An image segmentation-based power system model abnormal data cleaning method

Abstract

The application provides a power system model abnormal data cleaning method based on image segmentation, and the method comprises the following steps: step one, performing multi-scale analysis and processing on the characteristics of the power system model by constructing a multi-scale space to determine the optimal scale of the most stable and information-rich shape representation; step two, performing double-feature extraction on the WPC image to comprehensively describe the morphological characteristics of the power system model; step three, constructing a deformable WPC shape model which can adapt to the complex deformation of the power system model; and step four, cleaning the abnormal data in the model by using the image segmentation technology to remove noise and abnormal points. The application can effectively remove noise and abnormal points, improve the accuracy and reliability of the data, and has high feasibility and applicability in the aspect of power system model abnormal data cleaning.

Description

Technical Field

[0001] This invention relates to the field of carbon emission reduction technology in power systems, and in particular to a method for cleaning abnormal data in power system models based on image segmentation. Background Technology

[0002] Power system data processing and analysis is a crucial research area in the power sector. Its primary purpose is to monitor the real-time operating status of power systems, ensuring their stability and security. Power system models are an essential component of power system data processing and analysis; they are used to describe the structure and characteristics of power systems and to predict the operating status of various components within the system.

[0003] Because power system models contain a large amount of data, including many noise and outliers, these data can affect the accuracy and reliability of the power system models, thereby affecting the accuracy of power system operation status monitoring and analysis results.

[0004] Therefore, how to clean and process outlier data in power system models has become an important research topic in the power field. Developing a new algorithm for cleaning outlier data in power system models is therefore highly necessary and has significant application value. Summary of the Invention

[0005] This invention proposes a method for cleaning outlier data in power system models based on image segmentation. Addressing the problems of low efficiency and difficulty in handling large-scale data in traditional power system outlier data processing methods, this invention constructs a multi-scale space to determine the optimal scale for extracting the most stable and information-rich shape representation. A deformable WPC shape model is then built, and image segmentation technology is used to clean outlier data in the power system model. This effectively removes noise and outliers, improving the accuracy and reliability of the data. The method described in this invention has high feasibility and applicability in cleaning outlier data in power system models.

[0006] The present invention adopts the following technical solution.

[0007] A method for cleaning abnormal data in a power system model based on image segmentation, the method comprising the following steps;

[0008] Step 1: Analyze and process the power system model features at multiple scales by constructing a multi-scale space to determine the optimal scale for extracting the most stable and information-rich shape representation.

[0009] Step 2: Perform dual feature extraction on the WPC (Wavelet Packet Transform) image to comprehensively describe the morphological features of the power system model;

[0010] Step 3: Construct a deformable WPC shape model that can adapt to the complex deformation of the power system model;

[0011] Step 4: Clean up abnormal data in the model using image segmentation techniques to remove noise and outliers.

[0012] In step two, the WPC algorithm is used to decompose and reconstruct the power system model image to extract information of different frequencies of the power system model, that is, to use the WPC data binarization method to extract morphological features and text features.

[0013] The WPC data binarization method is performed using the following formula;

[0014]

[0015]

[0016] Where p max p min v max v min Let x represent the maximum and minimum values ​​of power and wind speed in the power model, respectively. i ,y i ) represents the data pixel index, (v i ,p i ) represents the corresponding power system data point; if it contains at least one data point, the pixel is set to 1 (data pixel, represented by DP); if it contains no data points, the pixel is set to 0 (background pixel); Δ x and Δ y It is the scaling factor used for binarization, and the image size M i ×N i In fact, it is determined by the scaling factor (Δx, Δy), and the specific representation is as follows:

[0017]

[0018] In step three, an algorithm based on a deformation model is used to adjust the WPC shape model according to the complex deformation of the power system model; it employs a quantitative method to calculate the optimal scale, and different scales used for binarization will produce binary images with different structures; specifically;

[0019] A binary image is generated at a predefined small scale. A multi-scale space is then constructed by filtering the image using a Gaussian kernel function, which is formulated as follows:

[0020]

[0021] Where a and b are the width and height of the Gaussian kernel window; g(x,y) is the sum of the products of the filter coefficients and the image pixels enclosed by the filter, where the filter is represented as...

[0022]

[0023] When performing dual feature extraction on WPC images, assume that outlier data pixels are scattered or unevenly distributed in the image space, and the data pixel density R c and divergence L c Within an n×n moving window, for each data pixel and composite measure T c , is defined as a descriptor for the spatial distribution properties of data pixels;

[0024] R c It is obtained by calculating the ratio of data pixels in the moving window, and the calculation method is as follows;

[0025]

[0026] Where Ω is the set of all data pixels in the binary WPC image, Ω c A moving window of size n×n centered on each data pixel p; for any pixel in Ω c For pixel c in the data, if the pixel is a data pixel, then φ(c) = 1, otherwise it is 0;

[0027] The spatial distribution description parameters of the data pixels within the moving window include R. c and data pixel divergence L c Data pixel divergence L c The calculation method is as follows:

[0028]

[0029] Where pi(xi,yi) defines any data pixel in Ωc; x m y m For x i y i The mean; M is Ω c The number of data pixels in the data. For each data pixel, the composite measure T c Calculated using the following formula

[0030]

[0031] Where w1 and w2 are R c and L c The weight.

[0032] In step four, an image segmentation algorithm is used to accurately locate and clean up abnormal data in the power system model, removing noise and outliers to improve data accuracy and reliability. Specifically:

[0033] Step A1: Use an image segmentation algorithm to segment the power system model data into different regions or parts;

[0034] Step A2: Identify and locate abnormal data by extracting and comparing features from the segmented data; the abnormal data includes noise, outliers, missing data, and data whose features differ from those of normal data.

[0035] Step A3: Based on the detection results of abnormal data, clean and repair the abnormal data using an abnormal data cleaning algorithm. The goal of cleaning is to remove abnormal data, while the goal of repair is to fill in data that may be missing or damaged. The specific cleaning method is to remove abnormal data based on its characteristics, and the specific repair method is to use interpolation methods, substitution rules, and statistical inference methods to estimate and fill in the data.

[0036] Step A4: Verify and evaluate the cleaned and repaired data to ensure its accuracy and reliability. This involves checking whether the data has been accurately cleaned and repaired and is consistent with other data. Specifically, this can be done by comparing the data with the original data, comparing it with other independent data sources, and validating the model.

[0037] The image segmentation algorithm uses a pixel similarity-based clustering algorithm to divide the pixels in the power system model image into several different regions during the image segmentation process.

[0038] The clustering algorithm, based on the color and texture features of pixels, groups similar pixels into the same region to achieve segmentation of the power system model.

[0039] The abnormal data cleaning algorithm includes a shape matching algorithm and an anomaly detection algorithm, specifically:

[0040] Shape matching algorithm: used to measure the shape similarity of multiple extracted regions, and then determine the target shape of the power system model; the shape matching algorithm uses template matching and contour matching methods to calculate the similarity between each region and the expected shape, and selects the most matching shape as the target shape;

[0041] Anomaly detection algorithm: used for post-processing of segmentation results to identify and remove anomalies in power system model images; the anomaly detection algorithm is based on the gradient and color information features of the image, and uses statistical methods or machine learning algorithms to determine whether a pixel is an anomaly, so as to improve the cleaning effect of abnormal data.

[0042] This invention addresses the problems of low efficiency and difficulty in handling large-scale data in traditional power system anomaly data processing methods. It constructs a multi-scale space to determine the optimal scale for extracting the most stable and information-rich shape representation, thus building a deformable WPC shape model. Image segmentation technology is then used to clean up anomaly data in the power system model, effectively removing noise and outliers, and improving data accuracy and reliability. This algorithm demonstrates high feasibility and applicability in anomaly data cleaning for power system models.

[0043] The core technology of this invention combines multi-scale spatial and WPC image dual-feature extraction with deformable WPC shape models and image segmentation techniques to achieve effective cleaning of anomalous data in power system models. This algorithm boasts high accuracy and reliability, improving the efficiency and precision of power system data processing and analysis.

[0044] The beneficial effects of this invention are:

[0045] 1. Anomaly cleaning algorithms can effectively remove noise and outliers from power system models, improving data accuracy and reliability. Cleaned data more accurately reflects the operating status of the power system, helping to monitor system health in real time, promptly identify potential problems, and take appropriate measures to maintain stable power system operation.

[0046] 2. By cleaning and optimizing power system model data, the structure and characteristics of the power system can be accurately analyzed, potential fault risks can be identified, and the reliability and lifespan of power equipment can be predicted. Early detection of system vulnerabilities and safety hazards helps in the development of corresponding protection strategies and emergency measures, thereby improving the safety and robustness of the power system.

[0047] 3. This invention can quickly and accurately remove abnormal data by automating the processing of large-scale power system data. This will greatly improve the efficiency of data processing and analysis, saving manpower and time costs. Accurate data helps to more precisely model and simulate power systems, enabling the formulation of optimization strategies and the rational allocation of resources, thereby improving the operating efficiency and energy utilization efficiency of the power system.

[0048] 4. Cleaned model data can be used for analysis and decision-making in multiple fields such as power system state assessment, prediction, and optimization, helping to improve the accuracy and precision of decision-making. For example, using cleaned data for power system fault diagnosis can improve the accuracy and speed of fault diagnosis. Furthermore, due to the high efficiency and accuracy of anomaly data cleaning algorithms, the time and cost of manual data cleaning can be significantly reduced, avoiding the need for a large amount of manpower required by traditional data cleaning methods.

[0049] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0050] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0051] Appendix Figure 1 This is a schematic diagram of the algorithm framework for cleaning abnormal data in a power system model based on image segmentation, as presented in this invention.

[0052] Appendix Figure 2 This is a schematic diagram of the process for generating a deformable WPC shape model based on dual-feature images according to the present invention. Detailed Implementation

[0053] The present invention will be further described below with reference to specific embodiments. The accompanying drawings are for illustrative purposes only, representing schematic diagrams rather than actual physical objects, and should not be construed as limiting the scope of this patent. To better illustrate the embodiments of the present invention, some components in the drawings may be omitted, enlarged, or reduced, and do not represent the actual dimensions of the product. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0054] As shown in the figure, a method for cleaning abnormal data in a power system model based on image segmentation is described, the method comprising the following steps;

[0055] Step 1: Analyze and process the power system model features at multiple scales by constructing a multi-scale space to determine the optimal scale for extracting the most stable and information-rich shape representation.

[0056] Step 2: Perform dual feature extraction on the WPC (Wavelet Packet Transform) image to comprehensively describe the morphological features of the power system model;

[0057] Step 3: Construct a deformable WPC shape model that can adapt to the complex deformation of the power system model;

[0058] Step 4: Clean up abnormal data in the model using image segmentation techniques to remove noise and outliers.

[0059] In step two, the WPC algorithm is used to decompose and reconstruct the power system model image to extract information of different frequencies of the power system model, that is, to use the WPC data binarization method to extract morphological features and text features.

[0060] The WPC data binarization method is performed using the following formula;

[0061]

[0062]

[0063] Where p max p min v max v min Let x represent the maximum and minimum values ​​of power and wind speed in the power model, respectively. i ,y i ) represents the data pixel index, (v i ,p i ) represents the corresponding power system data point; if it contains at least one data point, the pixel is set to 1 (data pixel, represented by DP); if it contains no data points, the pixel is set to 0 (background pixel); Δ x and Δ y It is the scaling factor used for binarization, and the image size M i ×N i In fact, it is determined by the scaling factor (Δx, Δy), and the specific representation is as follows:

[0064]

[0065] In step three, an algorithm based on a deformation model is used to adjust the WPC shape model according to the complex deformation of the power system model; it employs a quantitative method to calculate the optimal scale, and different scales used for binarization will produce binary images with different structures; specifically;

[0066] A binary image is generated at a predefined small scale. A multi-scale space is then constructed by filtering the image using a Gaussian kernel function, which is formulated as follows:

[0067]

[0068] Where a and b are the width and height of the Gaussian kernel window; g(x,y) is the sum of the products of the filter coefficients and the image pixels enclosed by the filter, where the filter is represented as...

[0069]

[0070] When performing dual feature extraction on WPC images, assume that outlier data pixels are scattered or unevenly distributed in the image space, and the data pixel density R c and divergence L c Within an n×n moving window, for each data pixel and composite measure T c, is defined as a descriptor for the spatial distribution properties of data pixels;

[0071] R c It is obtained by calculating the ratio of data pixels in the moving window, and the calculation method is as follows;

[0072]

[0073] Where Ω is the set of all data pixels in the binary WPC image, Ω c A moving window of size n×n centered on each data pixel p; for any pixel in Ω c For pixel c in the data, if the pixel is a data pixel, then φ(c) = 1, otherwise it is 0;

[0074] The spatial distribution description parameters of the data pixels within the moving window include R. c and data pixel divergence L c Data pixel divergence L c The calculation method is as follows:

[0075]

[0076] Where pi(xi,yi) defines any data pixel in Ωc; x m y m For x i y i The mean; M is Ω c The number of data pixels in the data. For each data pixel, the composite measure T c Calculated using the following formula

[0077]

[0078] Where w1 and w2 are R c and L c The weight.

[0079] In step four, an image segmentation algorithm is used to accurately locate and clean up abnormal data in the power system model, removing noise and outliers to improve data accuracy and reliability. Specifically:

[0080] Step A1: Use an image segmentation algorithm to segment the power system model data into different regions or parts;

[0081] Step A2: Identify and locate abnormal data by extracting and comparing features from the segmented data; the abnormal data includes noise, outliers, missing data, and data whose features differ from those of normal data.

[0082] Step A3: Based on the detection results of abnormal data, clean and repair the abnormal data using an abnormal data cleaning algorithm. The goal of cleaning is to remove abnormal data, while the goal of repair is to fill in data that may be missing or damaged. The specific cleaning method is to remove abnormal data based on its characteristics, and the specific repair method is to use interpolation methods, substitution rules, and statistical inference methods to estimate and fill in the data.

[0083] Step A4: Verify and evaluate the cleaned and repaired data to ensure its accuracy and reliability. This involves checking whether the data has been accurately cleaned and repaired and is consistent with other data. Specifically, this can be done by comparing the data with the original data, comparing it with other independent data sources, and validating the model.

[0084] The image segmentation algorithm uses a pixel similarity-based clustering algorithm to divide the pixels in the power system model image into several different regions during the image segmentation process.

[0085] The clustering algorithm, based on the color and texture features of pixels, groups similar pixels into the same region to achieve segmentation of the power system model.

[0086] The abnormal data cleaning algorithm includes a shape matching algorithm and an anomaly detection algorithm, specifically:

[0087] Shape matching algorithm: used to measure the shape similarity of multiple extracted regions, and then determine the target shape of the power system model; the shape matching algorithm uses template matching and contour matching methods to calculate the similarity between each region and the expected shape, and selects the most matching shape as the target shape;

[0088] Anomaly detection algorithm: used for post-processing of segmentation results to identify and remove anomalies in power system model images; the anomaly detection algorithm is based on the gradient and color information features of the image, and uses statistical methods or machine learning algorithms to determine whether a pixel is an anomaly, so as to improve the cleaning effect of abnormal data.

Claims

1. A method for cleaning abnormal data in a power system model based on image segmentation, characterized in that: The method includes the following steps; Step 1: Analyze and process the power system model features at multiple scales by constructing a multi-scale space to determine the optimal scale for extracting the most stable and information-rich shape representation. Step 2: Perform dual feature extraction on the WPC image to comprehensively describe the morphological features of the power system model; Step 3: Construct a deformable WPC shape model that can adapt to the complex deformation of the power system model; Step 4: Clean up abnormal data in the model using image segmentation techniques to remove noise and outliers; In step two, the WPC algorithm is used to decompose and reconstruct the power system model image in order to extract information of different frequencies of the power system model, that is, to use the WPC data binarization method to extract morphological features and text features. The WPC data binarization method is performed using the following formula; Formula 1; Image size M i ×N i It is determined by the scaling factor, and is expressed as follows: Formula 2; Where p max p min V represents the maximum and minimum power values ​​in the power model, respectively. max v min Let x and y represent the maximum and minimum wind speeds in the power model, respectively. i ,y i ) represents the data pixel index, (v i ,p i () represents the corresponding power system data point; if it contains at least one data point, the pixel is set to 1; if it contains no data point, the pixel is set to 0. and It is a scaling factor used for binarization. In step three, an algorithm based on a deformation model is used to adjust the WPC shape model according to the complex deformation of the power system model; it employs a quantitative method to calculate the optimal scale, and different scales used for binarization will produce binary images with different structures; specifically; A binary image is generated at a predefined small scale. A multi-scale space is then constructed by filtering the image using a Gaussian kernel function, which is formulated as follows: Formula 3; Where a and b are the width and height of the Gaussian kernel function windows; It is the sum of the products of the filter coefficients and the image pixels enclosed by the filter, where the filter is represented as... Formula 4; When performing dual feature extraction on WPC images, assume that outlier data pixels are non-uniformly distributed in the image space, and the data pixel density R... c and divergence L c Within an n×n moving window, for each data pixel and composite measure T c , is defined as a descriptor for the spatial distribution properties of data pixels; R c It is obtained by calculating the ratio of data pixels in the moving window, and the calculation method is as follows; Formula 5; in It is the collection of all data pixels in a binary WPC image. A moving window of size n×n centered on each data pixel p; for any... In the context of pixel c, if the pixel is a data pixel, then... Otherwise, it is 0; The spatial distribution description parameters of the data pixels within the moving window include R. c and data pixel divergence L c , Data pixel divergence L c The calculation method is as follows: Formula Six; in Defined Any data pixel in the x m y m For x i y i The mean; M is The number of data pixels in the data. ; For each data pixel, the composite measure T c Calculated using the following formula Formula 7; Where w1 and w2 are R c and L c The weights; In step four, an image segmentation algorithm is used to accurately locate and clean up abnormal data in the power system model, removing noise and outliers to improve data accuracy and reliability. Specifically: Step A1: Use an image segmentation algorithm to segment the power system model data into different regions; Step A2: Identify and locate abnormal data by extracting and comparing features from the segmented data; the abnormal data includes noise, outliers, missing data, and data whose features differ from those of normal data. Step A3: Based on the detection results of abnormal data, clean and repair the abnormal data using an abnormal data cleaning algorithm. The goal of cleaning is to remove abnormal data, while the goal of repair is to fill in data that may be missing or damaged. The specific cleaning method is to remove abnormal data based on its characteristics, and the specific repair method is to use interpolation methods, substitution rules, and statistical inference methods to estimate and fill in the data. Step A4: Verify and evaluate the cleaned and repaired data to ensure its accuracy and reliability. This involves verifying whether the data has been accurately cleaned, repaired, and is consistent with other data. Specifically, this can be achieved through comparison with the original data, comparison with other independent data sources, and model validation.

2. The method for cleaning abnormal data in a power system model based on image segmentation according to claim 1, characterized in that: The image segmentation algorithm uses a pixel similarity-based clustering algorithm to divide the pixels in the power system model image into several different regions during the image segmentation process. The clustering algorithm, based on the color and texture features of pixels, groups similar pixels into the same region to achieve segmentation of the power system model.

3. The method for cleaning abnormal data in a power system model based on image segmentation according to claim 1, characterized in that: The abnormal data cleaning algorithm includes a shape matching algorithm and an anomaly detection algorithm, specifically: Shape matching algorithm: used to measure the shape similarity of multiple extracted regions, and then determine the target shape of the power system model; the shape matching algorithm uses template matching and contour matching methods to calculate the similarity between each region and the expected shape, and selects the most matching shape as the target shape; Anomaly detection algorithm: used for post-processing of segmentation results to identify and remove anomalies in power system model images; the anomaly detection algorithm is based on the gradient and color information features of the image, and uses statistical methods or machine learning algorithms to determine whether a pixel is an anomaly, so as to improve the cleaning effect of abnormal data.

Images