Power system stability evaluation method based on ensemble learning anti-noise

Abstract

The application provides a power system stability evaluation method based on integrated learning anti-noise. According to different noise interference conditions in a power system, a plurality of stability evaluation models considering anti-noise and a noise identification model are trained, and are combined into an integrated learning model, so that the accuracy of power system stability evaluation under complex noise interference is improved.

Description

Technical Field

[0001] This invention belongs to the field of power system and stability assessment technology, specifically relating to a power system stability assessment method based on ensemble learning noise immunity. Background Technology

[0002] In power systems, faults occur very rapidly. Without timely control measures, large-scale power outages can occur, leading to significant social harm. Simultaneously, erroneous operations can exacerbate power system instability. Therefore, improving the accuracy of power system stability assessment is of paramount importance. With the development of power system stability measurement units (PMUs) in power systems, data mining-based machine learning has gradually become a research hotspot, offering high accuracy and fast computation speed. This process involves training with a large number of power system fault samples to find the mapping relationship between sample characteristics and stability conditions. When PMU measurement data is input, online assessments can be performed by building a model.

[0003] The accuracy of stability assessment depends on the reliability of the acquired data. However, PMU measurements can be affected by time synchronization or phase. Furthermore, data transmitted to the control center may be subject to interference from different noise distributions due to cyberattacks or communication transmission issues. Models trained offline often struggle to adapt to complex real-world data disturbances, thus impacting stability assessment results. Summary of the Invention

[0004] To overcome the shortcomings and deficiencies of existing technologies, this invention provides a power system stability assessment method based on ensemble learning for noise resistance. According to different noise interference situations occurring in the power system, multiple targeted stability assessment models considering noise resistance and a noise identification model are trained and combined into an ensemble learning model to improve the accuracy of power system stability assessment under complex noise interference. The specific steps are as follows:

[0005] 1. A large number of training samples are obtained through simulation. The feature quantities of each bus and the stability results of the samples are extracted as the raw data.

[0006] 2. Add n different noise interferences to the original data to construct n different noise datasets, thereby simulating the complex noise situation in reality.

[0007] 3. Based on n different noise datasets, n targeted noise-resistant stable evaluation models are trained using a Sparse Stacked Denoising Autoencoder (SSDAE). A noise recognition model is trained using a Support Vector Machine (SVR) based on the fit to the actual noise distribution. Then, the targeted stable evaluation models and the noise recognition model are combined into an adaptive stable evaluation model that considers noise resistance.

[0008] 4. By inputting the measured feature data into the combined evaluation model, it can adaptively output stable evaluation results with high confidence.

[0009] The specific technical solution adopted by this invention to solve its technical problem is as follows:

[0010] A power system stability assessment method based on ensemble learning noise resistance, characterized in that:

[0011] Based on different noise interference scenarios in the power system, multiple targeted stability assessment models considering noise resistance and a noise identification model are trained and combined into an ensemble learning model to improve the accuracy of power system stability assessment under complex noise interference. The specific steps include:

[0012] Step S1: Obtain training samples through simulation, extract the feature quantities of each bus, and use the stability results of the samples as raw data;

[0013] Step S2: Add n different noise interferences to the original data to construct n different noise datasets, thereby simulating the actual noise situation;

[0014] Step S3: Based on n different noise datasets, train n targeted noise-resistant stable evaluation models using a sparse stacked noise reduction autoencoder (SSDAE); train a noise recognition model using a support vector machine (SVR) based on the fit to the actual noise distribution; then, combine the targeted stable evaluation models and the noise recognition model into an adaptive stable evaluation model that considers noise resistance.

[0015] Step S4: Input the measured feature data into the combined evaluation model obtained in step S3 to adaptively output a stable evaluation result with high confidence.

[0016] Further, in step S1, information including voltage, voltage phase angle and frequency deviation of each bus is extracted, and the feature quantities at the time of fault occurrence, the time of fault clearance and 10 times after fault clearance, as well as the stability results of the sample are used as raw data.

[0017] Furthermore, in step S2, n different Gaussian white noises are added to the original data to construct noise data to simulate the real noise distribution, and different noise interference intensities are represented by different signal-to-noise ratios.

[0018] Furthermore, in step S3, n stable evaluation models for noise immunity against different noise interferences are trained based on n sets of different noise data. Using the SSDAE algorithm Perform training; train the i-th model M t i The characteristic is noisy data, as shown in equation (1);

[0019]

[0020] in, For the i-th group of noise data; x o This is the original data; The noise is Gaussian white noise in the i-th group;

[0021] For n M t The i-th model It targets the i-th group of noisy data. It has targeted noise reduction effects; during training, it needs to be improved. To improve the comprehensive evaluation capability of the ensemble model, the generalization ability is assessed under the remaining n-1 noise distributions. Firstly, the generalization ability is assessed using the remaining n-1 groups of noise data, excluding the i-th group. The stability labels of the samples are used to pre-train the SSDAE neural network, thereby enabling... It can fit this series of noisy data; the input data for the k-th training iteration is in the following training data format:

[0022]

[0023] in, This is the k-th group of noisy data, used as the input data for the model; y o The true and stable evaluation labels for the samples are used as output data for the model.

[0024] Then, through the i-th group of noise data SSDAE was trained in a targeted manner; the training set is shown below:

[0025]

[0026] in, y is the i-th set of noisy data, used as the input data for the model; o The true and stable evaluation labels for the samples are used as output data for the model.

[0027] Used to train the i-th model The feature set is shown in equation (4);

[0028]

[0029] For n Training was performed using different noisy datasets and real stable evaluation labels. The training set is shown in equation (5);

[0030]

[0031] Among them, T i for The training set; Let y be the i-th feature set; o For accurate and stable assessment labels;

[0032] Using the neural network structure of the SSDAE algorithm, and employing a sparsity reduction method, it is trained with n different training sets to obtain n... If noisy data is input, n It can output n stable evaluation results with values ​​between 0 and 1.

[0033] Furthermore, in step S3, a noise recognition model M is trained using the noise data from the training set. R1 and M R2 ;

[0034] Among them, M R1 The output set of weight values ​​reflects the actual noise distribution; each weight value is assigned one-to-one to n noise levels. The output stable evaluation result; M is obtained by training n sets of noisy data and corresponding n sets of weight labels. R1 ; in M R1 During the training process, it is necessary to construct a mapping relationship between noisy data and weight values; M R1 The training set is shown in equation (6);

[0035]

[0036] Where H is M R1 The training set; This is the i-th group of noise datasets; Let M be the weight label corresponding to the i-th group of noisy data; by training the neural network structure of the SVR algorithm using the training set H, we can obtain M. R1 ;

[0037] M R2 It can output the signal-to-noise ratio of noise, reflecting the actual noise distribution; M is obtained by training from n sets of noise data and corresponding n sets of signal-to-noise ratio labels. R2 ; in M R2 During the training process, it is necessary to establish a mapping relationship between noisy data and noise signal-to-noise ratio; M R2 The training set is shown in equation (7);

[0038]

[0039] Where G is M R2 The training set; For the i-th group of noise dataset; SNR iLet M be the signal-to-noise ratio label corresponding to the i-th group of noise data; train the neural network structure of the SVR algorithm using the training set G to obtain M. R2 .

[0040] Furthermore, in step S4, n items are... An M R1 And an M R2 The data are combined to construct a power system stability assessment model based on ensemble learning data to resist noise; when actual characteristic data subject to noise interference is input, the n data in the ensemble model... And an M R1 M R2 The model outputs results simultaneously; n of them. Output n evaluation results (R1, R2, ..., Rn) with values ​​between 0 and 1. n M R1 Output a set of weight values ​​(w1, w2, ..., w n M R2 Output the noise signal-to-noise ratio (SNR); then calculate the weight values ​​(w1, w2, ..., w n Assign n evaluation results (R1, R2, ..., R...) n ), and with M R2 The sub-model R with the closest signal-to-noise ratio SNR The output results are combined to adaptively obtain the final stable evaluation result R. CM Weighted stability assessment result R CM As shown in equation (8);

[0041]

[0042] Among them, R i for Output result; w i To give M R1 The weight values ​​of the output results, R SNR To be with M R2 The sub-model R with the closest signal-to-noise ratio result SNR The output result.

[0043] Compared with the prior art, the present invention and its preferred embodiments have the following beneficial effects:

[0044] 1. The provided noise identification model can evaluate the real noise distribution through two noise assessment models, resulting in more accurate results.

[0045] 2. Construct a stable evaluation model based on ensemble learning data denoising, which can adaptively output stable evaluation results with high confidence under complex noise interference.

[0046] 3. When training the sub-model, the sub-model was first trained using noise data of other intensities, and then trained using the selected noise data. This can ensure the sub-model's recovery performance for specific noises and improve its generalization ability to noises of other intensities. Attached Figure Description

[0047] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0048] Figure 1 This is the wiring diagram for the IEEE 39-node system.

[0049] Figure 2 This is a flowchart illustrating the overall process of an embodiment of the present invention. Detailed Implementation

[0050] To make the features and advantages of this patent more apparent and understandable, specific embodiments are provided below for detailed explanation:

[0051] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0052] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0053] like Figure 2 As shown, this embodiment of the invention provides a power system stability assessment method based on ensemble learning noise immunity. According to different noise interference situations in the power system, multiple targeted stability assessment models that consider data noise immunity and a noise identification model are trained and combined into an ensemble learning model, thereby improving the accuracy of power system stability assessment under complex noise interference.

[0054] The technical solutions adopted in the embodiments of the present invention are as follows:

[0055] 1) Obtain a large number of training samples through simulation based on different fault information. Extract information such as voltage, voltage phase angle and frequency deviation of each bus, and use the feature quantities at the time of fault occurrence, the time of fault clearance and 10 time points after fault clearance, as well as the stability results of the samples as the raw data.

[0056] 2) Add n different sets of Gaussian white noise to the original data to construct noise data to simulate the real noise distribution. Different noise interference intensities are represented by different signal-to-noise ratios. Train n stable evaluation models for noise immunity against different noise interferences based on the n different noise data sets. Using the SSDAE algorithm Training is performed. The i-th model M is trained. t i The characteristic is that the data is noisy, as shown in Equation (1).

[0057]

[0058] in, For the i-th group of noise data; x o This is the original data; Let be the Gaussian white noise of the i-th group.

[0059] For n M t The i-th model It targets the i-th group of noisy data. It has a targeted noise reduction effect. During training, it needs to be improved. The generalization ability under the remaining n-1 noise distributions is used to improve the overall evaluation capability of the ensemble model. First, the generalization ability is improved by considering the remaining n-1 groups of noise data, excluding the i-th group. The stability labels of the samples are used to pre-train the SSDAE neural network, thereby enabling... It can fit this series of noisy data. The input data for the k-th training iteration is in the following training data format:

[0060]

[0061] in, This is the k-th group of noisy data, used as the input data for the model; y o The true and stable evaluation labels for the samples are used as output data for the model.

[0062] Then, through the i-th group of noise data Targeted training was performed on SSDAE. The training set is shown below:

[0063]

[0064] in, y is the i-th set of noisy data, used as the input data for the model; o The true and stable evaluation labels for the samples are used as output data for the model.

[0065] Used to train the i-th model The feature set is shown in equation (4).

[0066]

[0067] For n The training was performed using different noisy datasets and real stable evaluation labels. The training set is shown in equation (5).

[0068]

[0069] Among them, T i for The training set; Let y be the i-th feature set; o This is a true stability assessment label.

[0070] By using the neural network structure of the SSDAE algorithm and employing sparsity processing, and training with n different training sets, n algorithms can be obtained. If noisy data is input, n It can output n stable evaluation results with values ​​between 0 and 1.

[0071] 4) Train the noise recognition model M using the noise data in the training set. R1 and M R2 .

[0072] M R1 The output provides a set of weight values ​​that reflect the actual distribution of noise. Each weight value is assigned one-to-one to one of the n noise levels. The output stable evaluation result. M is obtained by training n sets of noisy data and corresponding n sets of weight labels. R1 In M R1 During the training process, it is necessary to establish a mapping relationship between noisy data and weight values. R1 The training set is shown in equation (6).

[0073]

[0074] Where H is M R1 The training set; This is the i-th group of noise datasets; Let M be the weight label corresponding to the i-th group of noisy data. By training the neural network structure of the SVR algorithm using the training set H, we can obtain M. R1 .

[0075] M R2 It can output the signal-to-noise ratio (SNR) of noise, reflecting the actual noise distribution. M is obtained by training with n sets of noise data and corresponding n sets of SNR labels. R2 In M R2During the training process, it is necessary to establish a mapping relationship between noisy data and noise signal-to-noise ratio. R2 The training set is shown in equation (7).

[0076]

[0077] Where G is M R2 The training set; For the i-th group of noise dataset; SNR i Let M be the signal-to-noise ratio label corresponding to the i-th group of noise data. By training the neural network structure of the SVR algorithm using the training set G, we can obtain M. R2 .

[0078] 5) Take n items An M R1 And an M R2 The data are combined to construct a power system stability assessment model based on ensemble learning data to resist noise. When actual characteristic data subject to noise interference is input, the n data in the ensemble model of this invention... And an M R1 M R2 The model outputs results simultaneously. Among them, n... Output n evaluation results (R1, R2, ..., Rn) with values ​​between 0 and 1. n M R1 Output a set of weight values ​​(w1, w2, ..., w n M R2 Output the noise signal-to-noise ratio (SNR). Then, calculate the weight values ​​(w1, w2, ..., w). n Assign n evaluation results (R1, R2, ..., R...) n ), and with M R2 The sub-model R with the closest signal-to-noise ratio SNR The output results are combined to adaptively obtain the final stable evaluation result R. CM The weighted stability assessment result R CM As shown in equation (8).

[0079]

[0080] Among them, R i for Output result; w i To give M R1 The weight values ​​of the output results; R SNR To be with M R2 The sub-model R with the closest signal-to-noise ratio result SNR The output result; λ is M R1 The weight values ​​of the output results. The following specific application example further demonstrates the effectiveness of the solution of this invention.

[0081] The simulation software used was PSD-BPA, and the test system employed the IEEE-39-bus system. The system operated at five load levels: 90%, 95%, 100%, 105%, and 110%. Faults occurred at 10%, 50%, and 90% load on each line. All faults were three-phase short circuits. Fault durations ranged from 6 cycles to 18.5 cycles, increasing in 0.5-cycle intervals, for a total of 26 variations. Figure 1 As shown, a total of 15,012 samples were generated in the IEEE-39 node system, of which 7,821 were stable and 7,191 were unstable. Samples were extracted from these 15,012 fault samples in a 1:1 ratio to form training and test sets. Gaussian white noise interference was added to the training set with signal-to-noise ratios of 5dB, 10dB, 20dB, 30dB, and 50dB. Five targeted data denoising stability evaluation models (M1-M5) were trained according to the patented method and combined into an ensemble learning model.

[0082] Ten sets of test data were added to the test set with signal-to-noise ratios of 5dB, 7dB, 10dB, 15dB, 20dB, 25dB, 30dB, 35dB, 40dB, and 50dB. These ten sets of test data were then input into five models, and the results are shown in Table 1.

[0083] Table 1. Stability evaluation results of different models

[0084]

[0085] As shown in Table 1, thanks to the noise recognition model's ability to fit complex noise, the ensemble learning model in this paper can output stable evaluation results with high accuracy under interference of different noise levels.

[0086] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0087] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0088] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0089] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0090] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

[0091] This patent is not limited to the above-described preferred embodiment. Anyone can derive other forms of power system stability assessment method based on ensemble learning noise reduction under the guidance of this patent. All equivalent changes and modifications made within the scope of this patent application shall fall within the scope of this patent.

Claims

1. A power system stability assessment method based on ensemble learning noise immunity, characterized in that: Based on different noise interference scenarios in the power system, multiple targeted stability assessment models considering noise resistance and a noise identification model are trained and combined into an ensemble learning model to improve the accuracy of power system stability assessment under complex noise interference. The specific steps include: Step S1: Obtain training samples through simulation, extract the feature quantities of each bus, and use the stability results of the samples as raw data; Step S2: Add n different noise interferences to the original data to construct n different noise datasets, and then simulate the actual noise situation; Step S3: Based on n different noise datasets, train n targeted noise-resistant stable evaluation models using a sparse stacked noise reduction autoencoder (SSDAE); train a noise recognition model using a support vector machine (SVR) based on the fit to the actual noise distribution; then, combine the targeted stable evaluation models and the noise recognition model into an adaptive stable evaluation model that considers noise resistance. Step S4: Input the measured feature data into the combined evaluation model obtained in step S3 to adaptively output a stable evaluation result with high confidence. In step S3, n stable evaluation models for noise immunity against different noise interferences are trained based on n sets of different noise data. ; using the SSDAE algorithm Perform training; train the i-th model. The characteristic is that the data is noisy, as shown in equation (1); （1） in, This represents the i-th group of noise data; This is the original data; The noise is Gaussian white noise in the i-th group; For n The i-th model It targets the i-th group of noisy data. It has targeted noise reduction effects; during training, it needs to be improved. To improve the comprehensive evaluation capability of the ensemble model, the generalization ability is assessed under the remaining n-1 noise distributions. Firstly, the generalization ability is assessed using the remaining n-1 groups of noise data, excluding the i-th group. The stability labels of the samples are used to pre-train the SSDAE neural network, thereby enabling... It can fit this series of noisy data; the input data for the k-th training iteration is in the following training data format: （2） in, This is the k-th group of noisy data, used as the input data for the model; The true and stable evaluation labels for the samples are used as output data for the model. Then, through the i-th group of noise data SSDAE was trained in a targeted manner; the training set is shown below: （3） in, It is the i-th set of noise data, which serves as the input data for the model; The true and stable evaluation labels for the samples are used as output data for the model. Used to train the i-th model The feature set is shown in equation (4); （4） For n They were trained using different noisy datasets and real stable evaluation labels, respectively. The training set is shown in equation (5); （5） in, for The training set; Let i be the feature set of the i-th group; For accurate and stable assessment labels; Using the neural network structure of the SSDAE algorithm, and employing a sparsity reduction method, it is trained with n different training sets to obtain n... If noisy data is input, n It can output n stable evaluation results with values ​​between 0 and 1; In step S3, a noise recognition model is trained using the noise data from the training set. and ; in, The output set of weight values ​​reflects the actual noise distribution; each weight value is assigned one-to-one to n noise levels. The output stable evaluation result is obtained by training n sets of noisy data and corresponding n sets of weight labels. ;exist During the training process, it is necessary to establish a mapping relationship between noisy data and weight values; The training set is shown in equation (6); （6） in, for The training set; This is the i-th group of noise datasets; The weight label is the corresponding weight of the i-th group of noisy data; through the training set By training the neural network structure of the SVR algorithm, we can obtain... ; It can output the signal-to-noise ratio (SNR) of noise, reflecting the actual noise distribution; it is obtained through training with n sets of noise data and corresponding n sets of SNR labels. ;exist During the training process, it is necessary to establish a mapping relationship between noisy data and noise signal-to-noise ratio; The training set is shown in equation (7); （7） in, for The training set; This is the i-th group of noise datasets; The signal-to-noise ratio label corresponding to the i-th group of noise data; through the training set The neural network structure of the SVR algorithm is trained to obtain... .

2. The power system stability assessment method based on ensemble learning noise immunity according to claim 1, characterized in that: In step S1, information including voltage, voltage phase angle and frequency deviation of each bus is extracted, and the feature quantities at the time of fault occurrence, the time of fault clearance and 10 times after fault clearance, as well as the stability results of the sample are used as raw data.

3. The power system stability assessment method based on ensemble learning noise immunity according to claim 1, characterized in that: In step S2, n different groups of Gaussian white noise are added to the original data to construct noise data to simulate the real noise distribution. Different noise interference intensities are represented by different signal-to-noise ratios.

4. The power system stability assessment method based on ensemble learning noise immunity according to claim 1, characterized in that: In step S4, n items are... ,one and By combining these data, a power system stability assessment model based on ensemble learning data with noise resistance can be constructed. When actual characteristic data subject to noise interference is input, the n data in the ensemble model... , and a , The model outputs results simultaneously; n of them. Output n evaluation results with values ​​between 0 and 1. ; Output a set of weight values , Output noise signal-to-noise ratio Then the weight values Assign n evaluation results and with Sub-model with the closest signal-to-noise ratio The output results are combined to adaptively obtain the final stable evaluation result. Weighted stability assessment results As shown in equation (8); （8） in, for Output results; To give The weight values ​​of the output results To and The sub-model with the closest signal-to-noise ratio result The output result.

Images