A method for identifying power quality disturbances based on multi-source signal feature fusion

Abstract

The present invention relates to power quality disturbance identification, and in particular to a power quality disturbance identification method based on multi-source signal feature fusion, which includes collecting power quality disturbance signals, and corresponding voltage time series data and current time series data; generating corresponding grayscale images based on the collected voltage time series data and current time series data; performing image fusion on each grayscale image to obtain a two-dimensional feature map, and constructing an image data set; performing model training on a power quality disturbance identification model using the image data set to obtain a trained power quality disturbance identification model; and identifying voltage time series data and current time series data detected in real time using the trained power quality disturbance identification model to obtain a power quality disturbance identification result. The technical solution provided by the present invention can effectively overcome the defects of information loss and insufficient nonlinear feature capture capability existing in the prior art.

Description

Technical Field

[0001] The present invention relates to power quality disturbance identification, and in particular to a power quality disturbance identification method based on multi-source signal feature fusion. Background Art

[0002] With the large-scale development and utilization of new energy, the power system is facing the transformation from traditional centralized power generation to distributed power generation, which puts higher requirements on power quality. Distributed power generation has the characteristics of renewable energy, such as wind power, solar power, etc. Its output is unstable, random and intermittent, which leads to various power quality disturbances in the power grid, affecting the normal operation of power equipment, causing equipment damage or system failure. The widespread use of high-power motor equipment in industrial production has generated a large number of nonlinear, impact and asymmetric loads, causing the power quality problems of the power grid to become increasingly prominent.

[0003] At present, the identification of power quality disturbances at home and abroad mainly includes two steps: one is the extraction of power quality disturbance features, and the other is the design of classifiers. The traditional power quality disturbance feature extraction method is based on signal processing and mainly relies on time-frequency analysis. Commonly used methods include short-time Fourier transform, wavelet transform, discrete wavelet transform combined with fast Fourier transform, empirical mode decomposition, S transform, improved S transform, etc. However, these methods also have some defects. For example, feature extraction through empirical mode decomposition has problems such as modal errors and modal overlap, which affects its advantages in time-frequency analysis characteristics. In addition, feature extraction is performed through S transform, which is a time-frequency analysis tool based on Gaussian window function, which realizes the simultaneous observation of the energy distribution of a signal in the time and frequency range. Although it can identify power quality disturbances, the algorithm is relatively complex.

[0004] After obtaining the features, they need to be input into the classifier. Common methods for classifier design include support vector machine (SVM), BP (Back Propagation) neural network, probabilistic neural network (PNN), extreme learning machine (ELM), etc. However, the generalization ability of the above classifier design methods is limited, and the speed of processing large-scale data sets is slow.

[0005] Given the complexity of today's power systems, traditional power quality disturbance detection methods can no longer meet the requirements. With the continuous development of power systems, the accumulated power quality data samples are increasing, and deep learning networks have been applied in the field of power quality disturbance identification. By adopting an end-to-end approach and using deep learning networks for feature extraction and classification, the problems of insufficient feature extraction and feature grouping redundancy in traditional methods are effectively solved, and the recognition errors that may be caused by human experience in the feature extraction process are avoided.

[0006] In the study of power quality disturbance identification, the deep learning method mainly converts the one-dimensional signal into a two-dimensional image, and then uses the deep learning model to perform image classification to realize the automatic identification of power quality disturbance. For example, the method based on Markov transfer field converts the power quality disturbance data into an image modality, but it is only based on the voltage signal characteristics and has limited recognition ability. In addition, there is also a method in the prior art that converts the power quality disturbance signal into a trajectory circle image in a polar coordinate system for identification, but for non-periodic disturbance signals, polar coordinate conversion may lead to information loss.

[0007] In summary, existing technical solutions have certain limitations when processing power quality disturbance signals. Researchers often only focus on the single feature extraction of voltage signals and fail to fully explore the complex relationship between voltage and current, which leads to problems such as information loss and insufficient ability to capture nonlinear features. Summary of the invention

[0008] 1. Technical issues to be resolved

[0009] In view of the above-mentioned shortcomings of the prior art, the present invention provides a power quality disturbance identification method based on multi-source signal feature fusion, which can effectively overcome the defects of the prior art such as information loss and insufficient nonlinear feature capture capability.

[0010] (II) Technical solution

[0011] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0012] A method for identifying power quality disturbances based on multi-source signal feature fusion includes the following steps:

[0013] S1. Collect power quality disturbance signals, as well as corresponding voltage time series data and current time series data;

[0014] S2. Generate a corresponding grayscale image based on the collected voltage time series data and current time series data;

[0015] S3, performing image fusion on each grayscale image to obtain a two-dimensional feature map and construct an image data set;

[0016] S4. Using the image data set to perform model training on the power quality disturbance recognition model to obtain a trained power quality disturbance recognition model;

[0017] S5. Use the trained power quality disturbance identification model to identify the voltage time series data and current time series data detected in real time to obtain the power quality disturbance identification result.

[0018] Preferably, in S2, generating a corresponding grayscale image based on the collected voltage time series data and current time series data includes:

[0019] The collected voltage time series data and current time series data are converted based on the relative position matrix to generate corresponding voltage RPM grayscale image, current RPM grayscale image and current PMS grayscale image.

[0020] Preferably, in S3, each grayscale image is fused to obtain a two-dimensional feature map, and an image data set is constructed, including:

[0021] The voltage RPM grayscale image, the current RPM grayscale image and the current PMS grayscale image are fused according to the RGB channels to obtain a two-dimensional feature map;

[0022] The two-dimensional feature map is marked based on the collected power quality disturbance signal, and an image dataset is constructed.

[0023] Preferably, in S4, the image data set is used to perform model training on the power quality disturbance recognition model to obtain a trained power quality disturbance recognition model, including:

[0024] Divide the image dataset into training set, validation set and test set according to the preset ratio;

[0025] Build the ECA-ResNet neural network model and initialize the model parameters;

[0026] The training set is input into the ECA-ResNet neural network model for model training, and the validation set is used to optimize the model parameters of the ECA-ResNet neural network model until the model converges;

[0027] The test set is input into the ECA-ResNet neural network model to evaluate the generalization ability of the model, and finally a trained power quality disturbance identification model is obtained.

[0028] Preferably, in S5, the trained power quality disturbance identification model is used to identify the voltage time series data and the current time series data detected in real time to obtain the power quality disturbance identification result, including:

[0029] Real-time detection to obtain voltage time series data and current time series data;

[0030] Based on the relative position matrix, the voltage time series data and the current time series data detected in real time are converted to generate a voltage RPM grayscale map, a current RPM grayscale map and a current PMS grayscale map;

[0031] The voltage RPM grayscale image, the current RPM grayscale image and the current PMS grayscale image are fused according to the RGB channels to obtain a two-dimensional feature map;

[0032] The trained power quality disturbance recognition model is used to identify the two-dimensional feature map to obtain the power quality disturbance recognition result.

[0033] (III) Beneficial effects

[0034] Compared with the prior art, the power quality disturbance identification method based on multi-source signal feature fusion provided by the present invention has the following beneficial effects:

[0035] 1) The relative position information of the disturbance signal is introduced so that the power quality disturbance identification model can better capture the spatiotemporal relationship of the power quality disturbance, thereby effectively improving the accuracy of identification;

[0036] 2) By extracting and fusing the voltage signal features and the current signal features, it can effectively solve the problems of information loss, insufficient ability to capture nonlinear features, and weak ability to identify the critical states of certain transient disturbances caused by focusing only on the single feature extraction of the voltage signal and failing to fully explore the complex relationship between voltage and current. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0038] Figure 1 It is a schematic diagram of the process of the present invention;

[0039] Figure 2 It is a schematic diagram of obtaining a two-dimensional feature map by performing image fusion on each grayscale image in the present invention;

[0040] Figure 3 A model training loss curve diagram of the power quality disturbance identification model in the present invention;

[0041] Figure 4 It is a recognition accuracy curve diagram of the power quality disturbance recognition model in the present invention. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] A method for identifying power quality disturbances based on multi-source signal feature fusion, such as Figure 1 As shown, ① collect power quality disturbance signals, as well as corresponding voltage time series data and current time series data.

[0044] ② Generate corresponding grayscale images based on the collected voltage time series data and current time series data, including:

[0045] The collected voltage time series data and current time series data are converted based on the relative position matrix to generate corresponding voltage RPM grayscale image, current RPM grayscale image and current PMS grayscale image.

[0046] For sinusoidal signals, the grayscale value of the image after conversion based on the relative position matrix reflects the relative position relationship between each point and other points in the time series data, that is, the local characteristics of the original time series data; the texture characteristics of the image reflect the change trend between different regions in the time series data, that is, the global characteristics of the original time series data; the shape characteristics of the image reflect the structural characteristics such as periodicity, symmetry, monotonicity in the time series data, that is, the high-level characteristics of the original time series data. Therefore, converting a one-dimensional signal into a two-dimensional image based on the relative position matrix can provide rich feature information for subsequent image analysis and recognition.

[0047] ③ Such as Figure 2 As shown, each grayscale image is fused to obtain a two-dimensional feature map, and an image data set is constructed, which specifically includes:

[0048] The voltage RPM grayscale image, the current RPM grayscale image and the current PMS grayscale image are fused according to the RGB channels to obtain a two-dimensional feature map;

[0049] The two-dimensional feature map is marked based on the collected power quality disturbance signal, and an image dataset is constructed.

[0050] ④ Use the image data set to train the power quality disturbance recognition model to obtain a trained power quality disturbance recognition model, which specifically includes:

[0051] Divide the image dataset into training set, validation set and test set according to the preset ratio;

[0052] Build the ECA-ResNet neural network model and initialize the model parameters;

[0053] The training set is input into the ECA-ResNet neural network model for model training, and the validation set is used to optimize the model parameters of the ECA-ResNet neural network model until the model converges;

[0054] The test set is input into the ECA-ResNet neural network model to evaluate the generalization ability of the model, and finally a trained power quality disturbance identification model is obtained.

[0055] Overfitting is a common problem in machine learning. Common methods to prevent overfitting include data enhancement, regularization, cross-validation, early stopping, dropout, etc. In the technical solution of this application, the data enhancement method is used to perform scale transformation and random center rotation on the input two-dimensional feature map, and early stopping is also set in the model to solve the overfitting problem.

[0056] ⑤ Use the trained power quality disturbance identification model to identify the real-time detected voltage time series data and current time series data to obtain the power quality disturbance identification results, including:

[0057] Real-time detection to obtain voltage time series data and current time series data;

[0058] Based on the relative position matrix, the voltage time series data and the current time series data detected in real time are converted to generate a voltage RPM grayscale map, a current RPM grayscale map and a current PMS grayscale map;

[0059] The voltage RPM grayscale image, the current RPM grayscale image and the current PMS grayscale image are fused according to the RGB channels to obtain a two-dimensional feature map;

[0060] The trained power quality disturbance recognition model is used to identify the two-dimensional feature map to obtain the power quality disturbance recognition result.

[0061] Experimental analysis

[0062] 1) Composition of power quality disturbance sample library

[0063] According to IEEE Std 1159-2019

[20] , the circuit was simulated using Simulink in the Matlab environment, and sample data was obtained through the Workspace module to generate a transient power quality disturbance data set with a fundamental frequency of 50 Hz. The data set contains five different types of disturbance signals, namely: voltage sag, voltage swell, voltage interruption, transient oscillation, and impulse. For each type of disturbance signal, 1000 groups of samples were generated. In order to ensure the integrity of the disturbance information in each group of samples, the sampling time of each group of samples was 10 cycles, with a total of 1024 data points. After data enhancement processing, the image size of the voltage RPM grayscale image, current RPM grayscale image, and current PMS grayscale image input into the ECA-ResNet18 neural network model is 224×224.

[0064] 2) Evaluation indicators

[0065] In order to verify the effectiveness of various recognition models, the technical solution of this application uses the following performance indicators for evaluation: Accuracy, Precision, Recall and F1-score. Accuracy is the ratio of the number of samples correctly identified by the classifier in all recognitions to the total number of samples; Precision is the ratio of the number of samples correctly identified as positive by the classifier to the number of all samples identified as positive by the classifier; Recall is the ratio of the number of samples correctly identified as positive by the classifier to the number of actual positive samples; F1-score is the harmonic mean of precision and recall.

[0066] 3) Experimental setup

[0067] Using the transient power quality disturbance data set, the recognition effects of the ECA-ResNet18+RPM of the technical solution of this application on five different types of disturbance signals are compared, as shown in the following table:

[0068] Table 1. Recognition effect of ECA-ResNet18+RPM on 5 different types of disturbance signals

[0069]

[0070] Using the transient power quality disturbance data set, the recognition effects of ECA-ResNet18+RPM, 2D-ResNet+GDAF and ResNet+visualized trajectory circle of the technical solutions of this application are compared, as shown in the following table:

[0071] Table 2 Recognition effect of different recognition methods

[0072]

[0073] As can be seen from Table 2, the technical solution ECA-ResNet18+RPM of this application has a certain degree of improvement in the evaluation indicators of recall rate, precision, accuracy and F1 score compared with the ECA-ResNet18+RPM and 2D-ResNet+GDAF recognition methods, indicating that the current signal characteristics enhance the expression of disturbance characteristics in the identification of power quality disturbances. Therefore, the ECA-ResNet18+RPM technical solution adopted in this application can effectively solve the problems of information loss and insufficient nonlinear feature capture ability, and realize accurate identification of power quality disturbances.

[0074] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying power quality disturbances based on multi-source signal feature fusion, characterized in that: The following steps are involved: S1. Collect power quality disturbance signals, as well as corresponding voltage time series data and current time series data; S2. Generate a corresponding grayscale image based on the collected voltage time series data and current time series data; S3, performing image fusion on each grayscale image to obtain a two-dimensional feature map and construct an image data set; S4. Using the image data set to perform model training on the power quality disturbance recognition model to obtain a trained power quality disturbance recognition model; S5. Using the trained power quality disturbance identification model to identify the voltage time series data and the current time series data detected in real time, to obtain a power quality disturbance identification result; In S2, the corresponding grayscale image is generated based on the collected voltage time series data and current time series data, including: The collected voltage time series data and current time series data are converted based on the relative position matrix to generate corresponding voltage RPM grayscale image, current RPM grayscale image and current PMS grayscale image.

2. The method for identifying power quality disturbances based on multi-source signal feature fusion according to claim 1 is characterized in that: In S3, each grayscale image is fused to obtain a two-dimensional feature map, and an image dataset is constructed, including: The voltage RPM grayscale image, the current RPM grayscale image and the current PMS grayscale image are fused according to the RGB channels to obtain a two-dimensional feature map; The two-dimensional feature map is marked based on the collected power quality disturbance signal, and an image dataset is constructed.

3. The method for identifying power quality disturbances based on multi-source signal feature fusion according to claim 2 is characterized in that: In S4, the image data set is used to train the power quality disturbance recognition model to obtain a trained power quality disturbance recognition model, including: Divide the image dataset into training set, validation set and test set according to the preset ratio; Build the ECA-ResNet neural network model and initialize the model parameters; The training set is input into the ECA-ResNet neural network model for model training, and the validation set is used to optimize the model parameters of the ECA-ResNet neural network model until the model converges; The test set is input into the ECA-ResNet neural network model to evaluate the generalization ability of the model, and finally a trained power quality disturbance identification model is obtained.

4. The method for identifying power quality disturbances based on multi-source signal feature fusion according to claim 3 is characterized in that: In S5, the trained power quality disturbance identification model is used to identify the voltage time series data and current time series data detected in real time to obtain the power quality disturbance identification results, including: Real-time detection to obtain voltage time series data and current time series data; Based on the relative position matrix, the voltage time series data and the current time series data detected in real time are converted to generate a voltage RPM grayscale map, a current RPM grayscale map and a current PMS grayscale map; The voltage RPM grayscale image, the current RPM grayscale image and the current PMS grayscale image are fused according to the RGB channels to obtain a two-dimensional feature map; The trained power quality disturbance recognition model is used to identify the two-dimensional feature map to obtain the power quality disturbance recognition result.

Images