Power Grid Disturbance Identification Method and System Based on Improved FastICA

Abstract

The present invention discloses a power grid disturbance identification method and system based on improved FastICA. The method includes: setting power grid disturbance types of different types and at different bus positions based on the structure of the target power grid and the parameters of its various components, respectively simulating them to obtain the change data of the voltage amplitudes of each bus in the target power grid under different disturbance types, that is, disturbance data; using the improved fast independent component analysis algorithm to denoise the disturbance data; extracting the features of the denoised disturbance data; generating a disturbance identification sample data set from the extracted features; constructing a power grid disturbance identification model; training the power grid disturbance identification model through the disturbance identification sample data set to obtain a trained power grid disturbance identification model; and using the trained power grid disturbance identification model to identify power grid disturbances and outputting a disturbance identification result including the disturbance type and the disturbance occurrence position. The present invention has advantages such as good denoising effect and high disturbance identification accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid disturbance identification, and particularly to a power grid disturbance identification method and system based on improved FastICA. Background Art

[0002] With the large-scale penetration of non-traditional power grid entities such as renewable energy, power electronic devices, and controllable loads into the power grid, the operation scenarios of the power system are more complex. Traditional power grid disturbance identification methods based on power grid models may be difficult to adapt to the new characteristics of modern power systems, resulting in deteriorated effects. With the high degree of intelligence and automation of the power system, the amount of power data that can be obtained has increased significantly. Using data-driven methods that do not rely on traditional models to identify power grid disturbance signals helps to obtain higher identification accuracy.

[0003] Data-driven disturbance identification methods refer to using intelligent algorithms to autonomously learn from a large amount of power data to obtain corresponding characteristic parameters and the laws between different signal types, and then identifying the disturbance types. In existing research, the commonly used power grid disturbance signal type identification algorithms mainly include support vector machines, deep neural networks, and ensemble learning algorithms, etc. The support vector machine algorithm is mainly applied to pattern recognition and linear classification of small samples and non-linear samples. However, it is extremely susceptible to kernel function parameters and requires an optimization algorithm to find the best parameter combination to obtain better classification effects. Neural networks are widely used in the field of power grid disturbance signal identification due to their excellent non-linear function approximation and pattern recognition capabilities. However, their training duration and accuracy depend on the quantity and variety of sample data, and their generalization ability is average. With the continuous innovation of artificial intelligence technology, more and more intelligent algorithms have been gradually applied: some research is based on time-frequency analysis to extract signal features, and then uses decision tree algorithms to identify disturbance signals.

[0004] Through the analysis and performance comparison of data-driven disturbance identification methods, it is found that data-driven artificial intelligence algorithms can more accurately identify disturbance types. However, there has been less research on denoising in the case of high noise levels in the past; and currently, there are few studies on using intelligent algorithms to simultaneously identify and locate disturbance sources. In summary, it is necessary to carry out research on data-driven power grid disturbance identification methods considering high data noise. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a power grid disturbance identification method based on improved FastICA with excellent denoising performance and high identification accuracy in the case of high noise.

[0006] To solve the above technical problem, the present invention provides a power grid disturbance identification method based on improved FastICA, including:

[0007] Using Matlab / Simulink software, based on the structure of the target power grid and the parameters of its various components, different types of power grid disturbances at different bus positions are set, and simulations are carried out on them respectively to obtain the data of the voltage amplitude changes of each bus in the target power grid under different disturbance types, that is, disturbance data;

[0008] The improved fast independent component analysis algorithm is used to denoise the disturbance data;

[0009] Extract the features of the denoised disturbance data;

[0010] Generate a disturbance identification sample data set from the extracted features;

[0011] Construct a power grid disturbance identification model;

[0012] The power grid disturbance identification model is trained through the disturbance identification sample data set to obtain a trained power grid disturbance identification model;

[0013] The trained power grid disturbance identification model is used for power grid disturbance identification, and the disturbance identification results including the disturbance type and the disturbance occurrence location are output.

[0014] Furthermore, the improved fast independent component analysis algorithm is used to denoise the disturbance data, including:

[0015] The fast independent component analysis algorithm takes the maximum negative entropy as the objective function, and separates the noisy disturbance data into the original disturbance data and the noise data. The linear mathematical model of independent component analysis is:

[0016]

[0017] where s is composed of n independent source signals s i which are statistically independent of each other; A is composed of n components a i and is the mixing matrix;

[0018] In order to make the original disturbance data and the noise data satisfy the condition of mutual statistical independence, the original data X′ is centralized:

[0019] X = X′ - E[X′], E[s] = A -1 E[X]

[0020] And whitening:

[0021] Satisfy Q is the whitening matrix;

[0022] Take G(v) = tanh(v), and the negative entropy approximation formula is:

[0023]

[0024] Among them, is the output variable with zero mean and unit variance, and W is the separation matrix; is a Gaussian random variable, with the mean and unit variance being the same as ;

[0025] The signal amplitude ratio α and the similarity coefficient ζ are introduced ij as the constraints and judgment conditions of the algorithm:

[0026]

[0027]

[0028] Among them, y i is the initial output signal; y i,new is the final output signal after amplitude transformation; s is the source signal; N is the signal data length, that is, the number of sampling points; if the similarity coefficient ζ ij is closer to 1, it means that the i-th separated signal is more similar to the j-th source signal, that is, the separation effect of the algorithm is more ideal; if ζ ij approaches zero or is far from 1, it means that the separation is not completed and the signal needs to be separated again.

[0029] Furthermore, the discrete wavelet transform is used to extract the features of the denoised disturbance data, and the discrete wavelet transform is calculated by the following formula:

[0030] DWT h (a wt , b wt ) = 2 -j / 2 ∫2 -j / 2 h(t)ψ(2 -j t - k)dt

[0031] Among them, h(t) is the input signal, a wt and b wt are the scale factor and displacement factor of h(t), and ψ(t) is the wavelet function;

[0032] The db6 wavelet function is selected to construct the discrete wavelet transform algorithm, and the disturbance signal is decomposed into 7 layers, and eight characteristic quantities are selected to extract the features of the disturbance data, namely the maximum value of the fundamental frequency amplitude, the minimum value of the fundamental frequency amplitude, the mean value of the fundamental frequency amplitude, the standard deviation of the fundamental frequency amplitude, the amplitude difference of the fundamental frequency amplitude, the duration of the transient process, the energy E5 of the wavelet coefficients of the 5th layer, and the energy E6 of the wavelet coefficients of the 6th layer.

[0033] Further, for the characteristic sample data of different types and at different buses when the disturbance occurs, 75% is randomly selected as the training samples, and the remaining 25% is used as the test samples. The label is set as the corresponding disturbance type and the bus where the disturbance occurs; all the training samples form the training set, and all the test samples form the test set.

[0034] Further, the power grid disturbance identification model is a multi-label random forest classification model, and the power grid disturbance is identified through the trained multi-label random forest classification model.

[0035] To solve the above technical problems, the present invention further provides a power grid disturbance identification system based on improved FastICA for implementing the above power grid disturbance identification method, which includes:

[0036] A disturbance data acquisition module, which uses Matlab / Simulink software to set different types of power grid disturbances at different bus positions based on the structure of the target power grid and the parameters of its various components, and simulates them respectively to obtain the data of the voltage amplitude change of each bus of the target power grid under different disturbance types;

[0037] A denoising module, which uses an improved fast independent component analysis algorithm to denoise the disturbance data;

[0038] A feature extraction module, which is used to extract the features of the denoised disturbance data;

[0039] A data set generation module, which is used to generate a disturbance identification sample data set from the extracted features;

[0040] An identification model construction module, which is used to construct a power grid disturbance identification model;

[0041] An identification model training module, which is used to train the constructed power grid disturbance identification model;

[0042] A disturbance identification module, which uses the trained power grid disturbance identification model to identify the power grid disturbance and obtains the disturbance identification result including the disturbance type and the disturbance occurrence position.

[0043] To solve the above technical problems, the present invention further provides a computer device, which includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of the above power grid disturbance identification method are implemented.

[0044] To solve the above technical problems, the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above power grid disturbance identification method are implemented.

[0045] The beneficial effects of implementing the present invention are as follows:

[0046] 1. For the noisy perturbation data, an improved fast independent component analysis is used for denoising, which greatly improves the denoising effect.

[0047] 2. A multi-label random forest classification model is adopted for perturbation identification. In the case of combining with the improved fast independent component analysis for denoising, the accuracy of perturbation identification is greatly improved. Description of the Drawings

[0048] Figure 1 It is the principle flowchart of a power grid perturbation identification method based on improved FastICA of the present invention;

[0049] Figure 2 It is the principle flowchart of the multi-label random forest classification model in the power grid perturbation identification method based on improved FastICA of the present invention;

[0050] Figure 3 It is the connection block diagram of a power grid perturbation identification system based on improved FastICA of the present invention. Detailed Embodiment

[0051] To make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the drawings. It is hereby declared that the directional terms such as up, down, left, right, front, back, inside, outside, etc. that appear or will appear in the text of the present invention are only based on the drawings of the present invention, and they do not specifically limit the present invention.

[0052] As Figure 1 shown, this embodiment provides a power grid perturbation identification method based on improved FastICA, including the following steps:

[0053] S1. Using Matlab / Simulink software, based on the structure of the target power grid and the parameters of its various components, set the power grid perturbation types of different types and different bus positions, and simulate them respectively to obtain the target power grid bus voltage amplitude change data under different perturbation types, that is, perturbation data;

[0054] S2. Using the improved fast independent component analysis algorithm to denoise the perturbation data:

[0055] The fast independent component analysis algorithm takes the maximum negative entropy as the objective function, separates the noisy perturbation data into the original perturbation data and the noise data, and the linear mathematical model of independent component analysis is:

[0056]

[0057] where s is composed of n independent source signals s icomponents, which are statistically independent of each other; A consists of n components a i constitutes the mixing matrix;

[0058] To make the original perturbation data and noise data satisfy the condition of statistical independence from each other, the original data X′ is centered:

[0059] X = X′ - E[X'], E[s] = A -1 E[X]

[0060] and whitened:

[0061] satisfying Q is the whitening matrix;

[0062] Taking G(v) = tanh(v), the negative entropy approximation formula is:

[0063]

[0064] where, is the output variable with zero mean and unit variance, and W is the separation matrix; is a Gaussian random variable, with the mean and unit variance the same as ;

[0065] The signal amplitude ratio α and the similarity coefficient ζ are introduced ij as the constraints and decision conditions of the algorithm:

[0066]

[0067]

[0068] where, y i is the initial output signal; y i,new is the final output signal after amplitude transformation; s is the source signal; N is the signal data length, i.e., the number of sampling points; if the similarity coefficient ζ ij is closer to 1, it means that the i-th separated signal is more similar to the j-th source signal, that is, the separation effect of the algorithm is more ideal; if ζ ij approaches zero or is far from 1, it means that the separation is not completed and the signal needs to be separated again.

[0069] S3. Extract the features of the denoised perturbation data using the discrete wavelet transform, and the discrete wavelet transform is calculated by the following formula:

[0070] DWT h (a wt ,b wt ) = 2 -j / 2 ∫2 -j / 2 h(t)ψ(2 -j t - k)dt

[0071] Among them, h(t) is the input signal, and a wt and b wt are the scale factor and displacement factor of h(t), and ψ(t) is the wavelet function;

[0072] Select the db6 wavelet function to construct the discrete wavelet transform algorithm, decompose the disturbance signal into 7 layers, and select eight characteristic quantities to extract the characteristics of the disturbance data, namely the maximum value of the fundamental frequency amplitude, the minimum value of the fundamental frequency amplitude, the average value of the fundamental frequency amplitude, the standard deviation of the fundamental frequency amplitude, the amplitude difference of the fundamental frequency amplitude, the duration of the transient process, the energy E5 of the wavelet coefficients at the 5th layer, and the energy E6 of the wavelet coefficients at the 6th layer.

[0073] S4. Generate a disturbance identification sample data set from the extracted features;

[0074] In this step, for the characteristic sample data of different types and different buses when the disturbance occurs, randomly select 75% as the training samples, and the remaining 25% as the test samples, and set the label as the corresponding disturbance type and the bus where the disturbance occurs; all the training samples form the training set, and all the test samples form the test set.

[0075] S5. Construct a multi-label random forest classification model as the power grid disturbance identification model;

[0076] S6. Train the power grid disturbance identification model through the disturbance identification sample data set to obtain the trained power grid disturbance identification model;

[0077] S7. Use the trained power grid disturbance identification model to identify the power grid disturbance, and output the disturbance identification result including the disturbance type and the location where the disturbance occurs.

[0078] The working principle of the multi-label random forest classification model is as follows Figure 2 as shown.

[0079] To prove the superiority of the present invention and objectively and truly compare the denoising performances of three denoising methods, the average mean square error values after denoising the disturbance signals (s1 to s10) with different noise intensities by fast independent component analysis and improved fast independent component analysis in 200 different experiments are given in Table 1. It can be seen from Table 1 that the MSE values of the improved FastICA algorithm are all smaller than those of the original FastICA algorithm, verifying that the denoising performance of the improved FastICA algorithm is better, and at the same time verifying the effectiveness of introducing the amplitude ratio α and the similarity coefficient ζ ij to solve the uncertainty problem inherent in the original FastICA.

[0080] Table 1 Comparison of denoising effects of FastICA and improved FastICA under different noise intensities

[0081]

[0082] Considering the characteristics of different disturbance signal data, a simulation model of the IEEE standard 3-machine 9-bus system is built using the MATLAB / Simulink system to generate disturbance data of three-phase short circuit, single-phase short circuit to ground, two-phase short circuit to ground, two-phase short circuit, and large induction motor self-start at different buses, and the multi-label random forest classification model proposed by the present invention is trained. During testing, the Rank-SVM and ML-KNN, two common multi-label classification methods, are respectively used to classify and predict the sample data, and the results are compared with those of this method, as shown in Table 2.

[0083] Table 2 Comparison of classification results between the present invention and RankSVM and ML-KNN

[0084]

[0085] To implement the above method, this embodiment also provides a power grid disturbance identification system based on improved FastICA, as Figure 3 shown, which includes:

[0086] A disturbance data acquisition module, which is used to use Matlab / Simulink software to set different types of power grid disturbances at different bus positions based on the structure of the target power grid and the parameters of its various components, and simulate them respectively to obtain the data of the voltage amplitude change of each bus of the target power grid under different disturbance types;

[0087] A denoising module, which uses an improved fast independent component analysis algorithm to denoise the disturbance data;

[0088] A feature extraction module, which is used to extract the features of the denoised disturbance data;

[0089] A data set generation module, which is used to generate a disturbance identification sample data set from the extracted features;

[0090] An identification model construction module, which is used to construct a power grid disturbance identification model;

[0091] An identification model training module, which is used to train the constructed power grid disturbance identification model;

[0092] A disturbance identification module, which uses the trained power grid disturbance identification model to identify power grid disturbances and obtains disturbance identification results including disturbance types and disturbance occurrence positions.

[0093] In addition, this embodiment also provides a computer device and a computer-readable storage medium.

[0094] The computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of the above-mentioned power grid disturbance identification method are implemented.

[0095] A computer-readable storage medium stores a computer program thereon. When the computer program is executed by a processor, the steps of the above-mentioned power grid disturbance identification method are implemented.

[0096] Although the description of the present disclosure has been quite detailed and several of the described embodiments have been described in particular, it is not intended to be limited to any of these details or embodiments or any particular embodiment, but rather should be regarded as providing a broad interpretation of these claims in view of the prior art by reference to the appended claims, so as to effectively cover the intended scope of the present disclosure. In addition, the present disclosure has been described above in terms of embodiments foreseeable by the inventors for the purpose of providing a useful description, and those non-substantive modifications to the present disclosure that are not currently foreseeable may still represent equivalent modifications of the present disclosure.

Claims

1. A power grid disturbance identification method based on improved FastICA, characterized in that, Including: Using Matlab / Simulink software, based on the structure of the target power grid and the parameters of its various components, set different types of power grid disturbances at different bus positions, and simulate them respectively to obtain the voltage amplitude change data of each bus in the target power grid under different disturbance types, that is, disturbance data; Denoise the disturbance data using an improved fast independent component analysis algorithm; Extract the features of the denoised disturbance data; Generate a disturbance identification sample data set from the extracted features; Construct a power grid disturbance identification model; Train the power grid disturbance identification model through the disturbance identification sample data set to obtain a trained power grid disturbance identification model; Use the trained power grid disturbance identification model to identify power grid disturbances, and output disturbance identification results including disturbance types and disturbance occurrence positions; Denoise the disturbance data using an improved fast independent component analysis algorithm, including: The fast independent component analysis algorithm takes the maximum negative entropy as the objective function, separates the noisy disturbance data into the original disturbance data and noise data, and the linear mathematical model of independent component analysis is: where s consists of n independent source signals s i which are statistically independent of each other; A consists of n components a i and is the mixing matrix; In order to make the original disturbance data and noise data satisfy the condition of mutual statistical independence, centralize the original data X'; X = X' - E[X'], E[s] = A -1 E[X] And whiten it; Meet Q is a whitening matrix; Take G(v)=tanh(v), and the negative entropy approximation formula is: wherein, is an output variable with zero mean and unit variance, and W is a separation matrix; is a Gaussian random variable, with the mean and unit variance being the same as those of ; Introduce the signal amplitude ratio α and the similarity coefficient ζ ij As the constraints and judgment conditions of the algorithm: where y i is the initial output signal; y i,new is the final output signal after amplitude transformation; s is the source signal; N is the signal data length, i.e., the number of sampling points; if the similarity coefficient ζ ij is closer to 1, it indicates that the i-th separated signal is more identical to the j-th source signal, that is, the separation effect of the algorithm is more ideal; if ζ ij approaches zero or is far from 1, it indicates that the separation is not completed and the signal needs to be separated again.

2. The power grid disturbance identification method based on improved FastICA according to claim 1, characterized in that Extract the features of the denoised disturbance data using discrete wavelet transform, and the discrete wavelet transform is calculated by the following formula: DWT h (a wt ,b wt ) = 2 -j / 2 ∫2 -j / 2 h(t)ψ(2 -j t - k)dt where h(t) is the input signal, a wt and b wt are the scale factor and displacement factor of h(t), respectively, and ψ(t) is the wavelet function; Select the db6 wavelet function to construct a discrete wavelet transform algorithm, decompose the disturbance signal into 7 layers, and select eight characteristic quantities to extract the features of the disturbance data, namely the maximum value of the fundamental frequency amplitude, the minimum value of the fundamental frequency amplitude, the average value of the fundamental frequency amplitude, the standard deviation of the fundamental frequency amplitude, the amplitude difference of the fundamental frequency amplitude, the duration of the transient process, the energy E5 of the wavelet coefficients at the 5th layer, and the energy E6 of the wavelet coefficients at the 6th layer.

3. The power grid disturbance identification method based on improved FastICA according to claim 1, wherein For the extracted characteristic sample data of different types and at different buses when the disturbance occurs, randomly select 75% as training samples, and the remaining 25% as test samples, and set the label as the corresponding disturbance type and the bus where the disturbance occurs; all training samples constitute a training set, and all test samples constitute a test set.

4. The power grid disturbance identification method based on improved FastICA according to claim 1, characterized in that The power grid disturbance identification model is a multi-label random forest classification model, and the trained multi-label random forest classification model is used to identify power grid disturbances.

5. A power grid disturbance identification system based on improved FastICA, characterized in that it is used to implement the power grid disturbance identification method based on improved FastICA according to any one of claims 1-4, and it includes: A disturbance data acquisition module, which is used to use Matlab / Simulink software, based on the structure of the target power grid and the parameters of its various components, set different types of power grid disturbances at different bus positions, and simulate them respectively to obtain the voltage amplitude change data of each bus in the target power grid; A denoising module, which uses an improved fast independent component analysis algorithm to denoise the disturbance data; A feature extraction module, which is used to extract the features of the denoised disturbance data; A data set generation module, which is used to generate a disturbance identification sample data set from the extracted features; An identification model construction module for constructing a power grid disturbance identification model; An identification model training module for training the constructed power grid disturbance identification model; A disturbance identification module that uses the trained power grid disturbance identification model to identify power grid disturbances and obtains a disturbance identification result including the disturbance type and the disturbance occurrence location.

Images