Harmonic source integrated learning positioning method based on dominant scale feature transmission matching

Abstract

The application provides a harmonic source integrated learning positioning method based on dominant scale feature transmission matching, wavelet packet decomposition is used to perform multi-scale decomposition on each monitoring quantity, and unique wave fluctuation features of each harmonic source are extracted; a plurality of base learners are generated by combining voltage or current monitoring quantities in an arbitrary manner; on the basis of screening out the best window width and the data segment when only the dominant harmonic source acts by using an improved dynamic windowing algorithm, dominant scale features of each base learner are estimated d Regression coefficients corresponding to the sequence are obtained, and improved weighted voting is used to integrate voting results of the base learners, so that the position of the dominant harmonic source is output.

Description

Technical Field

[0001] This invention relates to the field of harmonic source localization technology in power systems, and in particular to a harmonic source integrated learning localization method based on dominant scale feature transfer matching. Background Technology

[0002] With economic development, a large number of distributed clean energy sources and nonlinear loads have been connected to the power grid, injecting significant harmonic pollution and making the harmonic situation in the grid more complex and difficult to estimate. Harmonic source location is crucial for identifying the sources of harmonic pollution and is a prerequisite for harmonic liability delineation and resolving economic disputes. Currently, the installation density of harmonic monitoring devices in distribution networks is low, failing to meet observability requirements and rendering existing harmonic source location methods inapplicable. Therefore, how to locate harmonic sources using existing statistical harmonic monitoring data with limited measurement points is an urgent problem to be solved. Furthermore, massive, multi-dimensional, and long-term harmonic monitoring data contains rich time-varying information about harmonic sources. How to extract harmonic source location information from statistical harmonic monitoring data using a data-driven approach combined with intelligent algorithms deserves in-depth research. Summary of the Invention

[0003] In view of the defects and shortcomings of the existing technology, the purpose of this invention is to provide a harmonic source integrated learning localization method based on dominant scale feature transmission matching. According to the data and distribution characteristics of existing harmonic monitoring devices, the method fully explores the wave characteristics of different harmonic sources and combines their transmission characteristics to realize harmonic source localization, thus avoiding the deployment of a large number of dedicated synchronous measurement devices.

[0004] The scheme first uses wavelet packet decomposition to decompose each monitored quantity into multiple scales, extracting the unique wave characteristics of each harmonic source. Second, considering the asynchronous nature of the monitored data and the lack of phase angle information, a transfer equation for the wave characteristics at each scale is established based on the system path information. An improved dynamic windowing algorithm is used to select the optimal window width and the data segment where only the dominant harmonic source is active. At the same time, the parameters of the transfer equation are identified, and then the nodes where the dominant harmonic source exists are matched based on the parameters of the transfer equation. Finally, several base learners are generated by arbitrarily combining voltage or current monitored quantities. An improved weighted voting method is used to integrate the voting results of each base learner to output the location of the dominant harmonic source.

[0005] The specific technical solution adopted is as follows:

[0006] A harmonic source ensemble learning localization method based on dominant scale feature transfer matching is proposed. This method utilizes wavelet packet decomposition to perform multi-scale decomposition of each monitored quantity, extracting the unique fluctuation characteristics of each harmonic source. Several base learners are generated by arbitrarily combining voltage or current monitored quantities. Based on the selection of the optimal window width and the data segment where only the dominant harmonic source is active using an improved dynamic windowing algorithm, the regression coefficients corresponding to the dominant scale d-sequence of each base learner are estimated. Finally, an improved weighted voting method is used to integrate the voting results of each base learner, outputting the location of the dominant harmonic source.

[0007] Furthermore, the step of using wavelet packet decomposition to perform multi-scale decomposition on each monitored quantity and extracting the unique wave characteristics of each harmonic source specifically involves:

[0008] Obtain the statistical harmonic monitoring sequences stored by each harmonic monitoring device in the network; select the wavelet basis and decomposition level to decompose all monitoring data sequences into wavelet coefficient sequences of different scale subspaces; select the decomposition results of the harmonic voltage average value sequence, and calculate the energy of the wavelet coefficients of each scale subspace. The calculation formula is as follows:

[0009]

[0010] In the formula, q represents the number of decomposition levels, F represents the set of measurement points, j represents the subspace scale, and c f (q,j) represents the wavelet coefficient sequence of the j-th scale subspace at level q, ||.|| 2 Represents the L2 norm;

[0011] The sum of the energy of wavelet coefficients at all scales except the 0-scale subspace is calculated using the following formula:

[0012]

[0013] The energy proportion of wavelet coefficients in each scale subspace is calculated using the following formula:

[0014]

[0015] Scales with a subspace energy percentage exceeding 5% are defined as dominant scales;

[0016] The wavelet coefficient sequence belonging to the dominant scale in the wavelet coefficient sequence is reconstructed.

[0017] Furthermore, the generation of several base learners from the arbitrary combination of voltage or current monitoring quantities specifically involves:

[0018] K base learners are formed by pairwise combining harmonic voltage or branch harmonic current data of different statistical types. The formula for calculating K is as follows:

[0019]

[0020] In the formula, C u and C i These represent the number of harmonic voltage and current state variables that the harmonic monitoring device in the network can acquire, respectively.

[0021] Furthermore, the specific steps of using the improved dynamic windowing algorithm to filter out the data segment that only dominates the harmonic source are as follows:

[0022] A linear model is constructed by reconstructing the sequence using wavelet coefficients of the dominant scale d of the harmonic monitoring quantities associated with each learner, as follows:

[0023] C ad =β d1 C bd +β d0 +ε

[0024] In the formula, C ad and C bd Let β represent the reconstructed sequence of wavelet coefficients of the dominant scale d of the harmonic monitoring quantities associated with each base learner. d1 and β d2 These are the regression coefficients of the linear model;

[0025] Set the initial window width and the maximum window width H. max ;

[0026] The input sequence of the linear model is divided according to the window width H;

[0027] Estimate the regression coefficient β and F-statistic for the corresponding sequence within each window using the following formulas;

[0028]

[0029] In the formula, x and y represent the input sequence of the k-th linear model, and g represents the window number;

[0030] like Then discard the corresponding window, and assume there are a total of G windows. Calculate the final regression coefficient corresponding to the window width H using the following formula. And the value of the F statistic; if G < 1, then decrease the threshold of the F statistic:

[0031]

[0032] If H <H max Let H = H+1, then re-divide the input sequence of the linear model according to the window width H and estimate the regression coefficients of the corresponding sequence within each window. and calculation

[0033] If H = H maxThen for Sort the windows by width, grouping them into sets of five, and calculate the average value of each group. And select The largest group represents the optimal window width range.

[0034] Furthermore, the estimation of the regression coefficients corresponding to the dominant scale d sequence of each base learner is specifically as follows:

[0035] The regression coefficients are estimated by averaging the regression estimates within the optimal window width range, as shown in the following formula:

[0036]

[0037] In the formula, H and H+5 represent the range of optimal window width.

[0038] Furthermore, the process of integrating the voting results of each base learner using the improved weighted voting method and outputting the location of the dominant harmonic source specifically involves:

[0039] Calculate the matching coefficient M of the dominant scale d for each base learner ds The calculation formula is as follows;

[0040]

[0041] In the formula, s represents the grid node number, a and b represent the node numbers associated with the harmonic voltage base learner input, and q, w, j, and k represent the node numbers associated with the harmonic current base learner input; Z is the impedance, and Y is the admittance.

[0042] Normalize the matching coefficients;

[0043] The difference in matching coefficients is calculated using the following formula:

[0044]

[0045] In the formula, max(·) represents taking the maximum value;

[0046] The weights of the base learner are calculated using the following formula:

[0047]

[0048] The voting value of the base learner is calculated using the following formula:

[0049]

[0050] Based on the voting results of each base learner, the probability that the harmonic source is located at node s is calculated using the following formula.

[0051]

[0052] Based on the probability calculation results, output the node with the highest probability of the harmonic source appearing;

[0053] Harmonic source localization is achieved by outputting the node with the highest probability of occurrence of the harmonic source at each dominant scale.

[0054] And an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the harmonic source integrated learning localization method based on dominant scale feature transfer matching as described above.

[0055] A non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the harmonic source ensemble learning localization method based on dominant scale feature transfer matching as described above.

[0056] Compared to existing technologies, this invention and its preferred scheme utilize wavelet packet decomposition to divide the time series according to frequency and screen the dominant scale through wavelet coefficient energy, effectively extracting the different wave characteristics of each harmonic source and improving the efficiency of subsequent analysis. The proposed improved dynamic window width regression parameter estimation method eliminates the problem of large differences in results with similar window widths, filters out the time periods of the dominant harmonic source, and accurately identifies the regression coefficients of each linear model. The proposed dominant harmonic source localization model based on ensemble learning fully utilizes the data provided by the harmonic monitoring device and effectively integrates the path information matching results of multiple base learners through weighted coefficients. This enables harmonic source localization even when the harmonic monitoring device does not meet observability requirements, and it also possesses certain robustness and generalization ability. Attached Figure Description

[0057] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0058] Figure 1 This is a schematic diagram illustrating the implementation process of the method in an embodiment of the present invention. Detailed Implementation

[0059] In the following, specific embodiments of this application will be described in detail with reference to the accompanying drawings. Based on these detailed descriptions, those skilled in the art will be able to clearly understand and implement this application. Without departing from the principles of this application, features from various embodiments can be combined to obtain new implementations, or certain features from some embodiments can be substituted to obtain other preferred implementations.

[0060] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0061] To make the features and advantages of this patent more apparent and understandable, specific embodiments are provided below for detailed explanation:

[0062] like Figure 1 As shown, this embodiment of the invention provides a harmonic source ensemble learning localization method based on dominant scale feature transfer matching, the implementation process of which includes the following steps:

[0063] Step S1: Obtain the statistical harmonic monitoring sequences stored by each harmonic monitoring device in the network;

[0064] Step S2: Select the wavelet basis and the number of decomposition levels to decompose all monitoring data sequences into wavelet coefficient sequences of different scale subspaces;

[0065] Step S3: Select the decomposition result of the harmonic voltage average value sequence of the wavelet coefficient sequence obtained in step S2, and calculate the energy of the wavelet coefficients in each scale subspace. The calculation formula is as follows:

[0066]

[0067] In the formula, q represents the number of decomposition levels, F represents the set of measurement points, j represents the subspace scale, and c f (q,j) represents the wavelet coefficient sequence of the j-th scale subspace at level q, ||.|| 2 Represents the L2 norm;

[0068] Step S4: Calculate the sum of the energy of wavelet coefficients at all scales except the 0-scale subspace for all measurement points. The calculation formula is as follows:

[0069]

[0070] Step S5: Calculate the energy proportion of wavelet coefficients in each scale subspace. The calculation formula is as follows:

[0071]

[0072] Step S6: Define the scales whose subspace energy accounts for more than 5% as the dominant scales;

[0073] Step S7: Reconstruct the wavelet coefficient sequence obtained in step S2 that belongs to the dominant scale.

[0074] Step S8: By combining harmonic voltage or branch harmonic current data of different statistical value types in pairs, K base learners are formed. The formula for calculating K is as follows:

[0075]

[0076] In the formula, C u and C i These represent the number of harmonic voltage and current state variables that the harmonic monitoring device in the network can acquire. Since statistical data includes statistical values ​​such as maximum, minimum, average, and 95% probability of a large value, the number of base learners should be multiplied by 4.

[0077] Step S8: Reconstruct the sequence using the wavelet coefficients of the dominant scale d of the harmonic monitoring quantity associated with the k-th base learner, and build a linear model in the following form:

[0078]

[0079] In the formula, and Let represent the reconstructed sequence of wavelet coefficients of the dominant scale d of the harmonic monitoring quantity associated with the k-th base learner. and These are the regression coefficients of the linear model;

[0080] Step S9: Set the initial window width and the maximum window width H max ;

[0081] Step S10: Divide the input sequence of the linear model according to the window width H;

[0082] Step S11: Estimate the regression coefficients and F-statistics of the corresponding sequences within each window. The calculation formula is as follows;

[0083]

[0084] In the formula, x and y represent the input sequence of the k-th linear model, and g represents the window number.

[0085] Step S12, if Then discard that window. Assuming there are ultimately G windows, calculate the final regression coefficient corresponding to the window width H using the following formula. And the value of the F statistic. If G < 1, then decrease the threshold of the F statistic:

[0086]

[0087] Step S13, if H = H max If yes, proceed to the next step. Otherwise, set H = H + 1 and return to step S10.

[0088] Step S14, for Sort the windows by width, grouping them into sets of five, and calculate the average value of each group. And select The largest group represents the optimal window width range;

[0089] Step S15: Take the average of the regression estimation results within the optimal window width range to obtain the final estimated regression coefficients. The calculation formula is as follows:

[0090]

[0091] In the formula, H and H+5 represent the range of optimal window widths.

[0092] Step S16: Through the iterative calculations from Step S9 to Step S14, the regression coefficients corresponding to the dominant scale d sequence of each base learner are estimated.

[0093] Step S17: Calculate the matching coefficient M of the dominant scale d for each base learner. ds The calculation formula is as follows;

[0094]

[0095] In the formula, s represents the grid node number, a and b represent the node numbers associated with the harmonic voltage base learner input, and q, w, j, and k represent the node numbers associated with the harmonic current base learner input.

[0096] Step S18: Normalize the matching coefficients;

[0097] Step S19: Calculate the difference in matching coefficients. The calculation formula is as follows:

[0098]

[0099] In the formula, max(·) represents taking the maximum value;

[0100] Step S20: Calculate the weights of the base learner using the following formula:

[0101]

[0102] Step S21: Calculate the voting value of the base learner. The calculation formula is as follows:

[0103]

[0104] Step S22: Based on the voting results of each base learner, calculate the probability that the harmonic source is located at node s. The calculation formula is as follows:

[0105]

[0106] Step S23: Based on the probability calculation results, output the node with the highest probability of occurrence of the harmonic source.

[0107] The harmonic source localization is thus completed by outputting the node with the highest probability of occurrence of the harmonic source at each dominant scale.

[0108] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0109] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0110] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0111] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0112] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0113] This patent is not limited to the above-described preferred embodiments. Anyone can derive other forms of harmonic source integrated learning localization methods based on dominant scale feature transfer matching under the guidance of this patent. All equivalent changes and modifications made within the scope of this patent application shall fall within the scope of this patent.

Claims

1. A harmonic source ensemble learning localization method based on dominant scale feature transfer matching, characterized in that: Wavelet packet decomposition method is used to decompose each monitoring quantity into multiple scales and extract the unique wave characteristics of each harmonic source. Several base learners are generated by arbitrarily combining voltage or current monitoring quantities. After using an improved dynamic windowing algorithm to select the optimal window width and the data segment that only dominates the harmonic source, the regression coefficients corresponding to the dominant scale d sequence of each base learner are estimated based on the regression estimation results corresponding to the optimal window width. Then, the voting results of each base learner are integrated using an improved weighted voting method to output the location of the dominant harmonic source. The improved dynamic windowing algorithm sets an initial window width and a maximum window width H max , divides the input sequence according to the window width, calculates the regression coefficient and F statistic in each window, removes the window with F statistic ≤2000, and selects the group with the maximum average value as the best window width range after grouping and averaging the F statistic under the maximum window width. The improved weighted voting method calculates the matching coefficients and differences between the matching coefficients of each base learner, determines the weights of the base learners based on the differences, and then calculates the probability of the existence of harmonic sources at each node through weighted voting. The estimation of the regression coefficients corresponding to the dominant scale d sequence of each base learner is specifically as follows: The regression coefficients are estimated by averaging the regression estimates within the optimal window width range, as shown in the following formula: In the formula, H and H+5 represent the range of the optimal window width; The process of integrating the voting results of each base learner using the improved weighted voting method and outputting the location of the dominant harmonic source is as follows: Calculating the matching coefficient M of each base learner dominant dimension d ds The calculation formula is as follows; In the formula, s represents the grid node number, a and b represent the node numbers associated with the harmonic voltage base learner input, and q, w, j, and k represent the node numbers associated with the harmonic current base learner input; Z is the impedance, and Y is the admittance. Normalize the matching coefficients; The difference in matching coefficients is calculated using the following formula: In the formula, max(·) represents taking the maximum value; The weights of the base learner are calculated using the following formula: The voting value of the base learner is calculated using the following formula: Based on the voting results of each base learner, the probability that the harmonic source is located at node s is calculated using the following formula. Based on the probability calculation results, output the node with the highest probability of the harmonic source appearing; Harmonic source localization is achieved by outputting the node with the highest probability of occurrence of the harmonic source at each dominant scale.

2. The harmonic source ensemble learning localization method based on dominant scale feature transfer matching according to claim 1, characterized in that: The specific steps involve using wavelet packet decomposition to perform multi-scale decomposition of each monitored quantity and extracting the unique wave characteristics of each harmonic source: Obtain the statistical harmonic monitoring sequences stored by each harmonic monitoring device in the network; select the wavelet basis and decomposition level to decompose all monitoring data sequences into wavelet coefficient sequences of different scale subspaces; select the decomposition results of the harmonic voltage average value sequence, and calculate the energy of the wavelet coefficients of each scale subspace. The calculation formula is as follows: where q represents the decomposition level, F denotes the set of measurement points, j represents the subspace scale, c f (q, j) denotes the wavelet coefficient sequence of the qth level j-scale subspace, ||.||2 2 denotes the two-norm; The sum of the energy of wavelet coefficients at all scales except the 0-scale subspace is calculated using the following formula: The energy proportion of wavelet coefficients in each scale subspace is calculated using the following formula: Scales whose subspace energy accounts for more than 5% are defined as dominant scales; The wavelet coefficient sequence belonging to the dominant scale in the wavelet coefficient sequence is reconstructed.

3. The harmonic source ensemble learning localization method based on dominant scale feature transfer matching according to claim 2, characterized in that: The arbitrary combination of voltage or current monitoring quantities generates several base learners specifically as follows: K base learners are formed by pairwise combining harmonic voltage or branch harmonic current data of different statistical types. The formula for calculating K is as follows: In the formula, C u and C i respectively represent the number of state variables of harmonic voltage and current that the harmonic monitoring device in the network can obtain.

4. The harmonic source ensemble learning localization method based on dominant scale feature transfer matching according to claim 3, characterized in that: The improved dynamic windowing algorithm is used to select the optimal window width and the data segment that only dominates the harmonic source. A linear model is constructed by reconstructing the sequence of wavelet coefficients of the dominant scale d using the harmonic monitoring quantities associated with each base learner, as follows: In the formula, C ad and C bd respectively represent the reconstructed sequence of wavelet coefficients of the dominant scale d of the harmonic monitoring quantity associated with each base learner, β d1 and β d2 are the regression coefficients of the linear model; Setting an initial window width and a maximum window width H max ; The input sequence of the linear model is divided according to the window width H; Estimate the regression coefficient β and F-statistic for the corresponding sequence within each window using the following formulas; In the formula, x and y represent the input sequence of the k-th linear model, and g represents the window number; If then the corresponding window is discarded, and if there are G windows left, the regression coefficients and the F-statistic value corresponding to the window width H are calculated by the following formula; if G < 1, the threshold of the F-statistic is reduced: If H < H max , let H = H + 1, re-divide the input sequence of the linear model according to the window width H and estimate the regression coefficients of the corresponding sequence in each window and calculate ; If H = H max , then for Sort by window width size, form a group every 5, calculate the average value of each group , and select The largest group is the best window width range.

5. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the harmonic source ensemble learning localization method based on dominant scale feature transfer matching as described in any one of claims 1-4.

6. A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the harmonic source ensemble learning localization method based on dominant scale feature transfer matching as described in any one of claims 1-4.

Images