Load Identification Method and System Based on Voltage-Current Trajectory Images

Abstract

The present disclosure relates to a load recognition method and system based on voltage-current trajectory images. The recognition method converts the voltage-current trajectory image into a color image by performing color coding on the voltage-current trajectory image, constructs a cubical complex based on the voltage-current trajectory image and the color image, then extracts topological features on the cubical complex, splices the topological features with the trajectory features to obtain spliced features, and finally screens out load features, and obtains a load recognition result relying on the screened load features. Based on the voltage-current trajectory image, this solution constructs a cubical complex through the trajectory image and the color image to filter the data in the image to obtain topological features, combines the trajectory features and the topological features to obtain spliced features, enriches and expands the feature space from more levels, so that the finally obtained load features can more comprehensively express the characteristics of the corresponding load, and improves the accuracy of load recognition.

Description

Technical Field

[0001] The present disclosure belongs to the field of load identification, and particularly relates to a load identification method, system, electronic device and storage medium based on voltage-current trajectory images. Background Art

[0002] With the development of smart grids, the energy utilization efficiency in the power generation, transmission, distribution, and consumption links of the power system has been increasingly improved, and it has become more and more important to analyze the electricity consumption patterns of consumers and provide energy-saving guidance. However, as the terminal of electric energy transmission, the power management system on the user side has not yet been developed maturely. Among them, the lack of sub-item electricity consumption statistical data has become one of the reasons restricting the development of this system.

[0003] When managing the electricity consumption of the user side, a relatively simple and direct common method is to set an intrusive device that can sample at an interval of several seconds or a higher sampling rate to periodically determine the total household power consumption. However, in actual scenarios, it is neither convenient nor more expensive to monitor the electricity consumption in this way, and it is impossible to achieve sub-item electricity metering. The electricity consumption structure on the user side is not clear, and it is impossible to manage the electricity consumption behavior in a targeted manner. Summary of the Invention

[0004] To solve the above problems, the present disclosure provides a load identification method, system, electronic device and storage medium based on voltage-current trajectory images. Enrich and expand the feature space from more levels, so that the finally obtained load features can more comprehensively express the characteristics of the corresponding load, and improve the accuracy of load identification.

[0005] To solve the above technical problems, a first aspect of the present invention proposes a load identification method based on voltage-current trajectory images, and the identification method includes:

[0006] Generating a voltage-current trajectory image based on the original steady-state voltage and current data; generating a color image by color coding the data in the voltage-current trajectory image;

[0007] Extracting the trajectory features in the voltage-current trajectory image, and respectively constructing a cubical complex based on the voltage-current trajectory image and the color image, and extracting the topological features of each cubical complex;

[0008] Stitching each of the trajectory features and each of the topological features respectively to generate a plurality of stitched features, and screening out load features from the stitched features according to the correlation between each of the stitched features;

[0009] Identifying each of the load features to obtain a load identification result.

[0010] According to a preferred embodiment of the present invention, generating a color image by color coding based on the data in the voltage-current trajectory image includes:

[0011] Determine the active voltage and active current according to the voltage-current trajectory image, and generate a first color coding representing the direction of the voltage-current trajectory based on the active voltage and the active current;

[0012] Determine the active power and apparent power according to the voltage-current trajectory image, and generate a second color coding representing the power factor based on the active power and the apparent power;

[0013] Generate binary images of multiple periods according to the voltage-current trajectory image, and generate a third color coding according to the average value of the pixel values in the binary images;

[0014] Generate the color image according to the first color coding, the second color coding and the third color coding.

[0015] According to a preferred embodiment of the present invention, constructing cubical complexes based on the voltage-current trajectory image and the color image respectively includes:

[0016] Use the pixel points in the voltage-current trajectory image and the color image as vertices respectively, and construct high-dimensional cubes between the vertices according to the pixel values of each pixel point;

[0017] Assemble the high-dimensional cubes corresponding to the voltage-current trajectory image to obtain the cubical complex corresponding to the voltage-current trajectory image;

[0018] Assemble the high-dimensional cubes corresponding to the color image to obtain the cubical complex corresponding to the color image.

[0019] According to a preferred embodiment of the present invention, extracting the topological features of each cubical complex includes:

[0020] Filter each cubical complex according to the pixel value size to obtain filtered cubical complexes corresponding to different pixel values;

[0021] Extract the Betti number, persistent amplitude and persistent entropy in the filtered cubical complex as the topological features; wherein, the Betti number is the number of points in the filtered cubical complex, the persistent amplitude is the farthest distance of the points in the filtered cubical complex from the diagonal line, and the persistent entropy is the Shannon entropy of the filtered cubical complex.

[0022] According to a preferred embodiment of the present invention, extracting the trajectory features in the voltage-current trajectory image includes:

[0023] Rotate the trajectory in the voltage-current trajectory image, and use the distance between the trajectories before and after rotation as the first trajectory feature;

[0024] And / or, calculate the area of the closed region between the trajectory in the voltage-current trajectory image and the coordinate axis travel as the second trajectory feature;

[0025] And / or, add a broken line to the trajectory in the voltage-current trajectory image to divide the trajectory into two parts, and use the degree of bending of the broken line as the third trajectory feature;

[0026] And / or, evenly divide the trajectory in the voltage-current trajectory image into three segments, and select the starting point and ending point of the middle segment to calculate the slope of the middle segment as the fourth trajectory feature;

[0027] And / or, evenly divide the trajectory in the voltage-current trajectory image into three segments, and use the peak value of the middle segment as the fifth trajectory feature;

[0028] And / or, calculate the active power and reactive power according to the voltage-current trajectory image as the sixth trajectory feature;

[0029] And / or, determine the power harmonics according to the voltage-current trajectory image as the seventh trajectory feature;

[0030] And / or, determine the original current data according to the voltage-current trajectory image as the eighth trajectory feature;

[0031] And / or, determine the power harmonics and the original current data according to the voltage-current trajectory image, and perform principal component analysis on the power harmonics and the original current data as the ninth trajectory feature.

[0032] According to a preferred embodiment of the present invention, the screening of the load characteristics from the splicing characteristics according to the correlation between the respective splicing characteristics includes:

[0033] Calculate the mutual information between the respective splicing characteristics as the correlation;

[0034] Compare the mutual information with a preset standard, and use the splicing characteristics corresponding to the mutual information that meets the preset standard as the load characteristics.

[0035] According to a preferred embodiment of the present invention, the identification of the respective load characteristics to obtain a load identification result includes:

[0036] Input the load characteristics into a pre-trained machine learning classifier to obtain a load identification result.

[0037] According to a preferred embodiment of the present invention, the method further includes:

[0038] Select an open dataset including voltages and currents of various loads as the training sample set;

[0039] Generate sample voltage-current trajectory images based on the voltages and currents of the loads in the training sample set respectively; and color-code each of the sample voltage-current trajectory images to generate sample color images;

[0040] Extract the sample trajectory features in each of the sample voltage-current trajectory images, and construct sample cubical complexes based on the sample voltage-current trajectory images and the sample color images respectively, and extract the sample topological features of each of the sample cubical complexes;

[0041] Stitch the sample trajectory features and each of the sample topological features respectively to generate a plurality of sample stitching features, and screen out sample load features from the sample stitching features according to the correlation between each of the sample stitching features;

[0042] Iteratively train a machine learning classifier with each load and its corresponding sample load features to obtain a trained machine learning classifier for identifying loads.

[0043] According to a preferred embodiment of the present invention, the method further includes:

[0044] Use a part of the data in the open dataset as the training sample set; use another part of the data in the open dataset as the test sample set;

[0045] When the machine learning classifier is trained, test the machine learning classifier with the test sample set to verify whether the machine learning classifier is accurate.

[0046] To solve the above technical problems, a second aspect of the present invention proposes a load identification system based on voltage-current trajectory images, and the identification system includes:

[0047] An image generation module, configured to generate voltage-current trajectory images based on original steady-state voltage and current data; and color-code the data in the voltage-current trajectory images to generate color images;

[0048] A feature extraction module, configured to extract the trajectory features in the voltage-current trajectory images, and construct cubical complexes based on the voltage-current trajectory images and the color images respectively, and extract the topological features of each of the cubical complexes;

[0049] A feature screening module, configured to stitch each of the trajectory features and each of the topological features respectively to generate a plurality of stitching features, and screen out load features from the stitching features according to the correlation between each of the stitching features;

[0050] A load identification module, configured to identify each of the load characteristics to obtain a load identification result.

[0051] According to a preferred embodiment of the present invention, the image generation module is specifically configured to determine the active voltage and active current according to the voltage-current trajectory image, and generate a first color code representing the direction of the voltage-current trajectory based on the active voltage and the active current; determine the active power and apparent power according to the voltage-current trajectory image, and generate a second color code representing the power factor based on the active power and the apparent power; generate binary images of multiple cycles according to the voltage-current trajectory image, and generate a third color code according to the average value of the pixel values in the binary images; generate the color image according to the first color code, the second color code, and the third color code.

[0052] According to a preferred embodiment of the present invention, the feature extraction module is specifically configured to use the pixel points in the voltage-current trajectory image and the color image as vertices respectively, and construct a high-dimensional cube between the vertices according to the pixel values of each of the pixel points; assemble each of the high-dimensional cubes corresponding to the voltage-current trajectory image to obtain a cubical complex corresponding to the voltage-current trajectory image; assemble each of the high-dimensional cubes corresponding to the color image to obtain a cubical complex corresponding to the color image.

[0053] According to a preferred embodiment of the present invention, the feature extraction module is specifically configured to filter each of the cubical complexes according to the pixel value size to obtain filtered cubical complexes corresponding to different pixel values; extract the Betti number, persistent amplitude, and persistent entropy in the filtered cubical complexes as the topological features; wherein, the Betti number is the number of points in the filtered cubical complex, the persistent amplitude is the farthest distance of the points in the filtered cubical complex from the diagonal line, and the persistent entropy is the Shannon entropy of the filtered cubical complex.

[0054] According to a preferred embodiment of the present invention, the feature screening module is specifically configured to calculate the mutual information between each of the splicing features as the correlation degree; compare the mutual information with a preset standard, and use the splicing features corresponding to the mutual information that meets the preset standard as the load characteristics.

[0055] To solve the above technical problems, a third aspect of the present invention proposes an electronic device, including:

[0056] A processor; and

[0057] A memory storing computer-executable instructions, and when the computer-executable instructions are executed, the processor executes the method described in any one of the above embodiments.

[0058] To solve the above technical problems, a fourth aspect of the present invention provides a computer storage medium, wherein the computer storage medium stores one or more programs, and when the one or more programs are executed by a processor, the method described in any one of the above embodiments is implemented.

[0059] Compared with the prior art, the present disclosure has the following advantages: The present disclosure generates a voltage-current trajectory image from voltage-current data, performs color coding on the voltage-current trajectory image to convert it into a color image, constructs a cubical complex based on the voltage-current trajectory image and the color image, then extracts topological features on the cubical complex, and splices these topological features with trajectory features to obtain a spliced feature containing a richer representation of load features. Finally, the spliced features that meet the requirement of relevance are selected as load features, and a load recognition result is obtained based on the selected load features. Based on the voltage-current trajectory image, the present solution further converts the voltage-current trajectory image into a color image corresponding to each data in the voltage-current trajectory image. The data in the image is filtered by constructing a cubical complex from the trajectory image and the color image to obtain topological features. By combining the trajectory features and the topological features, a spliced feature is obtained, enriching and expanding the feature space from more levels, so that the finally obtained load features can more comprehensively express the characteristics of the corresponding load, improving the accuracy of load recognition.

[0060] Other features and advantages of the present disclosure will be described in the following specification, and, in part, will be obvious from the specification, or will be understood by implementing the present disclosure. The objectives and other advantages of the present disclosure can be realized and obtained by the structures pointed out in the specification, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] To more clearly illustrate the technical solutions in the embodiments of the present disclosure or the prior art, the following will briefly introduce the drawings required for describing the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present disclosure. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0062] Figure 1 FIG. shows a schematic flow chart of a load recognition method based on a voltage-current trajectory image according to an embodiment of the present disclosure;

[0063] Figure 2 FIG. shows a schematic flow chart of a load recognition method based on a voltage-current trajectory image according to an embodiment of the present disclosure (first);

[0064] Figure 3 FIG. shows a schematic flow chart of a load recognition method based on a voltage-current trajectory image according to an embodiment of the present disclosure (second);

[0065] Figure 4 Shows a schematic structural diagram of a cubical complex and a filtered cubical complex according to an embodiment of the present disclosure;

[0066] Figure 5 Shows a schematic flowchart of a method for training a machine learning classifier according to an embodiment of the present disclosure;

[0067] Figure 6 Shows a block diagram of a load identification system based on a voltage - current trajectory image according to an embodiment of the present disclosure;

[0068] Figure 7 Shows a schematic structural diagram of an electronic device according to an embodiment of the present disclosure. Detailed implementation manners

[0069] To make the objectives, technical solutions and advantages of the embodiments of the present disclosure clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present disclosure with reference to the accompanying drawings in the embodiments of the present disclosure. Obviously, the described embodiments are some, but not all, of the embodiments of the present disclosure. Based on the embodiments in the present disclosure, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present disclosure.

[0070] The same reference numerals in the drawings represent the same or similar elements, components or parts. Therefore, the repeated description of the same or similar elements, components or parts may be omitted hereinafter. It should also be understood that although ordinal adjectives such as first, second, third, etc. may be used herein to describe various devices, elements, components or parts, these devices, elements, components or parts should not be limited by these ordinal adjectives. That is, these ordinal adjectives are only used to distinguish one from another. For example, the first device may also be called the second device without departing from the essential technical solution of the present invention. In addition, the terms "and / or", "or / and" mean all combinations including any one or more of the listed items.

[0071] Please refer to Figure 1 , Figure 1 is a schematic flowchart of a load identification method based on a voltage - current trajectory image provided by the present invention. As shown in Figure 1 , the method includes:

[0072] S11. Generate a voltage - current trajectory image based on the original steady - state voltage - current data; generate a color image by performing color coding on the data in the voltage - current trajectory image.

[0073] In this embodiment, the original steady-state voltage and current data generally refer to the specific values of voltage and current after they reach a steady state in a circuit or system. This kind of data is very important in fields such as power systems, electronic devices, and circuit analysis, and is used for evaluating system performance, conducting fault diagnosis, and optimizing designs, etc. The voltage and current values in the circuit are directly measured using a voltmeter and an ammeter. Under steady-state conditions, these values should be constant or periodically repeated (in an AC circuit). Ensure that the measuring equipment is accurate and reliable, and operate it according to the correct measuring method. Considering various factors in the actual system (such as load changes, temperature changes, component aging, etc.), multiple measurements or long-term monitoring may be required to obtain accurate data. Specifically, the voltage and current data of the power grid are measured in real time. When the voltage and current data are stable or periodically repeated within a preset time period, the voltage and current data are the original steady-state voltage and current data.

[0074] In this embodiment, the original steady-state voltage and current data can be the voltage and current data of the incoming line of a region or a room. The region can be an industrial park or an entire office building. This solution does not make special limitations on this.

[0075] In this embodiment, the voltage-current trajectory image, that is, the V-I trajectory image, the Voltage-Current trajectory image, is a commonly used graphical representation method in circuit analysis for showing the relationship between voltage and current in a circuit. This kind of image is usually obtained through experimental measurement or simulation. Among them, the horizontal axis represents the current (I), and the vertical axis represents the voltage (V). The V-I trajectory image is an intuitive and useful tool for analyzing and understanding the relationship between voltage and current in a circuit and the operating characteristics of the circuit. By observing and analyzing the V-I trajectory image, engineers and scientists can deeply understand the performance, stability, and reliability of the circuit, and accordingly conduct circuit design and optimization.

[0076] In this embodiment, color coding is performed based on the data in the voltage-current trajectory image. The color coding can refer to the three channels of the RGB image. The values for the R channel, G channel, and B channel are respectively calculated through the data in the voltage-current trajectory image. For example, different values can be calculated through current information, voltage information, and power information, thereby converting the characteristics in the voltage and current data into parameters in the image. Similarly, the color coding can also refer to the HSV image. The HSV color space represents colors through three dimensions: Hue, Saturation, and Value. By converting the characteristic data in the voltage and current data into a color image through the above method, more information can be provided, so that more detailed analysis and identification of different types of loads can be carried out.

[0077] S12. Extract the trajectory features in the voltage-current trajectory image, and respectively construct cubical complexes based on the voltage-current trajectory image and the color image, and extract the topological features of each cubical complex.

[0078] In this embodiment, in the voltage-current trajectory image, the trajectory features may be the trajectory direction of the voltage and current, amplitude fluctuation, the area of the region enclosed by the trajectory and the coordinate axes, etc. In different electrical appliances, the physical features of these trajectories will have obvious differences. For example, refrigerators and microwave ovens.

[0079] In this embodiment, the cubical complex mainly involves the specific structures and properties of simplicial complexes in mathematics and combinatorial topology. Cubical complexes or more general simplicial complexes have extensive applications in mathematics and physics. For example, in algebraic topology, simplicial complexes are used to define and calculate the homology groups of topological spaces, so as to study the properties of topological spaces. In combinatorial geometry and discrete geometry, simplicial complexes also play important roles. Respectively constructing cubical complexes based on the voltage-current trajectory image and the color image realizes the conversion of the features in the voltage-current trajectory image and the color image into the spatial structure of the cubical complex, and finally extracts the topological features of each cubical complex to study the properties of the topological space in algebraic topology, that is, to study the characteristics of voltage-current data and corresponding derivative data from more levels and analyze and identify the load from more perspectives.

[0080] Specifically, the trajectory features in the voltage-current trajectory image include: These features include: 1. Asymmetry: That is, whether the original trajectory is consistent with its rotation by 180 degrees. Usually, the distance between the pre-rotation and post-rotation trajectories (such as Haussdorff, etc.) is used to quantitatively characterize the asymmetry. 2. Area: That is, the area of the closed region formed by the trajectory. This feature is related to the active power of the load and some characteristics of the load. For example, for a pure resistance type load, the V-I trajectory area is 0. 3. Curvature of the average line: This feature divides the trajectory into two parts with a broken line, and then examines the degree of bending of this broken line. 4. Slope of the middle section: This feature divides the trajectory into three sections according to the abscissa: left section / middle section / right section, and the slope of the middle section is the slope of the straight line connecting the starting point and the ending point of the middle section. Some small power electrical appliances such as computers have the characteristic of a stable middle section. 5. Areas of the left and right sections: This feature calculates the sum of the areas of the left and right sections. 6. Peak value of the middle section: This feature calculates the maximum value of the ordinate of the middle section. In the present invention, mainly the area, the slope of the middle section, the areas of the left and right sections, and the peak value of the middle section are used as the basic features of the trajectory.

[0081] In addition, some physical characteristics that can be calculated from the original data of voltage and current are added. These characteristics include: 1. Active power and reactive power; 2. Power harmonics; 3. Original current data; 4. Principal component analysis of power harmonics and original current data.

[0082] Specifically, rotate the trajectory in the voltage-current trajectory image, and use the distance between the trajectories before and after rotation as the first trajectory feature; and / or, calculate the area of the closed region formed by the trajectory and the coordinate axes in the voltage-current trajectory image as the second trajectory feature; and / or, add a broken line to the trajectory in the voltage-current trajectory image to divide the trajectory into two parts, and use the degree of bending of the broken line as the third trajectory feature; and / or, divide the trajectory in the voltage-current trajectory image into three equal segments on average, select the starting point and the ending point of the middle segment to calculate the slope of the middle segment as the fourth trajectory feature; and / or, divide the trajectory in the voltage-current trajectory image into three equal segments on average, and use the peak value of the middle segment as the fifth trajectory feature; and / or, calculate the active power and reactive power according to the voltage-current trajectory image as the sixth trajectory feature; and / or, determine the power harmonics according to the voltage-current trajectory image as the seventh trajectory feature; and / or, determine the original current data according to the voltage-current trajectory image as the eighth trajectory feature; and / or, determine the power harmonics and the original current data according to the voltage-current trajectory image, and perform principal component analysis on the power harmonics and the original current data as the ninth trajectory feature.

[0083] S13. Concatenate each of the trajectory features and each of the topological features respectively to generate a plurality of concatenated features, and screen out the load features from the concatenated features according to the correlation between the concatenated features.

[0084] In this embodiment, each of the trajectory features and each of the topological features in the above steps are concatenated respectively to obtain concatenated features, resulting in an expanded feature space, further expanding the number of features for identifying the load and improving the accuracy of load identification.

[0085] In this embodiment, the correlation between the concatenated features can be calculated by means of calculating the Euclidean distance or cosine value of the features in the vector space, etc., and the mutual information between the features can also be calculated to determine the degree of mutual dependence between the concatenated features, so as to screen out more effective features.

[0086] Specifically, calculate the mutual information between each of the concatenated features as the correlation; compare the mutual information with a preset standard, and use the concatenated feature corresponding to the mutual information that meets the preset standard as the load feature.

[0087] Mutual information is an important concept in information theory, which is used to measure the degree of mutual dependence between two random variables. It represents the amount of information contained in another random variable given one random variable. The larger the value of mutual information, the stronger the correlation between the two random variables; conversely, if the value of mutual information is close to 0, it indicates that the correlation between the two random variables is weak.

[0088] The formula for calculating mutual information is as follows:

[0089] ;

[0090] where, is the mutual information of random variables X and Y, and represent the entropies of random variables X and Y respectively, while represents the conditional entropy of random variable X given random variable Y. Entropy is a measure of the uncertainty of a random variable, and the higher the entropy, the greater the uncertainty. By calculating mutual information, we can quantitatively describe the relationship between two random variables, providing valuable information for fields such as data analysis and machine learning. In our experiment, the concatenated features are calculated for mutual information according to the above mutual information formula, and we select features on the condition that the mutual information is not less than 0.6.

[0091] S14. Identify each load feature to obtain a load identification result.

[0092] In this embodiment, the historical load features can be constructed through the above steps with the known load and the historical voltage and current data of the known load. In this solution, the selected load features are compared with the historical load features, and the load corresponding to the load feature is obtained as the load identification result through the comparison result of the load feature and the historical load feature.

[0093] In this embodiment, the corresponding steady-state voltage and current data can also be collected for each load respectively. Based on the same steps as in this solution, the corresponding load features of each load are generated as comparison samples. The similarity between each load feature obtained in this solution and each comparison sample is calculated respectively, and the load corresponding to the comparison sample with the largest similarity to each load feature is determined as the load identification result.

[0094] In this embodiment, the sample load features can also be constructed through this solution with the known load and the voltage and current data of each known load. The load identification model is trained through the sample load features and the known load. The trained load identification model is used to identify each load feature to obtain a load identification result. Specifically, the load feature is input into a pre-trained machine learning classifier to obtain a load identification result.

[0095] The present disclosure generates a voltage-current trajectory image from voltage-current data, color-codes the voltage-current trajectory image to convert it into a color image, constructs a cubical complex based on the voltage-current trajectory image and the color image, extracts topological features on the cubical complex, and splices these topological features with trajectory features to obtain a spliced feature that contains a richer representation of load features. Finally, the spliced features that meet the requirement of relevance are selected as load features, and a load recognition result is obtained based on the selected load features. Based on the voltage-current trajectory image, this solution further converts the color image corresponding to each data in the voltage-current trajectory image, constructs a cubical complex from the trajectory image and the color image to filter the data in the image to obtain topological features, combines the trajectory features and the topological features to obtain spliced features, enriches and expands the feature space from more levels, so that the finally obtained load features can more comprehensively express the characteristics of the corresponding load and improve the accuracy of load recognition.

[0096] As Figure 2 shown, Figure 2 FIG. is a schematic flow chart of a load recognition method based on a voltage-current trajectory image provided by the present invention. Compared with Figure 1 the recognition method shown, the difference lies in the following steps:

[0097] S21. Determine the active voltage and active current according to the voltage-current trajectory image, and generate a first color code representing the direction of the voltage-current trajectory based on the active voltage and active current.

[0098] In this embodiment, a first color code representing the direction of the voltage-current trajectory is generated based on the active voltage and active current in the voltage-current trajectory image. Specifically, the direction of the voltage-current trajectory can be represented by the derivatives of the active voltage and active current, or sampling points can be selected at preset time intervals in the voltage-current trajectory, the directions of each two adjacent sampling points are calculated, and the directions of the adjacent sampling points are averaged to obtain the direction of the voltage-current trajectory.

[0099] In this embodiment, the color code is specifically the HSV color space, and the HSV color space represents colors through three dimensions: Hue, Saturation, and Value.

[0100] In this embodiment, the first color code is calculated by the following formula:

[0101] ;

[0102] ;

[0103] where is the first color code, and are the active voltage and active current at the j-th sampling point, is the maximum value of the voltage, is the maximum value of the current, A is the set formed by the serial numbers of the sampling points in the voltage-current trajectory image, is the number of elements in the set A, and arctan() is the arctangent function.

[0104] S22. Determine the active power and apparent power according to the voltage-current trajectory image, and generate a second color coding representing the power factor based on the active power and the apparent power.

[0105] In this embodiment, the power factor (abbreviated as PF) is the ratio of the active power P to the apparent power S, and is usually represented by cosΦ. It is a coefficient that measures the efficiency of electrical equipment, indicating the ratio of the active power to the apparent power in an AC circuit. The higher the power factor, the higher the utilization rate of the equipment, the better the benefits of the equipment, and the more fully the power generation equipment can be utilized.

[0106] In this embodiment, the second color coding is calculated by the following formula:

[0107] ;

[0108] where, is the second color coding, is the active power, is the apparent power.

[0109] S23. Generate binary images for multiple cycles according to the voltage-current trajectory image, and generate a third color coding based on the average value of the pixel values in the binary images.

[0110] In this embodiment, the voltage-current trajectory image is split into M binary images, that is, the binary voltage-current trajectory images, and the average value of the pixel values of the M binary images is used as the third color coding.

[0111] In this embodiment, the third color coding is calculated by the following formula:

[0112] ;

[0113] where, is the third color coding, is the pixel value of the binary voltage-current trajectory image of the m-th cycle, and M is the number of binary voltage-current trajectory images.

[0114] S24. Generate a color image according to the first color coding, the second color coding, and the third color coding.

[0115] In this embodiment, the above-mentioned first color coding, second color coding, and third color coding are respectively used as the numerical values of the three channels of the HSV image to obtain a color image.

[0116] As Figure 3 shown, Figure 3 FIG. is a schematic flow chart of a load identification method based on a voltage-current trajectory image provided by the present invention. Compared with Figure 1 the identification method shown, the difference lies in the following steps:

[0117] S31. Respectively take the pixel points in the voltage-current trajectory image and the color image as vertices, and construct a high-dimensional cube between the vertices according to the pixel values of each pixel point.

[0118] In this embodiment, the basic idea of the cubical complex is to regard an image as a function (where represents the dimension of the image, is the set of integer points in the -dimensional space), regard the elements in

[0119] S32. Assemble the high-dimensional cubes corresponding to the voltage-current trajectory image to obtain a cubical complex corresponding to the voltage-current trajectory image.

[0120] S33. Assemble the high-dimensional cubes corresponding to the color image to obtain a cubical complex corresponding to the color image.

[0121] In this embodiment, by assembling the cubes corresponding to the voltage-current trajectory image and the color image, a cubical complex is obtained, realizing the conversion of the image into the spatial structure of the cubical complex, so as to study the characteristics of voltage-current data based on the spatial structure and better identify the characteristics of different loads.

[0122] In this embodiment, each cubical complex is filtered according to the pixel value size to obtain a filtered cubical complex corresponding to different pixel values; the Betti number, persistent amplitude, and persistent entropy in the filtered cubical complex are extracted as topological features; among them, the Betti number is the number of points in the filtered cubical complex, the persistent amplitude is the farthest distance of the points in the filtered cubical complex from the diagonal line, and the persistent entropy is the Shannon entropy of the filtered cubical complex.

[0123] As Figure 4 shown, for each cubical complex, the cubical complex can be filtered by the pixel value to obtain a cubical complex corresponding to different pixel values, Figure 4The first cube complex is the cube complex that contains each pixel value. The second cube complex is the filtered cube complex obtained by filtering with a pixel value of 1, that is, the cube complex that only contains the cube complex with a pixel value of 1. The third cube complex is the filtered cube complex with a pixel value of 2. The fourth cube complex is the filtered cube complex with a pixel value of 3. The fifth cube complex is the filtered cube complex with a pixel value of 4. Figure 4 The cube complex in [the text] has a cube structure, and it can also be other shapes and structures. This solution does not make special limitations on this.

[0124] In this embodiment, the present invention considers the HSV image separately in three channels of H / S / V, and directly uses the original image for the binary image without color coding. For these images, we construct a cube complex and select different filters to extract the Betti numbers, amplitudes, and persistent entropies under different metrics as topological features. The calculation formulas of these topological features are as follows (the following is a persistence diagram):

[0125] • Betti number: that is, the number of points in the persistence diagram D considering the multiplicity;

[0126] • Persistent amplitude: that is, the farthest distance of the points in the persistence diagram D from the diagonal;

[0127] • Persistent entropy: that is, the Shannon entropy of the persistence diagram D.

[0128] Please refer to Figure 5 , Figure 5 which is a schematic diagram of the machine learning classifier training method provided by the present invention. As shown in Figure 5 , the method includes:

[0129] S41. Select a publicly available dataset including voltages and currents with multiple loads as the training sample set.

[0130] In this embodiment, this solution can select the publicly available dataset PLAID as the training sample set. PLAID is a high-frequency dataset that contains the voltages and currents of a single load, with a sampling frequency of 30 kHz. It includes 11 common household appliances, 312 different electrical appliances, and 1793 samples.

[0131] S42. Generate sample voltage-current trajectory images according to the voltages and currents of the loads in the training sample set respectively; and color-code each sample voltage-current trajectory image to generate a sample color image.

[0132] In this embodiment, sample voltage-current trajectory images are generated through the voltage and current data of each load in the training sample set. The same as the solution disclosed in the above embodiment, the sample voltage-current trajectory images are color-coded to generate sample color images.

[0133] S43. Extract the sample trajectory features from each sample voltage-current trajectory image, and respectively construct a sample cube complex based on the sample voltage-current trajectory image and the sample color image, and extract the sample topological features of each sample cube complex.

[0134] S44. Concatenate the sample trajectory features and each sample topological feature respectively to generate multiple sample concatenated features, and screen out the sample load features from the sample concatenated features according to the correlation between each sample concatenated feature.

[0135] In this embodiment, the sample load features are finally generated according to the sample voltage-current trajectory image and the sample color image for training the recognition model.

[0136] S45. Iteratively train the machine learning classifier through each load and the corresponding sample load features to obtain a trained machine learning classifier for recognizing loads.

[0137] In this embodiment, XGBoost can be selected as the machine learning classifier for training. Using the sample load features as the input of the machine learning classifier and the types of loads as the output of the machine learning classifier, iteratively train the machine learning classifier. On the one hand, it can be determined whether the training is completed according to the accuracy of the model training results, and on the other hand, according to the recall rate of the model training results.

[0138] In this embodiment, a part of the data in the public dataset can also be used as the training sample set; another part of the data in the public dataset can be used as the test sample set; when the machine learning classifier is trained, the test sample set is used to test the machine learning classifier to verify whether the machine learning classifier is accurate. Specifically, this solution selects the public dataset PLAID to train the model, where 20% of the data is randomly selected as the test set and 80% as the training set, and the model is trained through 5-fold cross-validation.

[0139] Please refer to Figure 6 , Figure 6 which is a block diagram of a load recognition system based on voltage-current trajectory images provided by the present invention. As Figure 6 shown, the system includes: an image generation module 11, a feature extraction module 12, a feature screening module 13, and a load recognition module 14.

[0140] In this embodiment, the image generation module 11 is used to generate a voltage-current trajectory image based on the original steady-state voltage-current data; and generate a color image through color coding of the data in the voltage-current trajectory image.

[0141] In this embodiment, the feature extraction module 12 is configured to extract the trajectory features in the voltage-current trajectory image, and respectively construct cubical complexes based on the voltage-current trajectory image and the color image, and extract the topological features of each cubical complex.

[0142] In this embodiment, the feature screening module 13 is configured to splice each of the trajectory features and each of the topological features to generate a plurality of spliced features, and screen out the load features from the spliced features according to the correlation between the spliced features.

[0143] In this embodiment, the load identification module 14 is configured to identify each of the load features to obtain a load identification result.

[0144] In this embodiment, the image generation module 11 is specifically configured to determine the active voltage and active current according to the voltage-current trajectory image, and generate a first color coding representing the direction of the voltage-current trajectory based on the active voltage and active current; determine the active power and apparent power according to the voltage-current trajectory image, and generate a second color coding representing the power factor based on the active power and apparent power; generate binary images of multiple periods according to the voltage-current trajectory image, and generate a third color coding according to the average value of the pixel values in the binary images; generate a color image according to the first color coding, the second color coding, and the third color coding.

[0145] In this embodiment, the feature extraction module 12 is specifically configured to use the pixel points in the voltage-current trajectory image and the color image as vertices respectively, and construct high-dimensional cubes between the vertices according to the pixel values of each pixel point; assemble the high-dimensional cubes corresponding to the voltage-current trajectory image to obtain the cubical complex corresponding to the voltage-current trajectory image; assemble the high-dimensional cubes corresponding to the color image to obtain the cubical complex corresponding to the color image.

[0146] In this embodiment, the feature extraction module 12 is specifically configured to filter each of the cubical complexes according to the pixel value size to obtain the filtered cubical complexes corresponding to different pixel values; extract the Betti number, the persistent amplitude, and the persistent entropy in the filtered cubical complexes as topological features; where the Betti number is the number of points in the filtered cubical complex, the persistent amplitude is the farthest distance from the points in the filtered cubical complex to the diagonal line, and the persistent entropy is the Shannon entropy of the filtered cubical complex.

[0147] In this embodiment, the feature extraction module 12 is specifically configured to rotate the trajectory in the voltage-current trajectory image, and use the trajectory spacing before and after rotation as the first trajectory feature; and / or calculate the area of the closed region formed by the trajectory and the coordinate axis travel in the voltage-current trajectory image as the second trajectory feature; and / or add a broken line to the trajectory in the voltage-current trajectory image to divide the trajectory into two parts, and use the degree of bending of the broken line as the third trajectory feature; and / or divide the trajectory in the voltage-current trajectory image into three equal parts on average, select the starting point and the ending point of the middle part to calculate the slope of the middle part as the fourth trajectory feature; and / or divide the trajectory in the voltage-current trajectory image into three equal parts on average, and use the peak value of the middle part as the fifth trajectory feature; and / or calculate the active power and reactive power according to the voltage-current trajectory image as the sixth trajectory feature; and / or determine the power harmonics according to the voltage-current trajectory image as the seventh trajectory feature; and / or determine the original current data according to the voltage-current trajectory image as the eighth trajectory feature; and / or determine the power harmonics and the original current data according to the voltage-current trajectory image, and perform principal component analysis on the power harmonics and the original current data as the ninth trajectory feature.

[0148] In this embodiment, the feature screening module 13 is specifically configured to calculate the mutual information between each splicing feature as the correlation degree; compare the mutual information with a preset standard, and use the splicing feature corresponding to the mutual information that meets the preset standard as the load feature.

[0149] In this embodiment, the load recognition module 14 is specifically configured to input the load feature into a pre-trained machine learning classifier to obtain a load recognition result.

[0150] In this embodiment, the system further includes: a model training module, which is specifically configured to select a public data set including voltages and currents of multiple loads as a training sample set; generate sample voltage-current trajectory images according to the voltages and currents of the loads in the training sample set respectively; and color-code each sample voltage-current trajectory image to generate a sample color image; extract sample trajectory features in each sample voltage-current trajectory image, and construct a sample cubical complex based on the sample voltage-current trajectory image and the sample color image respectively, and extract sample topological features of each sample cubical complex; splice the sample trajectory features and each sample topological feature respectively to generate multiple sample splicing features, and screen out sample load features from the sample splicing features according to the correlation degree between each sample splicing feature; perform iterative training on the machine learning classifier through each load and the corresponding sample load feature to obtain a trained machine learning classifier for load recognition.

[0151] In this embodiment, the model training module is specifically configured to use a part of the data in the public dataset as a training sample set; use another part of the data in the public dataset as a test sample set; when the machine learning classifier training is completed, test the machine learning classifier through the test sample set to verify whether the machine learning classifier is accurate.

[0152] As Figure 7 shown, an embodiment of the present invention provides an electronic device, including a processor 1110, a communication interface 1120, a memory 1130, and a communication bus 1140. Among them, the processor 1110, the communication interface 1120, and the memory 1130 complete mutual communication through the communication bus 1140;

[0153] The memory 1130 is used to store a computer program;

[0154] The processor 1110, when executing the program stored on the memory 1130, implements any of the above recognition methods.

[0155] For the electronic device provided in the embodiment of the present invention, the processor 1110 generates a voltage-current trajectory image based on the original steady-state voltage and current data; generates a color image by performing color coding on the data in the voltage-current trajectory image; extracts the trajectory features in the voltage-current trajectory image, and respectively constructs a cubical complex based on the voltage-current trajectory image and the color image, and extracts the topological features of each cubical complex; splices each trajectory feature and each topological feature respectively to generate a plurality of spliced features, and screens out the load features from the spliced features according to the correlation between the spliced features; identifies each load feature to obtain a load identification result.

[0156] The communication bus 1140 mentioned in the above electronic device may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus 1140 can be divided into an address bus, a data bus, a control bus, etc. For the sake of simplicity, only a thick line is used to represent it in the figure, but it does not mean that there is only one bus or one type of bus.

[0157] The communication interface 1120 is used for communication between the above electronic device and other devices.

[0158] The memory 1130 may include a random access memory 1130 (RAM), or may also include a non-volatile memory 1130, such as at least one disk memory 1130. Optionally, the memory 1130 may also be at least one storage device located far away from the aforementioned processor 1110.

[0159] The aforementioned processor 1110 may be a general-purpose processor 1110, including a central processing unit 1110 (CPU), a network processor 1110 (NP), etc.; it may also be a digital signal processor 1110 (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0160] An embodiment of the present invention provides a computer-readable storage medium. The computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors 1110 to implement the recognition method of any of the above embodiments.

[0161] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions according to the embodiments of the present invention are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more integrated available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid-state disk (SSD)).

[0162] Although the present disclosure has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments or perform equivalent replacements for some of the technical features; and these modifications or replacements do 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 disclosure.

Claims

1. A load identification method based on voltage and current trajectory images, characterized in that The recognition method includes: Generating a voltage-current trajectory image based on original steady-state voltage and current data; generating a color image by performing color coding on the data in the voltage-current trajectory image; Extracting trajectory features in the voltage-current trajectory image, and using the pixel points in the voltage-current trajectory image and the color image as vertices respectively. According to the pixel values of each pixel point, a high-dimensional cube is constructed between the vertices. The extracting of the trajectory features in the voltage-current trajectory image includes: rotating the trajectory in the voltage-current trajectory image, and using the trajectory spacing before and after rotation as the first trajectory feature; and / or, calculating the area of the closed region formed by the trajectory in the voltage-current trajectory image and the coordinate axes as the second trajectory feature; and / or, adding a broken line to the trajectory in the voltage-current trajectory image to divide the trajectory into two parts, and using the degree of bending of the broken line as the third trajectory feature; and / or, evenly dividing the trajectory in the voltage-current trajectory image into three segments, and selecting the starting point and the ending point of the middle segment to calculate the slope of the middle segment as the fourth trajectory feature; and / or, evenly dividing the trajectory in the voltage-current trajectory image into three segments, and using the peak value of the middle segment as the fifth trajectory feature; and / or, calculating the active power and reactive power according to the voltage-current trajectory image as the sixth trajectory feature; and / or, determining power harmonics according to the voltage-current trajectory image as the seventh trajectory feature; and / or, determining the original current data according to the voltage-current trajectory image as the eighth trajectory feature; and / or, determining power harmonics and the original current data according to the voltage-current trajectory image, and performing principal component analysis on the power harmonics and the original current data as the ninth trajectory feature; Assembling each of the high-dimensional cubes corresponding to the voltage-current trajectory image to obtain a cubical complex corresponding to the voltage-current trajectory image; Assembling each of the high-dimensional cubes corresponding to the color image to obtain a cubical complex corresponding to the color image; Extracting topological features of each of the cubical complexes; including: filtering each of the cubical complexes according to the pixel value size to obtain filtered cubical complexes corresponding to different pixel values; extracting the Betti number, persistent amplitude, and persistent entropy in the filtered cubical complexes as the topological features; Respectively splicing each of the trajectory features and each of the topological features to generate a plurality of spliced features, and screening out load features from the spliced features according to the correlation between each of the spliced features; Identifying each of the load features to obtain a load identification result, including: inputting the load features into a pre-trained machine learning classifier to obtain a load identification result.

2. The recognition method according to claim 1, wherein The generating a color image by performing color coding on the data in the voltage-current trajectory image includes: Determining the active voltage and active current according to the voltage-current trajectory image, and generating a first color coding representing the direction of the voltage-current trajectory based on the active voltage and the active current; Determine the active power and apparent power according to the voltage-current trajectory image, and generate a second color coding representing the power factor based on the active power and the apparent power; Generate binary images of multiple cycles according to the voltage-current trajectory image, and generate a third color coding according to the average value of pixel values in the binary images; Generate the color image according to the first color coding, the second color coding, and the third color coding.

3. The recognition method according to claim 1, characterized in that The extracting the topological features of each of the cubical complexes includes: The Betti number is the number of points in the filtered cubical complex, the persistence amplitude is the farthest distance of a point in the filtered cubical complex from the diagonal, and the persistence entropy is the Shannon entropy of the filtered cubical complex.

4. The recognition method according to claim 1, characterized in that The screening out the load features from the splicing features according to the correlation between each of the splicing features includes: Calculate the mutual information between each of the splicing features as the correlation; Compare the mutual information with a preset standard, and use the splicing feature corresponding to the mutual information that meets the preset standard as the load feature.

5. The recognition method according to claim 1, wherein The method further includes: Select a publicly available dataset of voltages and currents including multiple loads as the training sample set; Generate sample voltage-current trajectory images according to the voltages and currents of the loads in the training sample set respectively; and perform color coding on each of the sample voltage-current trajectory images to generate sample color images; Extract the sample trajectory features in each of the sample voltage-current trajectory images, and respectively construct sample cubical complexes based on the sample voltage-current trajectory images and the sample color images, and extract the sample topological features of each of the sample cubical complexes; Splice the sample trajectory features and each of the sample topological features respectively to generate multiple sample splicing features, and screen out sample load features from the sample splicing features according to the correlation between each of the sample splicing features; Iteratively train a machine learning classifier with each load and the corresponding sample load features to obtain a trained machine learning classifier for identifying loads.

6. The recognition method according to claim 5, characterized in that The method further includes: Use a part of the data in the publicly available dataset as the training sample set; use another part of the data in the publicly available dataset as the test sample set; When the machine learning classifier is trained, test the machine learning classifier with the test sample set to verify whether the machine learning classifier is accurate.

7. A load recognition system based on voltage-current trajectory images, characterized in that, The recognition system includes: An image generation module, configured to generate a voltage-current trajectory image based on original steady-state voltage and current data; perform color coding on the data in the voltage-current trajectory image to generate a color image; A feature extraction module for extracting trajectory features from the voltage-current trajectory image, including: rotating the trajectory in the voltage-current trajectory image, and taking the trajectory spacing before and after rotation as the first trajectory feature; and / or, calculating the area of the closed region formed by the trajectory and the coordinate axis travel in the voltage-current trajectory image as the second trajectory feature; and / or, adding a broken line to the trajectory in the voltage-current trajectory image to divide the trajectory into two parts, and taking the degree of bending of the broken line as the third trajectory feature; and / or, evenly dividing the trajectory in the voltage-current trajectory image into three segments, selecting the starting point and the ending point of the middle segment to calculate the slope of the middle segment as the fourth trajectory feature; and / or, evenly dividing the trajectory in the voltage-current trajectory image into three segments, and taking the peak value of the middle segment as the fifth trajectory feature; and / or, calculating the active power and reactive power according to the voltage-current trajectory image as the sixth trajectory feature; and / or, determining power harmonics according to the voltage-current trajectory image as the seventh trajectory feature; and / or, determining the original current data according to the voltage-current trajectory image as the eighth trajectory feature; and / or, determining power harmonics and original current data according to the voltage-current trajectory image, and performing principal component analysis on the power harmonics and the original current data as the ninth trajectory feature; and constructing a cubical complex based on the voltage-current trajectory image and the color image respectively, and extracting the topological features of each cubical complex; The feature extraction module is specifically configured to use the pixel points in the voltage-current trajectory image and the color image as vertices respectively, and construct high-dimensional cubes between the vertices according to the pixel values of each pixel point; assemble the high-dimensional cubes corresponding to the voltage-current trajectory image to obtain the cubical complex corresponding to the voltage-current trajectory image; assemble the high-dimensional cubes corresponding to the color image to obtain the cubical complex corresponding to the color image; The feature extraction module is specifically configured to filter each cubical complex according to the pixel value size to obtain filtered cubical complexes corresponding to different pixel values; extract the Betti numbers, persistent amplitudes, and persistent entropies in the filtered cubical complexes as the topological features; A feature screening module for splicing each of the trajectory features and each of the topological features to generate a plurality of spliced features, and screening out load features from the spliced features according to the correlation between the spliced features; A load recognition module for recognizing each of the load features to obtain a load recognition result, including: inputting the load features into a pre-trained machine learning classifier to obtain a load recognition result.

8. The recognition system according to claim 7, characterized in that The image generation module is specifically configured to determine the active voltage and active current according to the voltage-current trajectory image, and generate a first color code representing the direction of the voltage-current trajectory based on the active voltage and the active current; Determine the active power and apparent power according to the voltage-current trajectory image, and generate a second color coding representing the power factor based on the active power and the apparent power; generate binary images of multiple periods according to the voltage-current trajectory image, and generate a third color coding according to the average value of the pixel values in the binary images; generate the color image according to the first color coding, the second color coding, and the third color coding.

9. The recognition system according to claim 7, characterized in that, The feature extraction module is specifically configured to filter each of the cubical complexes according to the pixel value size to obtain filtered cubical complexes corresponding to different pixel values; extract the Betti number, persistent amplitude, and persistent entropy in the filtered cubical complexes as the topological features; wherein, the Betti number is the number of points in the filtered cubical complex, the persistent amplitude is the farthest distance from the points in the filtered cubical complex to the diagonal line, and the persistent entropy is the Shannon entropy of the filtered cubical complex.

10. The recognition system according to claim 7, wherein The feature screening module is specifically configured to calculate the mutual information between each of the splicing features as the correlation degree; compare the mutual information with a preset standard, and use the splicing features corresponding to the mutual information that meets the preset standard as the load features.

11. An electronic device, characterized in that, Comprising: A processor; And A memory storing computer-executable instructions, which when executed cause the processor to execute the method according to any one of claims 1-6.

12. A computer storage medium, characterized in that, Wherein, The computer storage medium stores one or more programs, which when executed by the processor, implement the method according to any one of claims 1-6.

Images