Distribution network measurement data repair method and device

Abstract

The present invention provides a method and device for repairing distribution network measurement data, which can be applied to the field of distribution network measurement data processing technology. The distribution network measurement data repair method includes: based on different multiple sequence time steps, determining multiple historical distribution network measurement data sequences that match the missing measurement data from the historical data set; generating a repair vector based on the historical distribution network measurement data sequence and a reference data sequence that matches the historical distribution network measurement data sequence; obtaining multiple target repair vectors based on the repair vector and the repair vector after dimensionality reduction, wherein the repair vector after dimensionality reduction is obtained by reducing the dimension of the repair vector to be reduced; inputting the multiple target repair vectors into the repair model respectively to obtain multiple repair data; and obtaining target repair data that matches the missing measurement data based on the multiple repair data.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network measurement data processing, and more specifically to a distribution network measurement data repair method and device. Background Art

[0002] As the scale of power systems continues to increase, the measurement data of power systems shows a rapid growth trend. However, the process of collecting, measuring, transmitting and storing massive data may lead to the problem of missing measurement data. In particular, the quality and missingness of distribution network measurement data are worse than those of transmission network. If not solved, it will seriously affect the observability of distribution network and threaten the security of power grid.

[0003] Distribution network measurement data usually has the problem of missing time periods, that is, measurement data at multiple times are usually missing. However, traditional measurement data repair methods mainly repair measurement data at a single time, and there is an error accumulation problem when repairing measurement data at multiple times. Therefore, when using traditional measurement data repair methods to repair distribution network measurement data with a high missing rate, the repair accuracy is low. Summary of the invention

[0004] In view of the above problems, the present invention provides a method and device for repairing distribution network measurement data with improved repair accuracy.

[0005] According to a first aspect of the present invention, a distribution network measurement data repair method is provided, comprising: determining, from a historical data set, a plurality of historical distribution network measurement data sequences matching missing measurement data based on a plurality of different sequence time steps; generating a repair vector based on the historical distribution network measurement data sequence and a reference data sequence matching the historical distribution network measurement data sequence; obtaining a plurality of target repair vectors based on the repair vector and the repair vector after dimensionality reduction, wherein the repair vector after dimensionality reduction is obtained by performing dimensionality reduction on the repair vector to be reduced, and the repair vector to be reduced is determined from a plurality of repair vectors; respectively inputting the plurality of target repair vectors into a repair model to obtain a plurality of repair data; and obtaining target repair data matching the missing measurement data based on the plurality of repair data.

[0006] The second aspect of the present invention provides a distribution network measurement data repair device, including: a history determination module, which is used to determine multiple historical distribution network measurement data sequences that match the missing measurement data from a historical data set based on different multiple sequence time steps; a vector generation module, which is used to generate a repair vector based on the historical distribution network measurement data sequence and a reference data sequence that matches the historical distribution network measurement data sequence; a vector dimension reduction module, which is used to obtain multiple target repair vectors based on the repair vector and the repair vector after dimension reduction, wherein the repair vector after dimension reduction is obtained by reducing the dimension of the repair vector to be reduced, and the repair vector to be reduced is determined from multiple repair vectors; a repair input module, which is used to input multiple target repair vectors into a repair model respectively to obtain multiple repair data; and a repair generation module, which is used to obtain target repair data that matches the missing measurement data based on the multiple repair data.

[0007] According to an embodiment of the present invention, multiple historical distribution network measurement data sequences are determined from a historical data set using different multiple sequence time steps to improve the richness of the repair vector used for prediction. At the same time, by combining the reference data sequence, the richness of the repair vector used for prediction is further improved. By performing dimensionality reduction processing on the repair vector, the richness of the repair vector is ensured while simplifying the dimension of the repair vector and improving processing efficiency. In addition, multiple repair data are fused to obtain target repair data, which improves repair accuracy and processing efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] The above contents and other objects, features and advantages of the present invention will become more apparent through the following description of the embodiments of the present invention with reference to the accompanying drawings, in which:

[0009] Figure 1 A flow chart of a method for repairing distribution network measurement data according to an embodiment of the present invention is shown.

[0010] Figure 2 A flow chart of repairing measurement data according to a specific embodiment of the present invention is shown.

[0011] Figure 3 A flowchart of adjusting the hyperparameters of the initial repair model according to an embodiment of the present invention is shown.

[0012] Figure 4 A repair curve diagram of the repair model of the embodiment of the present invention and the comparison model with a sequence time step of 96 is shown.

[0013] Figure 5 A structural block diagram of a distribution network measurement data repair device according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0014] Below, embodiments of the present invention will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present invention. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of embodiments of the present invention. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of concepts of the present invention.

[0015] The terms used herein are only for describing specific embodiments and are not intended to limit the present invention. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0016] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0017] When using expressions such as "at least one of A, B, and C, etc.", they should generally be interpreted according to the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0018] When repairing the distribution network measurement data, the measurement data at a single moment is mainly repaired, that is, the repair model with a sequence time step of 1 is usually used for repair. However, when repairing multiple missing measurement data continuously, starting from the second missing measurement data, the repair model will recursively predict based on the input data containing non-real data to repair the missing measurement data at multiple moments, and its repair accuracy will gradually decrease due to the error accumulation effect.

[0019] Figure 1 A flow chart of a method for repairing distribution network measurement data according to an embodiment of the present invention is shown.

[0020] like Figure 1 As shown, the distribution network measurement data repairing of this embodiment includes operations S110 to S150.

[0021] In operation S110 , a plurality of historical power distribution network measurement data sequences matching the missing measurement data are determined from a historical data set based on a plurality of different sequence time steps.

[0022] According to an embodiment of the present invention, the sequence time step may be used to characterize the time interval between two adjacent measurement moments. Specifically, the sequence time step may be 1, 2, 4, or 8.

[0023] According to an embodiment of the present invention, the historical data set may include measurement data corresponding to each historical measurement moment within a preset historical period. Specifically, the historical measurement moment is earlier than the measurement moment corresponding to the missing measurement data.

[0024] According to an embodiment of the present invention, multiple historical distribution network measurement data sequences may correspond one-to-one to multiple sequence time steps, and multiple historical distribution network measurement data sequences may be different. Specifically, it is possible to determine the historical distribution network measurement data sequences that match the missing measurement data when the sequence time steps are 1, 2, 4, and 8, respectively.

[0025] For example, when the sequence time step is 1, in the historical distribution network measurement data sequence, the time interval between the measurement times of two adjacent historical distribution network measurement data is 1.

[0026] In operation S120 , a repair vector is generated based on a historical power distribution network measurement data sequence and a reference data sequence that matches the historical power distribution network measurement data sequence.

[0027] According to an embodiment of the present invention, the reference data sequence may represent a sequence of reference data corresponding to the historical distribution network measurement data sequence. Specifically, the historical distribution network measurement data sequence includes a plurality of historical distribution network measurement data, and each historical distribution network measurement data may have at least one reference data.

[0028] According to an embodiment of the present invention, the repair vector may include multiple dimensions. For example, based on the historical distribution network measurement data sequence, the reference data sequence may be used as a new dimension to perform a vector stacking operation with the historical distribution network measurement data sequence to generate a repair vector.

[0029] According to the embodiment of the present invention, since there are multiple historical distribution network measurement data sequences, there should also be multiple repair vectors accordingly. Specifically, there is a one-to-one correspondence between the multiple historical distribution network measurement data sequences and the multiple repair vectors.

[0030] In operation S130, a plurality of target repair vectors are obtained based on the repair vector and the repair vector after dimension reduction.

[0031] According to an embodiment of the present invention, the repair vector after dimensionality reduction is obtained by reducing the dimensionality of the repair vector to be reduced, and the repair vector to be reduced is determined from multiple repair vectors.

[0032] According to an embodiment of the present invention, the repair vector to be reduced in dimensionality may be a repair vector with a higher dimensionality among multiple repair vectors. Since a repair vector with a higher dimensionality will increase the repair complexity, the repair vector with a higher dimensionality may be reduced in dimensionality to obtain a repair vector after dimensionality reduction to reduce the complexity. It is understandable that for a repair vector with a lower dimensionality, no dimensionality reduction process is required.

[0033] According to an embodiment of the present invention, the target repair vector may be obtained by performing a preprocessing operation on the repair vector or the repair vector after dimensionality reduction. Specifically, for a repair vector with a lower dimension, a preprocessing operation may be performed directly to obtain the target repair vector; for a repair vector with a higher dimension, i.e., a repair vector to be reduced in dimension, it is necessary to first reduce the dimension to obtain a repair vector after dimensionality reduction, and then perform a preprocessing operation on the repair vector after dimensionality reduction to obtain the target repair vector.

[0034] In operation S140, a plurality of target repair vectors are respectively input into the repair model to obtain a plurality of repair data.

[0035] According to an embodiment of the present invention, the repair model may be obtained based on a long short-term memory network (Long short-term memory, LSTM).

[0036] According to an embodiment of the present invention, the repair data can be used to characterize the prediction result output by the repair model based on the target repair vector. There is a one-to-one correspondence between the multiple target repair vectors and the multiple repair data.

[0037] In operation S150 , target repair data matching the missing measurement data is obtained based on the plurality of repair data.

[0038] According to an embodiment of the present invention, the target repair data may be used to repair missing measurement data. Specifically, the target repair data may be obtained by fusing multiple repair data.

[0039] According to an embodiment of the present invention, the weights of the plurality of repair data may be determined according to the sequence time steps corresponding to the plurality of repair data, and the repair data may be fused based on the weights.

[0040] According to an embodiment of the present invention, multiple historical distribution network measurement data sequences are determined from a historical data set using different multiple sequence time steps to improve the richness of the repair vector used for prediction. At the same time, by combining the reference data sequence, the richness of the repair vector used for prediction is further improved. By performing dimensionality reduction processing on the repair vector, the richness of the repair vector is ensured while simplifying the dimension of the repair vector and improving processing efficiency. In addition, multiple repair data are fused to obtain target repair data, which improves repair accuracy and processing efficiency.

[0041] According to an embodiment of the present invention, a repair vector is generated based on a historical distribution network measurement data sequence and a reference data sequence matching the historical distribution network measurement data, including: determining a reference data sequence related to time based on a sequence time step of the historical distribution network measurement data sequence; determining a reference data sequence related to external factors based on a measurement data type of the historical distribution network measurement data sequence; and obtaining a repair vector based on the historical distribution network measurement data sequence, the reference data sequence related to time, and the reference data sequence related to external factors.

[0042] According to an embodiment of the present invention, the time-related reference data sequence may include a measurement date sequence corresponding to the historical distribution network measurement data sequence, and a measurement date type sequence. Specifically, the measurement date type may indicate whether the measurement date is a holiday, and 1 may be used to indicate that the measurement date is a holiday, and 0 may be used to indicate that the measurement date is a working day.

[0043] According to an embodiment of the present invention, a time-related reference data sequence may be determined based on the sequence time step and the measurement date corresponding to the missing measurement data.

[0044] According to an embodiment of the present invention, the measurement data type may be used to characterize the type of measurement data of the distribution network, and may specifically include a power data type, a voltage data type, etc. of the distribution network.

[0045] According to an embodiment of the present invention, the external factors include at least one of the following: environment, equipment. Specifically, the equipment may be a measurement device for collecting and transmitting distribution network measurement data, and the environment may be the environment where the measurement device is located.

[0046] According to an embodiment of the present invention, the reference data sequence related to the external factors may include a temperature sequence, a device type sequence, and a light condition sequence corresponding to the historical power distribution network measurement data sequence.

[0047] According to an embodiment of the present invention, the repair vector may include multiple dimensions. Specifically, a historical distribution network measurement data sequence, a time-related reference data sequence, and a reference data sequence related to external factors may be respectively used as different dimensions of the repair vector.

[0048] For example, repair vectors can be constructed when the sequence time step is 1, 2, 4, and 8 respectively. For the repair vector when the sequence time step is 1, a 7-dimensional repair vector is constructed, which respectively represents the previous moment data corresponding to the missing measurement data, the corresponding moment data one day ago, the corresponding moment data two days ago, the corresponding moment data three days ago, the corresponding moment data one week ago, the corresponding moment data two weeks ago, and the corresponding moment data three weeks ago; for the repair vector when the sequence time step is 2, 4, and 8, an m-dimensional repair vector is constructed, where m=3*N+3, and N represents the total amount of measurement data for one day (related to the measurement sampling interval, such as the sampling interval is 15 minutes, then N is 96). Specifically, the historical distribution network measurement data series includes , the time-dependent reference data series include , the repair vector includes In addition, the selection of m can be determined according to the actual situation, and a reference data sequence related to external factors can be appropriately used as a new dimension of m.

[0049] According to an embodiment of the present invention, a repair vector is generated by utilizing a historical distribution network measurement data sequence of multiple dimensions, a reference data sequence related to time, and a reference data sequence related to external factors, thereby improving the richness of the repair vector and the repair accuracy.

[0050] According to an embodiment of the present invention, a time-related reference data sequence is determined based on a sequence time step of a historical distribution network measurement data sequence, including: determining a plurality of initial reference data related to time based on the sequence time step and the measurement moment corresponding to the missing measurement data; performing data encoding on the plurality of initial reference data related to time respectively to obtain a plurality of time-related reference data; and sorting the plurality of time-related reference data according to the sequence time step to obtain a time-related reference data sequence.

[0051] According to an embodiment of the present invention, multiple initial reference data related to time can be calculated based on the measurement moment corresponding to the missing measurement data using the sequence time step. Specifically, when the measurement moment of the missing measurement data is known, the measurement moment separated from the measurement moment of the missing measurement data by a sequence time step is determined as an initial reference data, and then based on the known initial reference data, the measurement moment separated from the known initial reference data by a sequence time step is determined as another initial reference data, and so on, until all the multiple initial reference data are obtained.

[0052] According to a specific embodiment of the present invention, the initial reference data can be used to represent the measurement date. In this case, the initial reference data can be encoded using polar coordinate encoding, for example, the initial reference data can be encoded using a sine (sin) function and a cosine (cos) function.

[0053] According to an embodiment of the present invention, the sequence between a plurality of time-related reference data may be determined according to the sequence time step, and then a time-related reference data sequence may be obtained according to the sequence.

[0054] For example, The time-dependent reference data series represented by The time-dependent reference sequence represented by can be expressed as follows:

[0055] (1)

[0056] Among them, Q represents the period, which is 6.

[0057] According to an embodiment of the present invention, encoding the initial reference data by polar coordinate encoding can enhance the time cognition ability of the repair model, better reflect the periodic characteristics of the measurement data, help improve the prediction accuracy of the repair model, and further improve the repair accuracy.

[0058] According to an embodiment of the present invention, the distribution network measurement data repair method also includes: based on multiple sequence time steps, determining the repair vector to be reduced in dimension from multiple repair vectors; performing covariance analysis on the repair vector to be reduced in dimension to obtain a covariance vector; based on the covariance vector, determining the importance of each of the multiple vector elements in the repair vector to be reduced in dimension; based on the importance of each of the multiple vector elements in the repair vector to be reduced in dimension, reducing the dimension of the repair vector to be reduced in dimension to obtain a repair vector after dimension reduction.

[0059] According to an embodiment of the present invention, when the sequence time step is large, the dimension of the corresponding repair vector may be large, so the repair vector to be reduced in dimension can be determined from multiple repair vectors based on the sequence time step. For example, when the multiple sequence time steps include 1, 2, 4, and 8, the repair vectors corresponding to the sequence time steps of 2, 4, and 8 can be determined as the repair vectors to be reduced in dimension.

[0060] According to an embodiment of the present invention, before reducing the dimension of the repair vector to be reduced, the dimension of the repair vector after dimensionality reduction may be preset. Specifically, the dimension of the repair vector to be reduced may be m, and the dimension of the repair vector after dimensionality reduction may be k=m / 2.

[0061] According to an embodiment of the present invention, before performing covariance analysis on the repair vector to be reduced in dimension, each row of the repair vector to be reduced in dimension may be zero-meaned, that is, each number in each row is subtracted from the mean of the row.

[0062] According to an embodiment of the present invention, the process of performing covariance analysis on the repair vector to be reduced in dimension can be shown in the following formula (2):

[0063] (2)

[0064] Among them, C is the covariance vector, X is the repair vector to be reduced in dimension, and m is the dimension of the repair vector to be reduced in dimension.

[0065] According to an embodiment of the present invention, the importance of each of the multiple vector elements in the repair vector may be determined according to the eigenvalue of the covariance vector. Specifically, the eigenvalue of the covariance vector may be determined by linear algebra solution or by using an eigenvalue solution (eig) function.

[0066] According to an embodiment of the present invention, the repair vector to be reduced in dimension includes multiple vector elements, and accordingly, each vector element has a corresponding eigenvector in the covariance vector. When reducing the dimension of the repair vector to be reduced in dimension, the eigenvectors corresponding to the vector elements can be sorted according to the importance of each of the multiple vector elements, and a feature matrix is ​​obtained based on the eigenvectors corresponding to each of the multiple vector elements with high importance, and the repair vector after dimensionality reduction is obtained based on the feature matrix and the repair vector to be reduced in dimension. Specifically, the above process can be shown in the following formula (3):

[0067] Y=PX (3)

[0068] Among them, Y is the repair vector after dimensionality reduction, and the dimension of Y is k, P is the feature matrix, and X is the repair vector to be reduced in dimension.

[0069] According to an embodiment of the present invention, by reducing the dimensionality of the repair vector to be reduced, a repair vector with reduced dimensionality having lower complexity is obtained while retaining the richness of the repair vector, thereby improving the repair efficiency.

[0070] According to an embodiment of the present invention, a preprocessing operation is performed on the repair vector or the repair vector after dimensionality reduction to obtain a target repair vector, including: performing data denoising operations and data normalization operations on the repair vector or the repair vector after dimensionality reduction.

[0071] According to an embodiment of the present invention, when the measurement equipment of the distribution network collects and transmits measurement data, the measurement equipment itself and the environment in which the measurement equipment is located may affect the measurement data. Therefore, it is necessary to perform data denoising operations on the repair vector or the repair vector after dimensionality reduction to restore the real measurement data as much as possible and ensure the repair quality.

[0072] According to an embodiment of the present invention, when performing a data denoising operation on a repair vector or a repair vector after dimensionality reduction, an approximately symmetrical compact support orthogonal wavelet (Symlet) denoising function can be used to perform preliminary processing on the measurement data. Since the Symlet denoising function has good symmetry, distortion can be reduced when performing a data denoising operation.

[0073] According to an embodiment of the present invention, after performing a data denoising operation on the repair vector or the repair vector after dimensionality reduction, a denoised repair vector or a denoised dimensionality reduced repair vector can be obtained, and then a data normalization operation can be performed on the denoised repair vector or the denoised dimensionality reduced repair vector.

[0074] According to an embodiment of the present invention, the data normalization operation can map the data to a fixed range interval according to a certain mapping relationship, such as [0, 1] or [-1, 1], which can eliminate the impact of singular data on prediction.

[0075] According to an embodiment of the present invention, when performing a data normalization operation on a denoised repair vector or a denoised and reduced repair vector, a minimum-maximum (Min-Max) normalization method may be used to map the denoised repair vector or the denoised and reduced repair vector to the [-1,1] interval according to a mapping relationship to implement a data normalization operation. Specifically, the mapping function for mapping may be as shown in the following formula (4):

[0076] (4)

[0077] Among them, y max =1,y min =-1, x is the denoised restoration vector or the restoration vector after denoising and dimensionality reduction, and y is the normalized denoised restoration vector or the restoration vector after denoising and dimensionality reduction.

[0078] According to an embodiment of the present invention, a plurality of target repair vectors are respectively input into a repair model to obtain a plurality of repair data, including: a plurality of target repair vectors are respectively input into a repair model matching the target repair vectors to obtain a plurality of repair data.

[0079] According to an embodiment of the present invention, a repair model may be set for each of the different sequence time steps, and according to the sequence time step corresponding to the target repair vector, the target repair vector is input into the repair model corresponding to the sequence time step.

[0080] According to an embodiment of the present invention, the repair model may be obtained through training. In order to improve the pertinence of the repair model, the sequence time steps of the multiple sample data sequences used for training of each repair model are the same, and the sequence time steps of the sample data sequences used for training of the multiple repair models are different.

[0081] According to an embodiment of the present invention, by training multiple repair models with targeted sequence time steps and inputting multiple target repair vectors into corresponding repair models respectively, the accuracy of repair data is improved, thereby improving the repair precision.

[0082] Specifically, an embodiment of the present invention trains four repair models with sequence time steps of 1, 2, 4, and 8 respectively, and inputs the target repair vectors corresponding to each sequence time step into the repair models respectively, so that the input data of the repair model contains more information, thereby reducing the impact of non-real data on the repair accuracy of the repair model and avoiding the occurrence of error accumulation problems.

[0083] According to an embodiment of the present invention, although the four repair models with sequence time steps of 1, 2, 4, and 8 respectively avoid the error accumulation problem when repairing missing data at multiple moments, the method of constructing the target repair vector of the three repair models with sequence time steps of 2, 4, and 8 respectively, relative to the method of constructing the target repair vector of the repair model with a sequence time step of 1, will slightly reduce the repair accuracy of a single repair model.

[0084] Therefore, the embodiment of the present invention adopts the covariance cross fusion (CI) algorithm to nonlinearly fuse the four restoration models to improve the overall restoration effect.

[0085] According to an embodiment of the present invention, target repair data matching the missing measurement data is obtained based on multiple repair data, including: determining the fusion functions of multiple repair models based on the variance value of each repair model; taking the minimum fusion function of the repair model as the goal, determining the weight value corresponding to the weight parameter of each repair model based on the weight limitation conditions and the variance value of each repair model; and determining the target repair data based on multiple repair data, the weight value of each repair model and the variance value of each repair model.

[0086] According to an embodiment of the present invention, the fusion function includes a variance value parameter corresponding to the variance value of each restoration model and a weight parameter of each restoration model. Specifically, the fusion function can be shown as the following formula (5):

[0087] (5)

[0088] Among them, P CI is the function value of the fusion function, , , , They represent the weight parameters of each repair model, and P1, P2, P4, and P8 represent the variance values ​​of each repair model.

[0089] According to an embodiment of the present invention, the smaller the fusion function is, the more stable and reliable the estimation is. Therefore, the process of determining the weight value corresponding to the weight parameter of the repair model can be converted into an optimization problem with the goal of minimizing the fusion function of the repair model. Specifically, the above process can be shown in the following formula (6):

[0090] (6)

[0091] Here, tr represents the trace of the matrix.

[0092] According to an embodiment of the present invention, a weight limiting condition may be used to limit the value of each weight value. Specifically, the weight limiting condition may be as shown in the following formula (7):

[0093] (7)

[0094] According to an embodiment of the present invention, the sequential least squares method may be used to optimize the fusion function so that the fusion function is minimized, and the weight value corresponding to the weight parameter of each restoration model is obtained.

[0095] According to an embodiment of the present invention, the process of determining target repair data can be shown as the following formula (8):

[0096] (8)

[0097] Among them, Z CI Repair data for the target, Z n (1) is the repair data output by the repair model with a sequence time step of 1, Z n (2) is the repair data output by the repair model with a sequence time step of 2, Z n (4) is the repair data output by the repair model with a sequence time step of 4, Z n (8) The repair data output by the repair model with a sequence time step of 8.

[0098] According to an embodiment of the present invention, a fusion function is used to accurately determine multiple weight values, so that multiple repair data can be accurately fused based on the weight values, thereby improving the accuracy of the target repair data and improving the repair efficiency.

[0099] According to an embodiment of the present invention, the variance values ​​of multiple repair models are determined, including: determining sample input data and labels from a sample data sequence; inputting the sample input data into the repair model to obtain sample repair data corresponding to the label; and obtaining the variance value of the repair model based on the sample repair data and the label.

[0100] According to an embodiment of the present invention, the variance values ​​of the multiple repair models may be determined in the process of training the multiple repair models.

[0101] According to an embodiment of the present invention, the label is used as real measurement data. The sample repair data can represent the predicted output obtained by the repair model based on the sample input data.

[0102] According to an embodiment of the present invention, , , , They represent the predicted outputs of the four repair models at the first moment with sequence time steps of 1, 2, 4, and 8, respectively, where n is the sequence length and the variance values ​​of the four repair models are , , , .in, The calculation process of is shown in the following formula (9), and the calculation process of other variances is similar.

[0103] (9)

[0104] in, For labels, Repair data for the sample.

[0105] According to the embodiment of the present invention, by determining the variance values ​​of the respective restoration models, the accuracy of the fusion function is improved, thereby improving the restoration precision.

[0106] Figure 2 A flow chart of repairing measurement data according to a specific embodiment of the present invention is shown.

[0107] like Figure 2 As shown, repairing measurement data includes operations S210 to S270.

[0108] In operation S210, measurement data at the current moment and a set of historical data are obtained.

[0109] According to an embodiment of the present invention, the historical data set may include historical measurement data.

[0110] In operation S220, it is determined whether the measurement data at the current moment is missing. If it is determined that the measurement data at the current moment is missing, operation S230 is performed, otherwise operation S260 is performed.

[0111] According to an embodiment of the present invention, when the measurement data at the current moment is not missing, the measurement data at the current moment is identified as T.

[0112] In operation S230 , it is determined that the measurement data at the current moment is missing measurement data.

[0113] In operation S240 , a historical power distribution network measurement data sequence is constructed based on the historical data set.

[0114] In operation S250, target repair data is generated using a repair model based on a historical distribution network measurement data sequence and a reference data sequence.

[0115] According to an embodiment of the present invention, the target repair data may be identified as F.

[0116] In operation S260, before the measurement data of the next moment arrives, the historical data set is updated.

[0117] According to an embodiment of the present invention, when the measurement data at the current moment is not missing, the historical data set is updated using the measurement data at the current moment; when the measurement data at the current moment is missing, the historical data set is updated using the target repair data.

[0118] In operation S270, when the measurement data of the next moment arrives, it is taken as a new current moment, and operation S210 is performed.

[0119] According to an embodiment of the present invention, a repair model that matches the target repair vector is obtained by training using the following training method: using the objective function of the particle swarm optimization algorithm as the loss function, and using multiple sample data sequences with the same sequence time step as the target repair vector to train the initial repair model to obtain the repair model.

[0120] According to an embodiment of the present invention, the initial repair model adopts a two-layer structure, the number of neurons in the hidden layer is 96*3, and each initial repair unit of the initial repair model is composed of a forget gate, an input gate, an output gate, and a cell state. Specifically, the calculation process of the three gate structures can be shown in the following formulas (10) to (12):

[0121] (10)

[0122] (11)

[0123] (12)

[0124] Among them, f(t), i(t), o(t) represent the outputs of the forget gate, input gate, and output gate at time t, respectively; x(t) represents the input of the initial repair unit at time t; h(t-1) represents the hidden layer state of the initial repair unit at time t-1; W f , W i , W o Respectively represent the weights of the input x(t) to each gate; U f , U i , U o Respectively represent the weights from the hidden layer h(t-1) to each gate; b f 、b i 、b o Respectively represent the bias values ​​corresponding to each gate; Represents the activation (Sigmoid) function.

[0125] According to an embodiment of the present invention, the updating calculation process of the cell state can be shown in the following formulas (13) to (14):

[0126] (13)

[0127] (14)

[0128] in, is the candidate state (intermediate state) of the initial repair unit at time t, indicating the preliminary feature extraction of x(t) and h(t-1) by the initial repair unit at time t; W c , U c 、b c They represent the weight of x(t), the weight of h(t-1), and the bias value in the feature extraction process respectively; c(t-1) and c(t) represent the cell states at time t-1 and time t respectively; tanh is the hyperbolic tangent function; the symbol represents the Hadamard product.

[0129] According to an embodiment of the present invention, the calculation process of the hidden layer state h(t) of the initial repair unit at time t is shown in the following formula (15):

[0130] (15)

[0131] According to an embodiment of the present invention, before training the initial repair model, the sample data sequence used to train the initial repair model may also be preprocessed, such as data normalization, which can eliminate singular sample data in the sample data sequence and speed up the gradient descent method, thereby obtaining the optimal solution more quickly and improving the convergence speed and training accuracy of the initial repair model.

[0132] According to an embodiment of the present invention, before the initial repair model is trained using multiple sample data sequences, a training set, a validation set, and a test set can be constructed based on the preprocessed multiple sample data sequences, wherein the training set is used to train the initial repair model and determine the parameters of the repair model, the validation set is used as the data basis for hyperparameter tuning, and the test set is used to test the accuracy of the initial repair model. Specifically, the preprocessed multiple sample data sequences can be divided into a training set, a validation set, and a test set in a ratio of 6:1:1.

[0133] According to an embodiment of the present invention, real measurement data can be used to construct a label matrix for training the initial repair model. Specifically, for the initial repair model corresponding to the sequence time step of 1, a 1*n-dimensional label matrix is ​​constructed. ,in , represents n data to be predicted; for the initial repair models corresponding to the sequence time steps of 2, 4, and 8, respectively, a p*n-dimensional label matrix is ​​constructed , where p represents the sequence time step, p=2, 4, 8, , i=1,2,…,p, They respectively represent the measurement data at the first moment, the measurement data at the second moment, ..., the measurement data at the pth moment of the n time periods to be predicted.

[0134] According to an embodiment of the present invention, before training the initial repair model, a deep network initialization (Xavier initialization) method may be used to initialize the weights of the initial repair model.

[0135] According to an embodiment of the present invention, when training the initial repair model, a training function such as a trainnetwork function can be used to train each initial repair model separately, and an optimizer (Adam) algorithm can be selected as the optimizer for training. At the same time, the gradient descent method is used to update the weights of the initial repair model, and the particle swarm algorithm is used to tune the hyperparameters of the initial repair model. The initial repair model is reconstructed based on the optimized parameters to obtain a trained repair model.

[0136] According to an embodiment of the present invention, in a particle swarm algorithm (PSO), each solution is abstracted as a particle, and the search space is assumed to be d-dimensional (there are d independent variables in the optimization function). Each particle has a position attribute, a velocity attribute, and a fitness value determined by the optimization function. Among them, the velocity attribute determines the direction and distance of the search. Specifically, the process of updating the particle attributes can be shown in the following formulas (16) to (17):

[0137] (16)

[0138] (17)

[0139] in, and denote the j-th component of the velocity vector of the ith particle at the t-th and t+1-th iterations respectively; and represents the j-th dimension component of the position vector of the ith particle at the t-th and t+1-th iterations respectively; w is the inertia weight; c1 and c2 are positive learning factors; r1 and r2 are uniformly distributed random numbers; and They represent the j-th dimension components of the individual optimal solution and the group optimal solution at the t-th iteration respectively; is the self-cognition term, indicating that particle updates are affected by their own experience; is the group cognition term, which means that the particle update is affected by the best experience of other particles, that is, the role of information sharing within the group.

[0140] According to an embodiment of the present invention, the loss function may be a root mean square error (RMSE) function. Specifically, the loss function may be shown in the following formula (18):

[0141] (18)

[0142] Where W is the total number of measurement data to be predicted, Z t is the true value of the measured data to be predicted, Z t * It is the predicted value of the measurement data output by the initial repair model.

[0143] According to an embodiment of the present invention, when verifying the performance of a trained repair model, in order to comprehensively consider the relative error and absolute error between the real measurement data and the sample repair data, the mean absolute percentage error (MAPE) and the root mean square error (RMSE) can be used to verify the repair accuracy of the repair model. Specifically, the process of determining the mean absolute percentage error can be shown in the following formula (19), and the process of determining the root mean square error can be shown in the following formula (20):

[0144] (19)

[0145] (20)

[0146] Where N represents the total number of data. represents the sample repair data, y i Indicates the actual measurement data.

[0147] According to an embodiment of the present invention, the initial repair model is trained using a particle swarm algorithm to improve the training speed of the initial repair model.

[0148] Figure 3 A flowchart of adjusting the hyperparameters of the initial repair model according to an embodiment of the present invention is shown.

[0149] like Figure 3 As shown, adjusting the initial repair model hyperparameters includes operations S310 to S380.

[0150] In operation S310, the topological structure and parameters of the initial repair model are set, and the synaptic weights and particle swarm are initialized.

[0151] According to an embodiment of the present invention, the particle swarm may characterize the hyperparameters of the initial repair model.

[0152] In operation S320, the current iteration number is set to 1.

[0153] In operation S330, a fitness function is selected.

[0154] According to an embodiment of the present invention, the training error may be used as a fitness function.

[0155] In operation S340 , the fitness value of each particle is evaluated, and the optimal fitness and position of a single particle and all particles are stored.

[0156] In operation S350, it is determined whether the convergence criterion is satisfied. If it is determined that the convergence criterion is satisfied, operation S360 is performed, otherwise operation S370 is performed.

[0157] In operation S360, the optimal result and the number of iterations are output.

[0158] In operation S370, the position and velocity of each particle are updated.

[0159] In operation S380, the number of iterations is incremented by 1.

[0160] According to an embodiment of the present invention, based on different multiple sequence time steps, multiple historical distribution network measurement data sequences matching the missing measurement data are determined from a historical data set, including: determining the measurement moment corresponding to the missing measurement data; and determining multiple historical distribution network measurement data sequences from the historical data set based on the measurement moment and different multiple sequence time steps.

[0161] According to an embodiment of the present invention, the historical distribution network measurement data sequence can be obtained by using the sequence time step based on the measurement time corresponding to the missing measurement data. Specifically, when the measurement time of the missing measurement data is known, according to the measurement time of each of the multiple historical measurement data in the historical data set, multiple historical measurement data whose measurement time is separated from the measurement time of the missing measurement data by one or more sequence time steps are determined, thereby obtaining the historical distribution network measurement data sequence.

[0162] According to an embodiment of the present invention, by dynamically determining a historical distribution network measurement data sequence corresponding to missing measurement data according to the measurement time, the reference of the historical distribution network measurement data sequence is improved, thereby improving the repair accuracy.

[0163] In order to verify the performance of the repair model of the embodiment of the present invention, examples can be set from the following two aspects: 1) Compare and analyze the repair accuracy of the repair model of the embodiment of the present invention with that of the ANN, GRU, LSTM, and PSO-LSTM models at different time steps to verify the effectiveness of the proposed model. 2) Compare and analyze the repair accuracy of the repair model of the embodiment of the present invention with that of 2, 4, and 8-step PCA-LSTM (respectively denoted as PCA-LSTM2, PCA-LSTM4, and PCA-LSTM8) and a model without PCA (LSTM-CI) at different time steps to illustrate the impact of PCA and CI methods on model performance.

[0164] According to an embodiment of the present invention, a 10kV power distribution system in a certain region is selected as the experimental object, and the A-phase current measurement data of the load switch at the feeder outlet from March 1 to May 31, 2022 is selected as the sample data set. The sampling interval is 15 minutes, and 96 sample points are collected every day, that is, the number of data samples included is 8832. Among them, the first 80% of the data is used as the training set and validation set, and the last 20% of the data is used as the test set to verify the model performance.

[0165] According to an embodiment of the present invention, in order to verify the effectiveness of the repair model of an embodiment of the present invention, the comparison models set are all single-step models (i.e., outputting a single-moment repair value), and a recursive method is adopted to implement multi-step repair, that is, the current-moment repair value is used as historical data for repairing the next-moment data and input into the trained model, and the repair results of multiple moments are gradually recursively obtained.

[0166] As the repair time step increases, the relative repair errors (measured by MAPE value) of all models increase to varying degrees. Although the relative repair error of the repair model of the embodiment of the present invention also increases, the error increase trend is significantly lower than that of other models, and the relative repair error and absolute repair error (measured by RMSE value) of the model at different time steps can be kept at the lowest, which also verifies that the repair model of the embodiment of the present invention has good repair accuracy and stability in multi-time missing data repair. In addition, the repair error of the repair model of the embodiment of the present invention at different time steps is also lower than that of the LSTM model, verifying the effect of the PSO algorithm on improving the model accuracy.

[0167] In the experimental results, the relative repair error under the long step length condition is significantly higher than the relative repair error of the short step length. This is because as the repair step length increases, the more non-real information is contained in the model input data, the more cumulative errors introduced will gradually increase, resulting in reduced multi-step repair performance, increased relative repair errors, and the longer the time step, the greater the relative repair error. The repair model of the embodiment of the present invention realizes the fusion of four PSO-LSTM prediction models with 1, 2, 4, and 8 sequence time steps, and the three models of 2, 4, and 8 sequence time steps use a unique method when constructing the target repair vector, so that the input data of the repair model contains more information, reducing the influence of non-real data information on the prediction accuracy of the repair model, thereby effectively suppressing the influence of cumulative errors on model performance, so that the repair model can still have a good repair effect when performing multi-moment repair.

[0168] Figure 4 A repair curve diagram of the repair model of the embodiment of the present invention and the comparison model with a sequence time step of 96 is shown.

[0169] like Figure 4 As shown in the figure, under the same time step conditions, the repair curve fitting effect of the repair model (PCA-LSTM-CI) of the embodiment of the present invention is always better than that of other comparison models (PSO-LSTM, LSTM, GRU, ANN), especially in the peak, trough, and time periods with large current fluctuations. The model captures the current change trend more accurately and has better fitting ability. Compared with single-step repair, the repair performance of each model under the condition of a sequence time step of 96 has declined to a certain extent due to the accumulation of errors, but the repair model of the embodiment of the present invention has a relatively small decline, and can still maintain a relatively good fitting effect. The repair performance advantage under multi-step conditions is more obvious.

[0170] According to an embodiment of the present invention, in order to verify the influence of the PCA and CI methods on the repair performance in the repair model of the embodiment of the present invention, the comparative experiments set up include 2, 4, and 8-step PCA-LSTM (respectively denoted as PCA-LSTM2, PCA-LSTM4, and PCA-LSTM8) and a model without PCA (LSTM-CI). The comparison models PCA-LSTM2, PCA-LSTM4, and PCA-LSTM8 adopt a recursive method to achieve multi-step repair (referring to repair after more than 2, 4, and 8 time steps, respectively). Except for removing the PCA dimensionality reduction operation, the LSTM-CI model is consistent with the repair model of the embodiment of the present invention.

[0171] Compared with the repair results of the LSTM-CI model, the error indicators of the PCA-LSTM-CI model have all decreased. The experimental results show that the introduction of the PCA dimensionality reduction method for feature extraction helps to improve the repair performance of the model.

[0172] Compared with the multi-step LSTM models (PCA-LSTM2, PCA-LSTM4, PCA-LSTM8), the repair performance of the PCA-LSTM-CI model using CI fusion is further improved. The experimental results show that the constructed PCA-LSTM-CI model can further enhance the repair performance of the model through nonlinear fusion.

[0173] Based on the above distribution network measurement data repair method, the present invention also provides a distribution network measurement data repair device. Figure 5 The device is described in detail.

[0174] Figure 5 The structure block diagram of the distribution network measurement data repair device according to an embodiment of the present invention is schematically shown.

[0175] like Figure 5 As shown, the distribution network measurement data repair device 500 of this embodiment includes a history determination module 510 , a vector generation module 520 , a vector dimension reduction module 530 , a repair input module 540 and a repair generation module 550 .

[0176] The history determination module 510 is used to determine multiple historical power distribution network measurement data sequences matching the missing measurement data from the historical data set based on multiple different sequence time steps. In one embodiment, the history determination module 510 can be used to perform the operation S110 described above, which will not be described in detail here.

[0177] The vector generation module 520 is used to generate a repair vector based on the historical distribution network measurement data sequence and the reference data sequence matching the historical distribution network measurement data sequence. In one embodiment, the vector generation module 520 can be used to perform the operation S120 described above, which will not be described in detail here.

[0178] The vector dimension reduction module 530 is used to obtain multiple target repair vectors based on the repair vector and the repair vector after dimension reduction, wherein the repair vector after dimension reduction is obtained by reducing the dimension of the repair vector to be reduced, and the repair vector to be reduced is determined from multiple repair vectors. In one embodiment, the vector dimension reduction module 530 can be used to perform the operation S130 described above, which will not be repeated here.

[0179] The repair input module 540 is used to input multiple target repair vectors into the repair model respectively to obtain multiple repair data. In one embodiment, the repair input module 540 can be used to perform the operation S140 described above, which will not be repeated here.

[0180] The repair generation module 550 is used to obtain target repair data matching the missing measurement data based on the multiple repair data. In one embodiment, the repair generation module 550 can be used to perform the operation S150 described above, which will not be described in detail here.

[0181] According to an embodiment of the present invention, the vector generation module 520 includes a time series determination submodule, an external sequence determination submodule and a vector generation submodule.

[0182] The time series determination submodule is used to determine a time-related reference data sequence based on the sequence time step of the historical distribution network measurement data sequence.

[0183] The external sequence determination submodule is used to determine a reference data sequence related to external factors based on the measurement data type of the historical distribution network measurement data sequence, wherein the external factors include at least one of the following: environment, equipment.

[0184] The vector generation submodule is used to obtain a repair vector based on a historical distribution network measurement data sequence, a time-related reference data sequence, and a reference data sequence related to external factors.

[0185] According to an embodiment of the present invention, the time series determination submodule includes an initial determination unit, a data encoding unit and a data sorting unit.

[0186] The initial determination unit is used to determine a plurality of initial reference data related to time based on the sequence time step and the measurement time corresponding to the missing measurement data.

[0187] The data encoding unit is used to respectively perform data encoding on a plurality of initial reference data related to time to obtain a plurality of reference data related to time.

[0188] The data sorting unit is used to sort a plurality of time-related reference data according to a sequence time step to obtain a time-related reference data sequence.

[0189] According to an embodiment of the present invention, the distribution network measurement data repairing device further includes a dimension reduction determination module, a covariance determination module, an importance determination module and a dimension reduction module.

[0190] The dimension reduction determination module is used to determine a repair vector to be reduced in dimension from multiple repair vectors based on multiple sequence time steps.

[0191] The covariance determination module is used to perform covariance analysis on the repair vector to be reduced in dimension to obtain a covariance vector.

[0192] The importance determination module is used to determine the importance of each of multiple vector elements in the repair vector to be reduced in dimension based on the covariance vector.

[0193] The dimension reduction module is used to reduce the dimension of the repair vector to be reduced based on the importance of each of the multiple vector elements in the repair vector to be reduced, so as to obtain the repair vector after dimension reduction.

[0194] According to an embodiment of the present invention, the repair input module 540 includes a repair input submodule.

[0195] The repair input submodule is used to input multiple target repair vectors into the repair models that match the target repair vectors, respectively, to obtain multiple repair data, wherein the sequence time step of the multiple sample data sequences used for training of each repair model is the same, and the sequence time step of the sample data sequences used for training of the multiple repair models are different.

[0196] According to an embodiment of the present invention, the repair generation module 550 includes a fusion determination submodule, a weight determination submodule and a repair generation submodule.

[0197] The fusion determination submodule is used to determine the fusion function of multiple repair models based on the variance value of each repair model, wherein the fusion function includes the variance value parameter corresponding to the variance value of each repair model and the weight parameter of each repair model.

[0198] The weight determination submodule is used to determine the weight value corresponding to the weight parameter of each repair model based on the weight limitation conditions and the variance values ​​of the repair models, with the goal of minimizing the fusion function of the repair model.

[0199] The repair generation submodule is used to determine target repair data based on multiple repair data, a weight value of each repair model and a variance value of each repair model.

[0200] According to an embodiment of the present invention, the repair generation module 550 further includes a data determination submodule, a sample repair submodule and a variance determination submodule.

[0201] The data determination submodule is used to determine sample input data and labels from the sample data sequence, wherein the labels serve as real measurement data.

[0202] The sample repair submodule is used to input the sample input data into the repair model to obtain the sample repair data corresponding to the label.

[0203] The variance determination submodule is used to obtain the variance value of the repair model based on the sample repair data and labels.

[0204] According to an embodiment of the present invention, the restoration model that matches the target restoration vector is obtained through training by a model training module.

[0205] The model training module is used to train the initial repair model using the objective function of the particle swarm optimization algorithm as the loss function and multiple sample data sequences with the same sequence time step as the target repair vector to obtain the repair model.

[0206] According to an embodiment of the present invention, the history determination module 510 includes a time determination submodule and a history determination submodule.

[0207] The time determination submodule is used to determine the measurement time corresponding to the missing measurement data.

[0208] The history determination submodule is used to determine multiple historical distribution network measurement data sequences from a historical data set based on the measurement time and multiple different sequence time steps.

[0209] According to an embodiment of the present invention, any multiple modules among the history determination module 510, the vector generation module 520, the vector dimension reduction module 530, the repair input module 540 and the repair generation module 550 can be combined into one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module. According to an embodiment of the present invention, at least one of the history determination module 510, the vector generation module 520, the vector dimension reduction module 530, the repair input module 540 and the repair generation module 550 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation modes of software, hardware and firmware or in a suitable combination of any of them. Alternatively, at least one of the history determination module 510, the vector generation module 520, the vector dimension reduction module 530, the repair input module 540 and the repair generation module 550 can be at least partially implemented as a computer program module, which can perform the corresponding function when executed.

[0210] According to an embodiment of the present invention, the program code for executing the computer program provided by the embodiment of the present invention can be written in any combination of one or more programming languages. Specifically, these computing programs can be implemented using high-level process and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, Java, C++, python, "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on the remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (e.g., using an Internet service provider to connect through the Internet).

[0211] The flowchart and block diagram in the accompanying drawings illustrate the possible architecture, functions and operations of the methods and devices according to various embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0212] It will be appreciated by those skilled in the art that the features described in the various embodiments of the present invention may be combined and / or combined in various ways, even if such combinations and / or combinations are not explicitly described in the present invention. In particular, without departing from the spirit and teachings of the present invention, the features described in the various embodiments of the present invention may be combined and / or combined in various ways. All of these combinations and / or combinations fall within the scope of the present invention.

[0213] The embodiments of the present invention are described above. However, these embodiments are only for the purpose of illustration, and are not intended to limit the scope of the present invention. Although each embodiment is described above, it does not mean that the measures in each embodiment cannot be used in combination advantageously. Without departing from the scope of the present invention, those skilled in the art may make various substitutions and modifications, which should all fall within the scope of the present invention.

Claims

1. A method for repairing distribution network measurement data, characterized in that: The method comprises: Based on different multiple sequence time steps, determining multiple historical distribution network measurement data sequences matching the missing measurement data from the historical data set; Determining a time-related reference data sequence based on the sequence time step of the historical distribution network measurement data sequence; Based on the measurement data type of the historical distribution network measurement data sequence, a reference data sequence related to external factors is determined, wherein the external factors include at least one of the following: environment, equipment; Obtaining a repair vector based on the historical distribution network measurement data sequence, the time-related reference data sequence and the external factor-related reference data sequence; Based on the repair vector and the repair vector after dimension reduction, a plurality of target repair vectors are obtained, wherein the repair vector after dimension reduction is obtained by reducing the dimension of the repair vector to be reduced, and the repair vector to be reduced is determined from the plurality of repair vectors; Inputting the plurality of target repair vectors into the repair models matching the target repair vectors respectively to obtain a plurality of repair data, wherein the sequence time step of the plurality of sample data sequences used for training of each of the repair models is the same, and the sequence time step of the sample data sequences used for training of the plurality of repair models are different; Based on the variance value of each of the repair models, determining a fusion function of the plurality of repair models, wherein the fusion function includes a variance value parameter corresponding to the variance value of each of the repair models and a weight parameter of each of the repair models; Taking the minimization of the fusion function of the repair model as the goal, based on the weight limiting conditions and the variance values ​​of the respective repair models, determining the weight value corresponding to the weight parameter of each of the repair models; and Target repair data is determined based on the plurality of repair data, the weight value of each repair model and the variance value of each repair model.

2. The method according to claim 1, characterized in that: The determining of a time-related reference data sequence based on the sequence time step of the historical distribution network measurement data sequence comprises: Determining a plurality of initial reference data related to time based on the sequence time step and the measurement time corresponding to the missing measurement data; respectively performing data encoding on the multiple time-related initial reference data to obtain multiple time-related reference data; and The plurality of time-dependent reference data are sorted according to the sequence time step to obtain the time-dependent reference data sequence.

3. The method according to claim 1, characterized in that The method further comprises: Based on a plurality of the sequence time steps, determining the repair vector to be reduced in dimension from a plurality of repair vectors; Performing covariance analysis on the repair vector to be reduced in dimension to obtain a covariance vector; Determining the importance of each of the plurality of vector elements in the repair vector to be reduced in dimension based on the covariance vector; Based on the importance of each of the multiple vector elements in the repair vector to be reduced in dimension, the repair vector to be reduced in dimension is reduced in dimension to obtain the repair vector after dimension reduction.

4. The method according to claim 1, characterized in that The method further comprises: Determining sample input data and labels from the sample data sequence, wherein the labels serve as real measurement data; Inputting the sample input data into the repair model to obtain sample repair data corresponding to the label; Based on the sample repair data and the label, a variance value of the repair model is obtained.

5. The method according to claim 1, characterized in that The repair model matching the target repair vector is trained by the following training method: The objective function of the particle swarm optimization algorithm is used as the loss function, and a plurality of sample data sequences with the same sequence time step as the target repair vector are used to train the initial repair model to obtain the repair model.

6. The method according to claim 1, characterized in that The method of determining a plurality of historical distribution network measurement data sequences matching the missing measurement data from the historical data set based on a plurality of different sequence time steps includes: Determining the measurement time corresponding to the missing measurement data; and Based on the measurement time and a plurality of different sequence time steps, a plurality of the historical power distribution network measurement data sequences are determined from the historical data set.

7. A distribution network measurement data repair device, characterized in that: The device comprises: A historical determination module, for determining, from a historical data set, a plurality of historical distribution network measurement data sequences matching the missing measurement data based on a plurality of different sequence time steps; A vector generation module, configured to generate a repair vector based on the historical distribution network measurement data sequence and a reference data sequence matching the historical distribution network measurement data sequence; A vector dimension reduction module, used to obtain multiple target repair vectors based on the repair vector and the repair vector after dimension reduction, wherein the repair vector after dimension reduction is obtained by reducing the dimension of the repair vector to be reduced, and the repair vector to be reduced is determined from the multiple repair vectors; A repair input module, used to input the plurality of target repair vectors into the repair model respectively to obtain a plurality of repair data; and A repair generation module, used for obtaining target repair data matching the missing measurement data based on the plurality of repair data; The vector generation module comprises: A time series submodule, for determining a time-related reference data sequence based on a sequence time step of the historical distribution network measurement data sequence; The external sequence determination submodule is used to determine a reference data sequence related to external factors based on the measurement data type of the historical distribution network measurement data sequence, wherein the external factors include at least one of the following: environment, equipment; A vector generation submodule, configured to obtain a repair vector based on the historical distribution network measurement data sequence, the time-related reference data sequence and the external factor-related reference data sequence; The repair input module comprises: An input repair submodule, used for inputting the plurality of target repair vectors into the repair models matching the target repair vectors respectively, to obtain a plurality of repair data, wherein the sequence time step of the plurality of sample data sequences used for training of each of the repair models is the same, and the sequence time step of the sample data sequences used for training of the plurality of repair models are different; The repair generation module comprises: A fusion determination submodule, used to determine a fusion function of the plurality of restoration models based on the variance value of each restoration model, wherein the fusion function includes a variance value parameter corresponding to the variance value of each restoration model and a weight parameter of each restoration model; A weight determination submodule, for determining a weight value corresponding to a weight parameter of each of the repair models based on a weight limiting condition and a variance value of each of the repair models with the goal of minimizing the fusion function of the repair model; and The repair generation submodule is used to determine target repair data based on the plurality of repair data, the weight value of each repair model and the variance value of each repair model.

Images