Multiple interpolation method for missing data in power collection system based on user quantitative evaluation

Abstract

The present invention proposes a multiple interpolation method for missing data of an electric power collection system based on user quantitative evaluation, including collecting archive data and curve data of user meters to form a user group set of segmented application scenarios, and for user groups whose missing data ratio is less than a set threshold, intercepting the part of their curve data that does not contain missing data as the original training data set for the user group; combining the missing pattern, relying on the original training data set, generating a simulation data set including simulated missing data; combining the real data corresponding to the simulation data set, sorting each algorithm according to the fitting accuracy, generating an algorithm set for the user group data subset, and then applying the optimal multiple interpolation model algorithm to the real data to complete the interpolation of the real data. The present invention can achieve the segmentation of interpolation scenarios from the two dimensions of user quantitative evaluation and data missing pattern, and then complete the optimization of interpolation algorithms in the segmented scenarios relying on the simulation missing experiment.

Description

Technical Field

[0001] The invention belongs to the field of electric power information technology, and in particular relates to a multiple interpolation method for missing data of an electric power acquisition system based on user quantitative evaluation. Background Art

[0002] Since the State Grid Corporation of China has been promoting smart meters, a large amount of real-time curve data such as voltage, current, and power has been continuously recorded by hundreds of millions of terminal devices and collected and aggregated to the data center through the power consumption information collection system. During this period, the data has undergone the conversion from the physical field simulation state to the digital discrete quantity, and has been transmitted through the layers of the communication network, so data loss is inevitable. How to correctly handle the loss of real data has become the primary problem facing big data analysis.

[0003] The processing of missing data can be divided into two categories: data elimination and data supplementation. Although the former ensures the credibility of the final data, more effective information is lost during the elimination process, and there is a problem of selective elimination of non-completely random data. Ideally, the latter can restore the real data to the greatest extent, but it puts forward higher requirements for the accuracy and effectiveness of the supplementation algorithm.

[0004] Existing data completion algorithms can be roughly divided into three categories: traditional interpolation based on data features, multiple interpolation based on missing data distribution prediction, and machine learning interpolation based on neural networks. Machine learning has high requirements on the quality of the training data set itself, and there are deficiencies in the model's interpretability and scalability to different scenarios; traditional interpolation based on data features, such as mean substitution, univariate regression and other interpolation methods, has the disadvantage that a single interpolation value reduces the variability of the data and may introduce fitting bias; multiple interpolation based on missing data distribution prediction does not only generate a single value to interpolate the missing data, but is based on the core interpolation algorithm with the help of Monte Carlo Markov chain or chain equation to generate a series of complete interpolation results, and finally relies on statistical evaluation to generate the final interpolation result. Its interpolation accuracy mainly depends on the specific core interpolation method. Commonly used core interpolation methods include: regression prediction method, Bayesian linear regression method, prediction mean matching method, bootstrap maximum expectation method, random forest method, etc.

[0005] Although the existing multiple interpolation methods have good applications in some specific scenarios, they often fail to take care of one thing while neglecting another in the complex scenarios of State Grid Big Data. The reason is that the existing multiple interpolation algorithms based on a single model have limitations in terms of applicable scenarios and fail to make favorable adaptive adjustments to the specific data characteristics of segmented applications. Summary of the invention

[0006] In order to solve the above problems existing in the prior art, a multiple interpolation method for missing data of a power acquisition system based on user quantitative evaluation is provided.

[0007] The technical solution adopted by the present invention to solve its technical problem is:

[0008] This technical solution proposes a multiple interpolation method for missing data in the power collection system based on user quantitative evaluation, including:

[0009] S1: Collect archive data and curve data of user electricity meters to form a user group set for segmented application scenarios. The user group set includes several user group data subsets.

[0010] S2: Traverse the user groups of the subdivided application scenarios, and conduct a statistical analysis of the curve-type data missing pattern for the user groups with missing data. The analysis content includes the data missing pattern and the proportion of missing data;

[0011] S3: For the user group whose missing data ratio is less than the set threshold, the part of its curve data that does not contain missing data is intercepted as the original training data set of the user group;

[0012] S4: Combined with the missing pattern, relying on the original training data set, a simulation data set including simulated missing data is generated. On the simulation data set, several multiple interpolation model algorithms are called in sequence to obtain the final interpolation results of each multiple interpolation model algorithm on the simulation data set, and the accuracy is evaluated;

[0013] S5: Combined with the real data corresponding to the simulation data set, sort the algorithms according to the fitting accuracy, generate an algorithm set for the user group data subset, and then apply the optimal multiple interpolation model algorithm to the real data to complete the interpolation of the real data.

[0014] Preferably, the method for forming a user group set for segmented application scenarios includes:

[0015] S11: After collecting the archive data and curve data of the user's electricity meter, a user quantitative evaluation system is established based on the industrial and commercial category of the user, the time series analysis results of electricity consumption data, and the statistical results of data distribution as basic parameters;

[0016] S12: Reduce the dimension of user quantitative evaluation, cluster the user quantitative evaluation results based on similarity, and form a data-driven user group set for segmented application scenarios.

[0017] Preferably, a user quantitative evaluation analysis is performed in combination with the user's electricity consumption characteristic statistics: specifically, a probability distribution estimation model is established for the user's historical electricity consumption indication curve data, and analysis is performed from the overall mean, overall standard deviation, skewness, kurtosis and statistical dimensions.

[0018] Preferably, the skewness is measured by adding the third-order standardized moment with unbiased correction, and the kurtosis is measured by the fourth-order standardized moment, where:

[0019]

[0020]

[0021]

[0022] Among them, skewness represents skewness, kurtosis represents kurtosis, and m i is the i-th order standardized moment, is the sampling mean.

[0023] Preferably, for each user's missing data pattern, a Boolean array of the same length as the curve data is selected to represent it, and the true value in the array indicates that the data at the corresponding time point is missing and needs to be interpolated. The missing patterns corresponding to the users in the same user group are collected to form a two-dimensional array as the missing pattern of the user group;

[0024] According to the distribution of missing data points in the time dimension and the continuity of missing data, the data missing patterns are divided into isolated data point missing at the beginning of the data segment, continuous data point missing at the beginning of the data segment, isolated data point missing in the middle of the data segment, continuous data point missing in the middle of the data segment, isolated data point missing at the end of the data segment, and continuous data point missing at the end of the data segment.

[0025] For the user group with two missing patterns, namely, isolated data points missing in the middle of the data and continuous data points missing in the middle of the data, the missing data in the middle of the data are repaired using an interpolation method based on Hermitian polynomials to ensure that the values ​​before and after the interpolated data points are equal and the reciprocals of the high-order derivatives are continuous.

[0026] Preferably, for user groups with obvious periodic electricity consumption characteristics, the missing data is eliminated from the time dimension with the significant period as the smallest unit; the data cliff that appears in the curve after the data is eliminated is repaired by data shifting, including shifting the data after the missing interval of each user separately so that the data points before and after the missing interval overlap, and then resetting the time index to make the time index continuous.

[0027] Preferably, the original training data set is adapted to the data missing template after adding random bias, and the data at the corresponding position of the missing template is recorded and eliminated to generate a simulation data set; the simulation data set includes: the original test data after the specified data is eliminated through simulated missing, as the unified input of the algorithm, and a list recording the values ​​and positions of the eliminated data, as the evaluation basis of the algorithm output.

[0028] Preferably, the method of adding bias to the original data missing template includes: random translation bias in the time dimension, random mapping reconstruction in the user dimension, and a comprehensive bias combining time and user;

[0029] The random shift bias of the time dimension is to keep the relative relationship between users unchanged when the missing occurs, advance or delay the time when the missing occurs by T time points, and ensure that the data missing pattern does not change before and after the time dimension shift;

[0030] The random mapping reconstruction of the user dimension is to keep the time when the missing occurs unchanged, but re-establish the mapping relationship between the user and the power consumption curve;

[0031] The comprehensive bias that combines time and user is a combination of the random translation bias in the above two time dimensions and the random mapping reconstruction in the user dimension.

[0032] Preferably, the interpolation method based on cluster similarity includes:

[0033] S21: in the user group of the subdivided application scenario, based on the curve data in the historical time window W before the time point T to be interpolated, establish a power consumption trend similarity matrix between users;

[0034] S22: record the most relevant M users as the same replacement electricity meter, and establish a linear regression model between the power consumption of the target electricity meter and the same replacement electricity meter according to the curve data in the historical time window T;

[0035] S23: using the electricity consumption of the same replacement meter at time point T to calculate the electricity consumption of the user to be interpolated at that moment, and recording it as a set of interpolation values;

[0036] S24: Change W and M, repeat the above process, form multiple groups of interpolation values, and take the average of the multiple groups of interpolation values ​​as the final interpolation result.

[0037] Preferably, the accuracy evaluation method includes the root mean square deviation (IQR-RMSD) of the relative interquartile range and the adjusted goodness of fit.

[0038]

[0039]

[0040] IQR=CDF -1 (0.75)-CDF -1 (0.25) (6)

[0041] in, and tare the observed values ​​and true values ​​of the tth group, CDF is the quantile function, and the smaller the root mean square deviation IQR-RMSD value relative to the interquartile range, the more accurate the interpolation result of the algorithm on the unified test data set;

[0042]

[0043] Among them, SS res is the residual sum of squares, SS tot is the overall sum of squares, dfe is the degrees of freedom for the error variance estimate, dft is the degrees of freedom for the overall variance estimate, and the adjusted goodness of fit The closer the value is to 1, the more accurate the interpolation result of the algorithm on the unified test data set is;

[0044] In the same user group, the final interpolation output of each algorithm was evaluated by the root mean square deviation (IQR-RMSD), the adjusted goodness of fit, and the The first and second priorities are sorted by accuracy to form a preferred algorithm set, and the first-ranked algorithm is used as the optimal interpolation algorithm for the user group.

[0045] Compared with the prior art, the present invention has the following advantages:

[0046] The present invention realizes the segmentation of interpolation scenarios based on two dimensions: user quantitative evaluation and data missing pattern. In the segmented scenarios, the interpolation algorithm is optimized based on the simulation missing experiment. Based on the user's real curve data, 76339 artificial missing data points are generated after multiple simulations. The method is used for logarithmic interpolation, and the interpolation success rate is 99.2%, of which 63.3% of the data points have a relative interpolation deviation better than 5%, and 12.6% of the data points have a relative interpolation deviation better than 1%. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0048] Figure 1 It is a framework diagram of the present invention;

[0049] Figure 2 is a flow chart of the method of the present invention;

[0050] Figure 3 It is the relationship between the basic parameters of the user quantitative evaluation system;

[0051] Figure 4 It is an example display of typical user electricity consumption in a collection of segmented scenarios. DETAILED DESCRIPTION

[0052] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0053] like Figure 1-4 As shown, this embodiment proposes a multiple interpolation method for missing data of a power collection system based on user quantitative evaluation, including:

[0054] S1: Collect archive data and curve data of user electricity meters to form a user group set for segmented application scenarios. The user group set includes several user group data subsets;

[0055] S2: Traverse the user groups of the subdivided application scenarios, and conduct a statistical analysis of the curve-type data missing pattern for the user groups with missing data. The analysis content includes the data missing pattern and the proportion of missing data;

[0056] S3: For the user group whose missing data ratio is less than the set threshold, the part of its curve data that does not contain missing data is intercepted as the original training data set of the user group;

[0057] S4: Combined with the missing pattern, relying on the original training data set, a simulation data set including simulated missing data is generated as the input of the next interpolation algorithm optimization process. On the simulation data set, several multiple interpolation model algorithms are called in sequence to obtain the final interpolation results of each multiple interpolation model algorithm on the simulation data set, and the accuracy is evaluated;

[0058] S5: Combined with the real data corresponding to the simulation data set, sort each algorithm according to the fitting accuracy, generate an algorithm set for the user group data subset, and then apply the optimal multiple interpolation model algorithm obtained based on the simulation process to the real data to complete the interpolation of the real data.

[0059] The method for forming a user group set for segmented application scenarios includes:

[0060] S11: After collecting the archive data and curve data of the user's electricity meter, a detailed user quantitative evaluation system is established based on the industrial and commercial category of the user, the time series analysis results of electricity consumption data, and the statistical results of data distribution as basic parameters;

[0061] S12: At the same time, in order to improve the utilization rate of cross-information between users and reduce the computational complexity, the dimension of user quantitative evaluation is further reduced, and the detailed user quantitative evaluation results are clustered based on similarity to form a data-driven user group set for segmented application scenarios.

[0062] In step S11, a user quantitative evaluation analysis is performed in combination with the user's electricity consumption characteristic statistics: specifically, a probability distribution estimation model is established for the user's historical electricity consumption indication curve data, and statistical dimensions such as overall mean, overall standard deviation, skewness, and kurtosis are analyzed.

[0063] The skewness is measured by adding the third-order standardized moment with unbiased correction, and the kurtosis is measured by the fourth-order standardized moment, where:

[0064]

[0065]

[0066]

[0067] Among them, skewness represents skewness, kurtosis represents kurtosis, and m i is the i-th order standardized moment, is the sampling mean.

[0068] Quantitative evaluation analysis of users combined with archival non-digital feature information: First, non-digital features such as user categories and industry information are digitally encoded. Then, combined with the multi-dimensional evaluation information of electricity consumption characteristics, user categories are streamlined, merged, split and expanded to reduce electricity consumption differences among users of the same type while increasing differences between different types of users.

[0069] To further highlight the distribution characteristics of users’ electricity consumption behavior in the time dimension, we defined the peak hours on weekdays (7:00-11:00 13:00-19:00), the non-peak hours on weekdays, the peak hours on non-working days (7:00-11:00 13:00-19:00) and the non-peak hours on non-working days.

[0070] User quantitative evaluation analysis combined with potential periodicity analysis: Apply the time series model to perform autocorrelation / partial autocorrelation analysis on the user indication curve data to identify implicit potential periodicity information.

[0071] The calculation methods of autocorrelation ACF and partial autocorrelation PACF are as follows:

[0072]

[0073]

[0074] At the same time, in order to improve the utilization rate of cross-information between users and reduce the computational complexity, the dimension of user quantitative evaluation is further reduced. The detailed user quantitative evaluation results are clustered based on similarity to form a data-driven user group set for segmented application scenarios.

[0075] In step S12, PCA principal component analysis is performed on the evaluation information, and then the DBSCAN method is applied to cluster the meter users. The clustering radius and the minimum density number are adjusted, and the clustering results are appropriately merged to finally form a data-driven segmented application scenario set.

[0076] Figure 3 The relationship between the basic parameters used to establish the user quantitative evaluation system is shown in Figure 1. The main diagonal line is the histogram of each parameter itself; the upper triangular area sub-graph shows the scatter plot of the parameters corresponding to the horizontal and vertical coordinates; the lower triangular area sub-graph shows the binomial distribution contour map of the corresponding parameters after 2D Gaussian approximation; the colors in the figure correspond to the user group sets of some subdivided application scenarios after clustering and merging.

[0077] Figure 4 The figure shows the electricity consumption examples of typical users in each set after clustering to obtain a set of segmented scenarios.

[0078] In step S2, for each user's missing data pattern, a Boolean array of the same length as the curve data is selected to represent it. The true value in the array indicates that the data at the corresponding time point is missing and needs to be interpolated. The missing patterns corresponding to each user in the same user group are collected to form a two-dimensional array as the missing pattern of the user group.

[0079] According to the distribution of missing data points in the time dimension and the continuity of missing data, the data missing patterns are divided into isolated data point missing at the beginning of the data segment, continuous data point missing at the beginning of the data segment, isolated data point missing in the middle of the data segment, continuous data point missing in the middle of the data segment, isolated data point missing at the end of the data segment, and continuous data point missing at the end of the data segment.

[0080] In step S3, the missing data percentage is divided into time dimension and user dimension statistics. The missing data percentage in the time dimension refers to the ratio of the number of users with missing data at that moment to the total number of users, and the default threshold for the missing data percentage in the time dimension is 25%. The missing data percentage in the user dimension refers to the ratio of the number of time points at which the user has missing data to the total number of time points, and the default threshold for the missing data percentage in the user dimension is 35%.

[0081] For the user group whose missing data ratio is less than the set threshold, the part of its curve data containing missing data is removed or repaired as the original training data set for this user group.

[0082] For the user group with two missing patterns, namely, isolated data points missing in the middle of the data and continuous data points missing in the middle of the data, the missing data in the middle of the data are repaired using an interpolation method based on Hermitian polynomials to ensure that the values ​​before and after the interpolated data points are equal and the reciprocals of the high-order derivatives are continuous.

[0083] For user groups with obvious periodic electricity consumption characteristics, the missing data is eliminated from the time dimension with the significant period as the smallest unit (if there are multiple significant periods, the least common multiple of each significant period is used as the common period for elimination).

[0084] The data cliff that appears in the curve after data removal is repaired and solved by data shifting. The data after the missing interval of each user is adjusted and shifted separately so that the data points before and after the missing interval overlap, and then the time index is reset to make the time index continuous.

[0085] In step S4, the original test data set in step S3 is adapted to the data missing template after adding random bias, and the data at the corresponding position of the missing template is recorded and removed to generate a simulation data set. The simulation data set consists of two parts: the original test data after the specified data is removed by simulating missing data as the unified input of the algorithm; and the list of the values ​​and positions of the removed data is recorded as the evaluation basis of the algorithm output.

[0086] Among them, the main ways to add bias to the missing template of the original data are: random translation bias in the time dimension, random mapping reconstruction in the user dimension, and comprehensive bias combining time and user.

[0087] The random shift bias in the time dimension means keeping the relative relationship between users unchanged when the missing occurs, advancing or delaying the time when the missing occurs by T time points, and ensuring that the data missing pattern does not change before and after the time dimension shift.

[0088] The random mapping reconstruction of the user dimension means keeping the time when the loss occurs unchanged, but re-establishing the mapping relationship between the user and the power consumption curve.

[0089] The comprehensive bias refers to the combination of the random translation bias in the above two time dimensions and the random mapping reconstruction in the user dimension.

[0090] The interpolation methods involved include predicted mean matching method, weighted predicted mean matching method, random observation sampling method, classification regression number method, random forest method, Bayesian linear regression method, normal heteroscedasticity method, normal homoscedasticity method, clustering similarity method, bootstrap maximum expectation method, etc.

[0091] Among them, the interpolation methods based on cluster similarity include:

[0092] S21: in the user group of the subdivided application scenario, based on the curve data in the historical time window W before the time point T to be interpolated, establish a power consumption trend similarity matrix between users;

[0093] S22: record the most relevant M users as the same replacement electricity meter, and establish a linear regression model between the power consumption of the target electricity meter and the same replacement electricity meter according to the curve data in the historical time window T;

[0094] S23: using the electricity consumption of the same replacement meter at time point T to calculate the electricity consumption of the user to be interpolated at that moment, and recording it as a set of interpolation values;

[0095] S24: Change W and M, repeat the above process, form multiple groups of interpolation values, and take the average of the multiple groups of interpolation values ​​as the final interpolation result.

[0096] In step S5, the final interpolation results of different algorithms on the unified simulation test set are evaluated for accuracy. The evaluation criteria for interpolation accuracy include the root mean square deviation (IQR-RMSD) of the relative interquartile range, the adjusted goodness of fit, and the accuracy of the interpolation results.

[0097]

[0098]

[0099] IQR=CDF -1 (0.75)-CDF -1 (0.25) (6)

[0100] in, and t are the observed values ​​and true values ​​of the tth group, CDF is the quantile function, and the smaller the root mean square deviation IQR-RMSD value relative to the interquartile range, the more accurate the interpolation result of the algorithm on the unified test data set.

[0101]

[0102] Among them, SS res is the residual sum of squares, SS tot is the overall sum of squares, dfe is the degrees of freedom for the error variance estimate, dft is the degrees of freedom for the overall variance estimate, and the adjusted goodness of fit The closer the value is to 1, the more accurate the interpolation result of the algorithm on the unified test data set is.

[0103] In the same user group, the final interpolation output of each algorithm was evaluated by the root mean square deviation (IQR-RMSD), the adjusted goodness of fit, and the The first and second priorities are sorted by accuracy to form a preferred algorithm set, and the first-ranked algorithm is used as the optimal interpolation algorithm for the user group.

[0104] Then, the optimal interpolation algorithm obtained based on the simulation process is applied to the real missing data to complete the interpolation of the real data. For the real missing data, the optimal interpolation algorithm of the user group is called first. When the optimal interpolation algorithm fails to interpolate, other algorithms in the preferred algorithm set are called in sequence until the interpolation is successful.

[0105] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A multiple interpolation method for missing data in power collection system based on user quantitative evaluation, characterized in that: include: S1: Collect archive data and curve data of user electricity meters to form a user group set for segmented application scenarios. The user group set includes several user group data subsets; S2: Traverse the user groups of the subdivided application scenarios, and conduct a statistical analysis of the curve-type data missing pattern for the user groups with missing data. The analysis content includes the data missing pattern and the proportion of missing data; S3: For the user group whose missing data ratio is less than the set threshold, the part of its curve data that does not contain missing data is intercepted as the original training data set of the user group; S4: Combined with the missing pattern, relying on the original training data set, a simulation data set including simulated missing data is generated. On the simulation data set, several multiple interpolation model algorithms are called in sequence to obtain the final interpolation results of each multiple interpolation model algorithm on the simulation data set, and the accuracy is evaluated; S5: Combined with the real data corresponding to the simulation data set, sort the algorithms according to the fitting accuracy, generate an algorithm set for the user group data subset, and then apply the optimal multiple interpolation model algorithm to the real data to complete the interpolation of the real data; For each user's missing data pattern, a Boolean array of the same length as the curve data is used to represent it. The true value in the array indicates that the data at the corresponding time point is missing and needs to be interpolated. The missing patterns corresponding to each user in the same user group are collected to form a two-dimensional array as the missing pattern of the user group. According to the distribution of missing data points in the time dimension and the continuity of missing data, the data missing patterns are divided into isolated data point missing at the beginning of the data segment, continuous data point missing at the beginning of the data segment, isolated data point missing in the middle of the data segment, continuous data point missing in the middle of the data segment, isolated data point missing at the end of the data segment, and continuous data point missing at the end of the data segment. For user groups with two missing modes, one with isolated data points missing in the middle of the data and the other with continuous data points missing in the middle of the data, the missing data in the middle of the data are repaired using an interpolation method based on Hermitian polynomials to ensure that the values ​​before and after the interpolated data points are equal and the high-order derivatives are continuous; The original training data set is adapted to the data missing template after adding random bias, and the data at the position corresponding to the missing template is recorded and removed to generate a simulation data set; The simulation data set includes: the original test data after the specified data is removed by simulating missing data, which is used as the unified input of the algorithm, and a list that records the values ​​and positions of the removed data, which is used as the evaluation basis of the algorithm output.

2. The missing data multiple interpolation method of the power acquisition system based on user quantitative evaluation according to claim 1 is characterized in that: The method for forming a user group set for segmented application scenarios includes: S11: After collecting the archive data and curve data of the user's electricity meter, a user quantitative evaluation system is established based on the industrial and commercial category of the user, the time series analysis results of electricity consumption data, and the statistical results of data distribution as basic parameters; S12: Reduce the dimension of user quantitative evaluation, cluster the user quantitative evaluation results based on similarity, and form a data-driven user group set for segmented application scenarios.

3. The missing data multiple interpolation method of the power acquisition system based on user quantitative evaluation according to claim 2 is characterized in that: Combined with the statistics of user electricity consumption characteristics, quantitative evaluation and analysis of users is carried out: specifically, a probability distribution estimation model is established based on the user's historical electricity consumption indication curve data, and analysis is performed from the perspectives of overall mean, overall standard deviation, skewness, kurtosis and statistical dimensions.

4. The missing data multiple interpolation method of the power acquisition system based on user quantitative evaluation according to claim 3 is characterized in that: The skewness is measured by adding the third-order standardized moment with unbiased correction, and the kurtosis is measured by the fourth-order standardized moment, where: Among them, skewness represents skewness, kurtosis represents kurtosis, mi is the i-th order standardized moment, is the sampling mean.

5. The missing data multiple interpolation method of the power acquisition system based on user quantitative evaluation according to claim 1 is characterized in that: For user groups with obvious periodic electricity consumption characteristics, missing data are eliminated from the time dimension with the significant period as the smallest unit; for the data cliff that appears in the curve after data elimination, it is repaired by data shifting, including shifting the data after the missing interval of each user separately so that the data points before and after the missing interval overlap, and then resetting the time index to make the time index continuous.

6. The missing data multiple interpolation method of the power acquisition system based on user quantitative evaluation according to claim 1 is characterized in that: Methods for adding bias to the missing template of the original data include: random translation bias in the time dimension, random mapping reconstruction in the user dimension, and comprehensive bias combining time and user; The random shift bias of the time dimension is to keep the relative relationship between users unchanged when the missing occurs, advance or delay the time when the missing occurs by T time points, and ensure that the data missing pattern does not change before and after the time dimension shift; The random mapping reconstruction of the user dimension is to keep the time when the missing occurs unchanged, but re-establish the mapping relationship between the user and the power consumption curve; The comprehensive bias that combines time and user is a combination of the random translation bias in the above two time dimensions and the random mapping reconstruction in the user dimension.

7. The missing data multiple interpolation method of the power acquisition system based on user quantitative evaluation according to claim 1 is characterized in that: It also includes an interpolation method based on cluster similarity, and the specific steps include: S21: in the user group of the subdivided application scenario, based on the curve data in the historical time window W before the time point T to be interpolated, establish a power consumption trend similarity matrix between users; S22: record the most relevant M users as the same replacement electricity meter, and establish a linear regression model between the power consumption of the target electricity meter and the same replacement electricity meter according to the curve data in the historical time window T; S23: using the electricity consumption of the same replacement meter at time point T to calculate the electricity consumption of the user to be interpolated at that moment, and recording it as a set of interpolation values; S24: Change W and M, repeat the above process, form multiple groups of interpolation values, and take the average of the multiple groups of interpolation values ​​as the final interpolation result.

8. The multiple interpolation method for missing data of power acquisition system based on user quantitative evaluation according to claim 1 is characterized in that: The accuracy evaluation method includes the root mean square deviation (IQR-RMSD) of the relative interquartile range and the adjusted goodness of fit (R 2 : IQR=CDF -1 (0.75)-CDF -1 (0.25) (6) in, and yt are the observed values ​​and true values ​​of the tth group, CDF is the quantile function, and the smaller the root mean square deviation IQR-RMSD value relative to the interquartile range, the more accurate the interpolation result of the algorithm on the unified test data set; Among them, SS res is the residual sum of squares, SS tot is the overall sum of squares, dfe is the degrees of freedom for the error variance estimate, dft is the degrees of freedom for the overall variance estimate, and the adjusted goodness of fit The closer the value is to 1, the more accurate the interpolation result of the algorithm on the unified test data set is; In the same user group, the final interpolation output of each algorithm was evaluated by the root mean square deviation (IQR-RMSD), the adjusted goodness of fit, and the The first and second priorities are sorted by accuracy to form a preferred algorithm set, and the first-ranked algorithm is used as the optimal interpolation algorithm for the user group.

Images