A method for completing missing power data

Abstract

The present invention discloses a method for completing missing power data, which includes obtaining original power data and dividing it into a complete data set and a missing data set; using the dynamic time warping algorithm to determine the dynamic time warping distance, and using the K-nearest neighbor algorithm to construct a nearest neighbor data matrix according to the dynamic time warping distance of the power sequence; optimizing the weight distribution of the nearest neighbor data matrix to obtain a first completion value; calculating the attribute correlation influence coefficient according to the nearest neighbor data matrix as the second completion value; determining the completion value of the power sequence in the missing data set according to the first completion value and the second completion value; moving the completed power sequence out of the missing data set and adding it to the complete data set. The present invention can accurately complete the missing data in the power data. The effective repair of the missing values through the completion method can truly reflect the real electricity consumption situation of users, providing complete and effective basic power data for the analysis of user-related electricity consumption behavior research.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power, and particularly relates to a method for completing missing electric power data. Background Art

[0002] With the continuous construction and improvement of the smart grid, electric power data presents the characteristics of a large number of users, a wide coverage range, and high time accuracy. Conducting data analysis and data modeling applications based on complete electric power data is an important foundation for the power system to carry out many fields such as electric power load forecasting, regional electric power allocation, and power consumption monitoring of key power consumption units. During the process of collecting electric power data, due to the influence of smart meter failures, data transmission channel blockages, etc., irregular data loss phenomena will occur. Among them, the loss of electric power data belongs to random loss, which is manifested as the uncertainty of both the size and the time point of the loss. The electric power data set containing losses can be processed by direct deletion and filling. When the amount of missing samples is not negligible relative to the data set, directly deleting the missing samples will cause a large amount of information loss, which may lead to conclusive misjudgments in the analysis results. In contrast, it is very necessary to study the completion method within a reasonable range of losses.

[0003] The existing electric power data completion methods are mainly divided into two categories. One category is to complete the electric power data by using classical completion methods such as the mean method, the difference method, and the regression method. The advantage of this category is that the method principle is simple and very convenient to use. However, such classical completion methods only consider the numerical distribution form of the data. If applied in the process of completing electric power data, the spatio-temporal distribution characteristics of the electric power data will be directly ignored, resulting in an unsatisfactory data completion effect. The other category is to use complex models such as deep learning to construct data completion methods. This method can effectively extract the spatio-temporal correlation of electric power data. However, due to the complexity of the model, the time cost is high and it is not easy to be popularized and applied. The occurrence point of the loss in the electric power sequence may be at any time point in the known time sequence, and may include complex loss scenarios such as continuous loss and isolated point loss. The above two categories of methods cannot handle complex loss situations.

[0004] The K-Nearest Neighbor algorithm (KNN) can relatively simply divide the complete data set according to whether the data is complete, so as to quickly construct sufficient complete training samples, and calculate the distance of the power consumption sequence of the same user to adaptively complete its own missing data. First of all, the KNN method can simply divide the data set, solving the defect of many loss scenarios in the electric power sequence. However, this directly destroys the time correlation of the data. Secondly, the KNN method needs to calculate the Euclidean distance between two sequences during the calculation process, and the existence of missing values affects the direct calculation of the distance. Finally, the KNN method can relatively reliably complete the missing data based on the complete data. However, during the calculation process, it is necessary to traverse the data set, and the time cost is relatively large. Summary of the Invention

[0005] In view of the above deficiencies in the prior art, the present invention provides a method for completing missing power data to accurately complete the missing data in the power data. The effective repair of the missing values through the completion method can truly reflect the real power consumption situation of users, providing complete and effective basic power data for the analysis of user-related power consumption behaviors.

[0006] In order to achieve the above object of the invention, the technical solution adopted by the present invention is as follows:

[0007] A method for completing missing power data, comprising the following steps:

[0008] S1. Obtain the original power data and divide it into a complete data set and a missing data set;

[0009] S2. Use the dynamic time warping algorithm to determine the dynamic time warping distance of the power energy sequences in the complete data set and the missing data set, and use the K-nearest neighbor algorithm to construct a nearest neighbor data matrix according to the dynamic time warping distance of the power energy sequences;

[0010] S3. Optimize the weight distribution of the nearest neighbor data matrix to obtain the first completion value;

[0011] S4. Calculate the attribute correlation influence coefficient according to the nearest neighbor data matrix as the second completion value;

[0012] S5. Determine the completion value of the power energy sequence in the missing data set according to the first completion value and the second completion value;

[0013] S6. Move the completed power energy sequence out of the missing data set and add it to the complete data set, and determine whether the missing data set is empty; if so, sort the complete data set and the missing data set in the original power set order; otherwise, select the next power energy sequence from the missing data set and return to step S2.

[0014] Optionally, step S2 specifically includes the following sub-steps:

[0015] S2-1. Select a power energy sequence containing missing values from the missing data set, traverse the power energy sequences in the complete data set to calculate the dynamic time warping distance, and construct a dynamic time warping distance matrix;

[0016] S2-2. Select a set number of complete sequences with the smallest dynamic time warping distance from the dynamic time warping distance matrix to construct a nearest neighbor data matrix.

[0017] Optionally, the calculation method for calculating the dynamic time warping distance in step S2-1 is:

[0018] d dtw_t =DTW(s i ,S train_t )

[0019] D dtw = {d dtw_1 , d dtw_2 , …, d dtw_t}

[0020] Among them, d dtw_t is the power sequence s containing missing values in the missing dataset i and the dynamic time warping distance calculated from the t-th power sequence in the complete dataset S train_t , D dtw is the dynamic time warping distance matrix.

[0021] Optionally, step S3 specifically includes the following sub-steps:

[0022] S3-1. Calculate the weight coefficient matrix of the nearest neighbor data matrix according to the power sequence containing missing values in the missing dataset and the nearest neighbor data matrix;

[0023] S3-2. Calculate the corresponding weight assignment matrix according to the weight coefficient matrix of the nearest neighbor data matrix;

[0024] S3-3. Calculate the first completion value according to the power data in the column corresponding to the missing value in the nearest neighbor data matrix and the weight assignment matrix.

[0025] Optionally, step S3-1 specifically includes:

[0026] Divide the power sequence containing missing values in the missing dataset by the nearest neighbor data in each row of the nearest neighbor data matrix to obtain the weight coefficient matrix of the nearest neighbor data matrix, denoted as

[0027]

[0028] Among them, s i is the power sequence containing missing values in the missing dataset, S neighbor is the nearest neighbor data matrix, W K is the weight coefficient vector of the K-th row of the nearest neighbor data matrix, W K = {w1, w2, …, w 24}, define w j = 0 at the missing point, when the denominator is 0, w j = 0.

[0029] Optionally, step S3-2 specifically includes:

[0030] Sum and average the weight coefficients in each row of the weight coefficient matrix to obtain the weight assignment matrix of the nearest neighbor data matrix, denoted as

[0031]

[0032] Among them, W K is the weight coefficient vector of the K-th row of the neighbor data matrix, and W K ={w1, w2, …, w 24}. At the missing point, w j =0. When the denominator is 0, w j =0; j is the sampling period.

[0033] Optionally, step S3-3 specifically includes:

[0034] Multiply the power data corresponding to the missing column in the neighbor data matrix by the weight distribution matrix to obtain the first completion value, denoted as

[0035] x i =∑W'y i

[0036] Among them, x i is the missing value of the missing electric energy sequence, i is the column where the missing value of the electric energy sequence is located, W' is the weight distribution matrix of the neighbor matrix, and y i is the neighbor matrix data corresponding to the column where the missing value is located.

[0037] Optionally, step S4 specifically includes the following sub-steps:

[0038] S4-1. Calculate the covariance matrix of the neighbor data matrix;

[0039] S4-2. Calculate the mean of the neighbor data in each column of the neighbor data matrix, and subtract the mean of the corresponding column from the neighbor data in each column of the neighbor data matrix to obtain the centered matrix of the neighbor data matrix;

[0040] S4-3. Multiply the columns of the centered matrix of the neighbor data matrix where the non-missing values are located by the covariance matrix to obtain the attribute correlation influence coefficient, which is used as the second completion value.

[0041] Optionally, the second completion value is denoted as:

[0042]

[0043] Among them, r is the number of columns where the non-missing values are located in the centered matrix, cov(Y, Y) r is the covariance matrix, Y is the column data of the neighbor matrix, and z r is the column data where the non-missing values are located in the centered matrix of the neighbor data matrix.

[0044] Optionally, step S5 specifically includes:

[0045] Sum the first completion value and the second completion value to obtain the completion value of the electric energy sequence in the missing dataset, denoted as:

[0046] x fill = x i + x'

[0047] Wherein, x fill is the completion value of the power sequence in the missing dataset, x i is the first completion value, and x' is the second completion value.

[0048] The present invention has the following beneficial effects:

[0049] (1) The present invention proposes a power data completion method based on DTWKNN. Based on the K-Nearest Neighbor (KNN) completion method, DTW is used as the distance metric to solve the problem that the unequal sequence lengths caused by data missing make it impossible to effectively calculate the distance. At the same time, the weight combination method is optimized to further improve the data completion accuracy. Facing the problem that KNN destroys the data correlation, the improved DTWKNN method in the present invention adds the calculation of attribute influence relationship to re-establish the data attribute correlation influence.

[0050] (2) The present invention effectively improves KNN. The improved DTWKNN algorithm has a better completion effect than the KNN completion method in various scenarios; at the same time, it reduces the time cost required for completion within a certain range, and is easy to promote and widely applied in the process of completing the electricity consumption data of enterprise users. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 is a schematic flow chart of a method for completing power missing data in the present invention;

[0052] Figure 2 is a schematic diagram of the dynamic time warping algorithm in the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0053] The following describes the specific embodiments of the present invention to facilitate those skilled in the art of the present technology to understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art of the present technology, as long as various changes are within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions and creations using the concept of the present invention are within the scope of protection.

[0054] As Figure 1 shown, the embodiment of the present invention provides a method for completing power missing data, including the following steps S1 to S6:

[0055] S1. Obtain the original power data and divide it into a complete dataset and a missing dataset;

[0056] In an alternative embodiment of the present invention, the purpose of step S1 is to access the industrial enterprise user data set and clarify the data missing situation. The positions of the missing data are counted and marked, and the data is regularized into a daily power consumption sequence matrix with n dimensions (columns) according to the meter sampling frequency. The data is divided into a complete data set and a missing data set through the detection data marks.

[0057] Specifically, the power data is collected every day with a certain collection period T, and the original power data is organized in matrix form. A power consumption time vector is constructed on a daily scale, and by accumulating data for several days, the original power consumption data matrix can be obtained, expressed as

[0058] S = [s1, s2, …, s n '

[0059] where S is the power data matrix; n is the number of days; s n = {y1, y2, …, y 24} is a vector composed of the power consumption at 24 different moments of each day, representing the power consumption situation of the day; the original power data is divided into a missing data set S miss , and a complete data set S train .

[0060] S2. Use the dynamic time warping algorithm to determine the dynamic time warping distances of the power sequences in the complete data set and the missing data set, and use the K-nearest neighbor algorithm to construct a nearest neighbor data matrix according to the dynamic time warping distances of the power sequences;

[0061] In an alternative embodiment of the present invention, in step S2, a power sequence containing missing values in the missing data set is taken out in sequence, and the dynamic time warping distance (DTW distance) of this power sequence is calculated by traversing the complete data set, thereby forming a DTW distance matrix. In the KNN algorithm, it is necessary to find K complete data with similar distances to supplement the missing values. Therefore, by presetting the K value, K complete power sequences with the smallest DTW distances are found, and the K complete power sequences are used to form a nearest neighbor data matrix.

[0062] Step S2 specifically includes the following sub-steps:

[0063] S2-1. Select a power sequence s i containing missing values from the missing data set, and calculate the dynamic time warping distance of this power sequence by traversing the power sequences in the complete data set, and construct a dynamic time warping distance matrix, expressed as:

[0064] d dtw_t = DTW(s i , S train_t )

[0065] Ddtw = {d dtw_1 , d dtw_2 , …, d dtw_t}

[0066] where d dtw_t is the power sequence s i containing missing values in the missing dataset train_t and the dynamic time warping distance calculated from the t-th power sequence in the complete dataset S dtw , and D

[0067] is the dynamic time warping distance matrix.

[0068] The calculation rule of the DTW distance is as follows: Figure 2 As shown in the figure, the DTW distance essentially aims to find the shortest path of the sequence. Suppose there are two power sequences E = {e1, e2, …, e n} and U = {u1, u2, …, u m}, where n and m are the lengths of their respective sequences. Construct the distance matrix of the power sequences E and U as

[0069]

[0070] Define d(i, j) = |e i - u j |. Find the training distance and recalculate the cumulative matrix, denoted as

[0071]

[0072] where: i = 1, 2, …, n; j = 1, 2, …, m; D(0, 0) = 0; D(i, 0) = d(i, 0) + D(i - 1, 0); D(0, j) = d(0, j) + D(0, j - 1);

[0073] The upper right corner element D(n, m) in the cumulative matrix D is the DTW distance measure for evaluating the correlation of the power sequences, that is, D(n, m) = DTW(E, U).

[0074] S2-2. Select a set number of complete sequences with the smallest dynamic time warping distance from the dynamic time warping distance matrix to construct the nearest neighbor data matrix.

[0075] Specifically, after calculating the DTW distance matrix D dtw , set the value of K in the KNN algorithm, that is, find the K complete data closest to the missing sequence to obtain the nearest neighbor matrix s i ∈ S train .

[0076] S3. Optimize the weight assignment of the neighbor data matrix to obtain the first completion value;

[0077] In an alternative embodiment of the present invention, the purpose of step S3 is to optimize the weight assignment of the neighbor matrix. Specifically, divide the missing value sequence by the neighbor matrix of K rows to obtain the weight coefficient of K rows, sum and average each row to obtain the weight assignment matrix of K sequences. Multiply the power consumption data corresponding to the missing column (dimension) in the neighbor matrix by the corresponding data and multiply by the weight and sum to obtain the first completion value.

[0078] Step S3 specifically includes the following sub-steps:

[0079] S3-1. Calculate the weight coefficient matrix of the neighbor data matrix according to the power sequence containing missing values in the missing data set and the neighbor data matrix, specifically including:

[0080] Divide the power sequence containing missing values in the missing data set by the neighbor data of each row of the neighbor data matrix to obtain the weight coefficient matrix of the neighbor data matrix, denoted as

[0081]

[0082] where s i is the power sequence containing missing values in the missing data set, S neighbor is the neighbor data matrix, W K is the weight coefficient vector of the Kth row of the neighbor data matrix, W K ={w1, w2,..., w 24}, define w j =0 at the missing point, and when the denominator is 0, w j =0.

[0083] S3-2. Calculate the corresponding weight assignment matrix according to the weight coefficient matrix of the neighbor data matrix, specifically including:

[0084] Sum and average the weight coefficients of each row in the weight coefficient matrix to obtain the weight assignment matrix of the neighbor data matrix, denoted as

[0085]

[0086] where W K is the weight coefficient vector of the Kth row of the neighbor data matrix, W K ={w1, w2,..., w 24}, define w j =0 at the missing point, and when the denominator is 0, w j =0; j is the sampling period.

[0087] S3-3. Calculate the first completion value according to the power data and the weight distribution matrix corresponding to the missing column in the neighbor data matrix, specifically including:

[0088] Multiply the power data corresponding to the missing column in the neighbor data matrix by the weight distribution matrix to obtain the first completion value, expressed as

[0089] x i =∑W'y i

[0090] where x i is the missing value of the missing electric energy sequence, i is the column where the missing value of the electric energy sequence is located, W' is the weight distribution matrix of the neighbor matrix, and y i is the data of the neighbor matrix corresponding to the column where the corresponding missing value is located.

[0091] S4. Calculate the attribute correlation influence coefficient based on the neighbor data matrix as the second completion value;

[0092] In an optional embodiment of the present invention, the purpose of step S4 is to calculate the attribute correlation coefficient. Specifically, calculate the covariance matrix of the neighbor data matrix, and at the same time calculate the mean value of each column in the original data set and subtract the mean value of each column from each column to centralize the original data. Multiply the non-missing columns corresponding to the centralization matrix values of the neighbor matrix by the covariance coefficient to calculate the attribute correlation coefficient and call it the second completion value.

[0093] Step S4 specifically includes the following sub-steps:

[0094] S4-1. Calculate the covariance matrix of the neighbor data matrix;

[0095] Specifically, covariance is a measure of the relationship between two variables. Given sequences A and B with lengths both l, the covariance calculation formula is:

[0096]

[0097] In the embodiment, S neighbor is represented by columns as S neighbor ={Y1, Y2, …, Y 24}, a covariance can be calculated between every two dimensions, and after calculating all relevant covariances, a 24x24 covariance matrix can be obtained, expressed as

[0098]

[0099] S4-2. Calculate the mean value of the neighbor data in each column of the neighbor data matrix, and subtract the mean value of the neighbor data in each column from the neighbor data in the corresponding column to obtain the centralized matrix of the neighbor data matrix;

[0100] Specifically, in order to eliminate the influence of dimension in the process of data analysis, the present invention performs a centering process on the nearest neighbor data matrix. The specific process is to subtract the numerical average value of all non-null values of the corresponding attribute from the data point, which is expressed as:

[0101]

[0102] where i is each column of the nearest neighbor data matrix, and ∑y i is the sum of the nearest neighbor data in each column of the nearest neighbor data matrix, and m is the number of non-zero data in each column of the nearest neighbor data matrix.

[0103] S4-3. Multiply the columns where non-missing values are located in the centering matrix of the nearest neighbor data matrix by the covariance matrix to obtain the attribute correlation influence coefficient as the second completion value.

[0104] Specifically, the calculation formula of the attribute correlation influence coefficient adopted by the present invention is:

[0105]

[0106] where r is the number of columns where non-missing values are located in the centering matrix, cov(Y,Y) r is the covariance matrix, and z r is the column where non-missing values are located in the centering matrix of the nearest neighbor data matrix.

[0107] S5. Determine the completion value of the electric energy sequence in the missing data set according to the first completion value and the second completion value;

[0108] In an optional embodiment of the present invention, step S5 specifically includes:

[0109] Sum the first completion value and the second completion value to obtain the completion value of the electric energy sequence in the missing data set, which is expressed as:

[0110] x fill = x i + x'

[0111] where x fill is the completion value of the electric energy sequence in the missing data set, x i is the first completion value, and x' is the second completion value.

[0112] S6. Remove the completed electric energy sequence from the missing data set, add it to the complete data set, and determine whether the missing data set is empty; if so, sort the complete data set and the missing data set in the original power set order; otherwise, select the next electric energy sequence from the missing data set and return to step S2.

[0113] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, and combinations of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general purpose computers, special purpose computers, embedded processors, or other programmable data processing devices to produce a machine, such that the instructions executed by the processors of the computer or other programmable data processing devices produce means for implementing the functions specified in the flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in multiple blocks.

[0114] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including instruction means that implement the functions specified in the flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in multiple blocks.

[0115] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in the flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in multiple blocks.

[0116] Specific embodiments are used in the present invention to elaborate on the principles and implementation manners of the present invention. The description of the above embodiments is only for helping to understand the method of the present invention and its core idea; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present invention.

[0117] Those of ordinary skill in the art will realize that the embodiments described herein are for helping the reader understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those of ordinary skill in the art can make various specific deformations and combinations that do not depart from the essence of the present invention based on the technical revelations disclosed in the present invention, and these deformations and combinations are still within the protection scope of the present invention.

Claims

1. A method for completing missing power data, characterized in that, It includes the following steps: S1. Obtain the original power data and divide it into a complete data set and a missing data set; S2. Use the dynamic time warping algorithm to determine the dynamic time warping distances of the power sequences in the complete data set and the missing data set, and use the K-nearest neighbor algorithm to construct a nearest neighbor data matrix based on the dynamic time warping distances of the power sequences; S3. Optimize the weight assignment of the nearest neighbor data matrix to obtain the first completion value. Step S3 specifically includes the following sub-steps: S3-1. Calculate the weight coefficient matrix of the nearest neighbor data matrix according to the power sequences containing missing values in the missing data set and the nearest neighbor data matrix; S3-2. Calculate the corresponding weight assignment matrix according to the weight coefficient matrix of the nearest neighbor data matrix; S3-3. Calculate the first completion value according to the power data in the column corresponding to the missing value in the nearest neighbor data matrix and the weight assignment matrix; S4. Calculate the attribute correlation influence coefficient according to the nearest neighbor data matrix as the second completion value. Step S4 specifically includes the following sub-steps: S4-1. Calculate the covariance matrix of the nearest neighbor data matrix; S4-2. Calculate the mean of the nearest neighbor data in each column of the nearest neighbor data matrix, and subtract the mean of the corresponding column from the nearest neighbor data in each column of the nearest neighbor data matrix to obtain the centralized matrix of the nearest neighbor data matrix; S4-3. Multiply the columns of the centralized matrix of the nearest neighbor data matrix where the non-missing values are located by the covariance matrix to obtain the attribute correlation influence coefficient as the second completion value; S5. Determine the completion value of the power sequence in the missing data set according to the first completion value and the second completion value; S6. Remove the completed power sequence from the missing data set, add it to the complete data set, and determine whether the missing data set is empty; if so, sort the complete data set and the missing data set in the order of the original power set; Otherwise, select the next power sequence from the missing data set and return to step S2.

2. The method for completing missing power data according to claim 1, wherein Step S2 specifically includes the following sub-steps: S2-1. Select a power sequence containing missing values from the missing data set, traverse the power sequences in the complete data set to calculate the dynamic time warping distances, and construct a dynamic time warping distance matrix; S2-2. Select a set number of complete sequences with the smallest dynamic time warping distances from the dynamic time warping distance matrix to construct a nearest neighbor data matrix.

3. The power outage data completion method according to claim 2, characterized in that The calculation method for calculating the dynamic time warping distance in step S2-1 is: ， ， Among them, is the power sequence containing missing values in the missing dataset and the complete dataset the dynamic time warping distance calculated with the th power sequence, and is the dynamic time warping distance matrix.

4. A method for completing missing power data according to claim 1, characterized in that, Step S3-1 specifically includes: Divide the power sequences containing missing values in the missing data set by the nearest neighbor data in each row of the nearest neighbor data matrix to obtain the weight coefficient matrix of the nearest neighbor data matrix, denoted as ， Among them, is the power sequence containing missing values in the missing dataset, is the nearest neighbor data matrix, is the weight coefficient vector of the K-th row of the nearest neighbor data matrix, , defined at the missing point , when the denominator is 0, .

5. A method for completing missing power data according to claim 1, characterized in that Step S3-2 specifically includes: Sum and average the weight coefficients in each row of the weight coefficient matrix to obtain the weight assignment matrix of the nearest neighbor data matrix, denoted as: ， Among them, is the weight coefficient vector of the K-th row of the neighbor data matrix, , and is defined at the missing point . When the denominator is 0, ; j is the sampling period.

6. The method for completing missing power data according to claim 1, characterized in that, Step S3-3 specifically includes: Multiply the power data in the column corresponding to the missing value in the nearest neighbor data matrix by the weight assignment matrix to obtain the first completion value, denoted as: ， Among them, is the missing value of the missing electric energy sequence, is the column where the missing value of the electric energy sequence is located, is the weight assignment matrix of the neighbor matrix, is the neighbor matrix data corresponding to the column where the corresponding missing value is located.

7. A method for completing power outage data according to claim 1, characterized in that The second completion value is denoted as: ， Among them, r is the number of columns where non-missing values are located in the centering matrix, is the covariance matrix, Y is the column data of the nearest neighbor matrix, is the column data where non-missing values are located in the centering matrix of the nearest neighbor data matrix.

8. A method for completing missing power data according to claim 1, characterized in that Step S5 specifically includes: Sum the first completion value and the second completion value to obtain the completion value of the power sequence in the missing data set, denoted as: ， Among them, is the completion value of the electric energy sequence in the missing dataset, is the first completion value, is the second completion value.

Images