[0050] Step 2 - 3: If the decomposition times are too high or too low, the cumulative energy of all components may decrease. Considering this situation, a method for selecting the decomposition times is proposed, which involves comparing the total energy of all components. The core of the improved empirical mode decomposition method is to compare the total energy of all components after calculating the mean values to dynamically adjust the decomposition times K to maximize the total energy and optimize the decomposition process. The energy ratio relationship between all components and the time series data is as follows:

[0051] The initial total energy can be defined as follows: ,

[0052] where represents the number of sampling points, represents the amplitude of the time series data at different sampling points.

[0053] After the mode decomposition operation, multiple components are obtained. The total energy of these components can be expressed as follows:

[0054]

[0055] And there is

[0056] where represents all when the decomposition times is K The total energy of the components. According to the above formula, all components and the time series data The energy ratio between them is calculated as follows: [[ID=⑥]]

[0057]

[0058] When the sampling frequency and sampling period are determined, the corresponding sampling points are fixed. That is, the initial total energy [[ID=②]] is a fixed value. When the energy ratio reaches the maximum value, the total energy of all corresponding also reaches the maximum value, indicating that the decomposition times K is optimal.

[0059] Step 3: Based on the training dataset , train a transient stability assessment model based on the improved fully convolutional neural network FCNs and bidirectional gated recurrent unit (BiGRU). Use the channel attention mechanism (Squeeze-and-Excitation Network, SENet) to complete the attention calculation of each convolutional layer, extract the relationship between the key features of the power system and the corresponding transient stability labels. The trained transient stability assessment model is used for online transient stability assessment.

[0060] In Step 3, the feature quantity processed by IEMD is input into the transient stability assessment model. The transient stability assessment model respectively constructs modules suitable for multi-dimensional feature mining, and simultaneously extracts the spatial features and time series relationship features in the feature quantity .

[0061] Step 3-1: The fully convolutional network (FCN) is a special network and belongs to the category of CNN. Compared with the conventional CNN, the FCN omits the fully connected layer at the end, which results in returning the overall features instead of the classification results. Using the FCN that only contains convolutional and pooling layers to extract spatial features is more simple and efficient. In this embodiment, as Figure 2 shown, the fully convolutional neural network is changed to a parallel two-channel processing sample to form an improved fully convolutional neural network FCNs. Among them, FCN1 processes the data to obtain time-domain spatial features, and FCN2 first processes the data through the fast Fourier transform FFT to obtain frequency-domain spatial features. The convolutional kernels in FCNs are adjusted to one-dimensional; a global average pooling GAP layer is applied after the last convolutional layer; the channel attention mechanism is used, and SENet is embedded after each convolutional operation. By modeling the correlation between feature channels, important features are enhanced. It should be noted that in the above translation, some tags are retained in their original form as they seem to be specific identifiers in a technical context and may not have a direct equivalent in English. Also, the translation is done as accurately as possible while following the given rules to maintain the integrity of the original text structure.

[0062] Compared with traditional RNNs and LSTMs, BiGRU is a lighter, more efficient, and memory-capable option, making it an ideal choice for extracting temporal relationship features of feature quantities. BiGRU first divides the feature quantity into T groups according to the period to form reshaped data, then inputs the reshaped data into a BiGRU network composed of forward and backward GRU cells in parallel, and finally merges the forward and backward time-related eigenvalue groups to form temporal relationship features.

[0063] Step 3-2: The transient stability assessment model includes an FCN, BiGRU, and feature fusion module. Two FCNs extract spatial features in the time-frequency domain through convolution, channel attention mechanism, and pooling; BiGRU extracts temporal relationship features; then, after integrating the spatial features and temporal relationship features through a fully connected layer, the Sigmoid function is used to map the result to the interval [0, 1] to represent the probability of belonging to transient stability or transient instability.

[0064] Step 3-3: According to the training dataset obtained in Step 2, input it into the transient stability assessment model obtained in Step 3 for training to establish a transient stability assessment model based on FCNs-BiGRU.

[0065] Step 4: Use a Phasor Measurement Unit (PMU) to obtain the required measurement data online and input it into the trained transient stability assessment model to output the transient stability assessment result.

[0066] In Step 4, for the transient stability assessment model based on FCNs-BiGRU, when applied online, the phasor measurement unit in the wide-area measurement system is used to collect the required real-time measurement data, and the measurement data is input into the trained transient stability assessment model to obtain the transient stability assessment result of the power system and complete the online transient stability assessment. If the assessment result is instability, a risk warning will be sent to the staff and emergency braking measures will be taken immediately; if the assessment result is stability, the transient stability assessment model will continue to monitor the system.

[0067] Embodiment:

[0068] The embodiment used in the present invention is based on the New England 10-machine 39-bus system, which consists of 10 generators, 39 buses, and 19 load nodes.

[0069] During the transient stability simulation, the wind power penetration rates are considered as 5%, 10%, and 15%; with a step size of 5%, five different load levels are set in the range of 85% to 115%; the faulty lines are 33 AC lines; three-phase short-circuit faults are set at 30%, 50%, and 70% of the AC lines; the fault durations are set as 9 cycles, 10 cycles, and 11 cycles. One cycle is 1 / 60 s, and 4455 original samples are generated through time-domain simulation, among which there are 3019 stable samples and 1436 unstable samples. Each sample contains: bus voltage amplitude, bus voltage phase angle, generator active power, generator reactive power, branch active power, branch reactive power, generator angular velocity, generator frequency deviation, mechanical power, and generator power angle; finally, 70% of the samples are randomly selected for model training, and the remaining 30% of the samples are used for model testing.

[0070] Select evaluation indicators and to test the classification performance of the transient stability assessment model based on FCNs-BiGRU. Let be the correctly evaluated stable samples, be the incorrectly evaluated stable samples, be the incorrectly evaluated unstable samples, be the correctly evaluated unstable samples. The specific expressions are as follows:

[0071]

[0072]

[0073]

[0074]

[0075] In the formula: represents the overall accuracy; the precision is the ratio of the correctly evaluated stable samples to all the evaluated stable samples; the recall is the ratio of the correctly evaluated stable samples to all the actual stable samples; is defined as the harmonic mean of the precision and the recall ; ranges from (0, 1), the closer it is to 1, the better the classification performance of the TSA model.

[0076] To demonstrate the remarkable classification performance of the transient stability assessment model based on FCNs-BiGRU of the present invention, it was tested and compared with other deep learning models. The same input samples were used to test four other methods, namely the fully convolutional network (FCN), the gated recurrent unit (GRU), the long short-term memory network (LSTM), and the parallel fully convolutional neural network-long short-term memory network (FCN-LSTM). In terms of the overall accuracy and the comparison results are shown in Table 1.

[0077] Table 1 Experimental results of the transient stability model

[0078] As can be seen from the table, compared with other models, the TSA model based on FCNs-BiGRU has more remarkable classification performance 、 up to 98.76% and 99.08% respectively, indicating that the classification performance meets the actual needs and conforms to the purpose to be achieved by the present invention.

Claims

1. The power system transient stability assessment method based on FCNs-BiGRU is characterized by: It includes the following steps: Step 1: Through time-domain simulation, batch simulation generation of samples is carried out to obtain the operation data set of the power system; Step 2: The improved empirical mode decomposition method IEMD is used to process the operation data set to establish the training data set of the transient stability assessment model; Step 3: Based on the training data set, the transient stability assessment model based on the improved fully convolutional neural network FCNs and bidirectional gated recurrent unit BiGRU is trained, and the trained transient stability assessment model is used for online transient stability assessment; Step 4: The phasor measurement unit PMU is used to obtain the required measurement data online and input it into the trained transient stability assessment model to output the transient stability assessment result.

2. The transient stability assessment method for power systems based on FCNs-BiGRU according to claim 1, wherein: In step 1, a real power system is built through time domain simulation, different fault times, fault locations and new energy output powers are set, and the time series data from before the fault occurs to after the fault is cleared in each simulation process is extracted. As the characteristic quantity, the transient stability label corresponding to the characteristic quantity is obtained after the simulation is completed, and different working conditions are simulated to finally generate an operation data set containing a large amount of operation status information of the power system.

3. The transient stability assessment method for power systems based on FCNs-BiGRU according to claim 2, characterized in that: In step 1, the voltage amplitude and phase angle of the busbar, and the active and reactive powers of the branches are specifically selected as characteristic quantities to form an operating data set ; denotes the characteristic quantity, denotes the transient stability label determined by the system, corresponding to the characteristic quantity ; before step 2, the characteristic quantity is also normalized.

4. The transient stability assessment method for power systems based on FCNs-BiGRU according to claim 3, characterized in that: The specific process of Step 2 is as follows: Step 2-1: First, for the normalized feature quantities in the time series data superimpose an independent white noise , which is expressed as follows: . After performing the Empirical Mode Decomposition (EMD) operation, it is decomposed into components and residuals . The number of decomposition times is K, and the decomposition result is shown as follows: ; Step 2-2: Calculate the corresponding mean value and residual of the component The mean value is calculated according to the following formula: , , where represents the mean value of the p-th component, and represents the mean value of the residual; Based on these mean values of the components and the mean value of the residuals, the processed time series data are obtained: ; The processed time series data and the corresponding transient stability labels form a training data set ; Step 2-3: Compare the total energy of all components after calculating the mean value to dynamically adjust the decomposition times K to maximize the total energy and optimize the decomposition process; all components and the energy ratio relationship with the time series data is as follows: , represents the total energy of all components when the decomposition times is K, represents the initial total energy, which is a fixed value. When the energy ratio reaches the maximum value, the total energy of the corresponding all components also reaches the maximum value, indicating that the decomposition times K is optimal.

5. The transient stability assessment method of the power system based on FCNs-BiGRU according to claim 4, characterized in that: In step 3, by improving the fully convolutional neural network FCNs and the bidirectional gated recurrent unit BiGRU, the spatial feature and the temporal relationship feature in are extracted respectively.

6. The transient stability assessment method for power systems based on FCNs-BiGRU according to claim 5, wherein The specific training process of the transient stability assessment model is as follows: Step 3-1: The improved fully convolutional neural network FCNs realizes dual-channel processing of samples by using two fully convolutional neural networks FCN in parallel. One channel processes the data to obtain time-domain spatial features, and the other channel first processes the data through the fast Fourier transform FFT to obtain frequency-domain spatial features; the convolutional kernels in FCNs are adjusted to one-dimensional; a global average pooling GAP layer is applied after the last convolutional layer; the channel attention mechanism is used, and SENet is embedded after each convolutional operation; Step 3-2: The transient stability assessment model includes FCN, BiGRU and a feature merging module. Two FCNs extract spatial features in the time-frequency domain through convolution, channel attention mechanism and pooling; The BiGRU module extracts the time-series relationship features; then, after integrating the spatial features and time-series relationship features through a fully connected layer, the Sigmoid function is used to map the result to the interval [0,1] to represent the probability of belonging to transient stability or transient instability; Step 3-3: According to the training data set obtained in Step 2, it is input into the transient stability assessment model for training to establish a transient stability assessment model based on FCNs-BiGRU.

7. The transient stability assessment method for power systems based on FCNs-BiGRU according to claim 1, characterized in that: In Step 4, based on the established transient stability assessment model, when applied online, the actual measurement data is input into the transient stability assessment model, and the transient stability assessment model outputs the assessment result, thereby obtaining the transient stability assessment result of the power system.

Images