Quality evaluation method for multivariate time series data of power distribution network

Abstract

The invention belongs to the technical field of power distribution network data quality management, and particularly relates to a power distribution network multivariate time series data-oriented quality evaluation method, which comprises the following steps of: selecting power distribution network multivariate time series data quality evaluation dimensions based on power distribution network data characteristics and distribution characteristics; comprising integrity, normalization, consistency, timeliness, uniqueness and accuracy; designing an evaluation method of the power distribution network multivariate time series data quality evaluation dimension; designing a weight distribution scheme of each dimension, and determining the weight of each dimension; and according to the quality score of each dimension and the weight of each dimension, calculating the total score of the quality of the to-be-measured data, and evaluating the data quality grade. According to the method, multi-level, multi-azimuth and multi-angle analysis is carried out on the power distribution network data, so that one-sidedness and limitation in a traditional evaluation method are avoided, and the data quality can be reflected more truly by the result, thereby optimizing the enterprise decision flow, promoting the intelligent development of the power distribution network, and realizing more efficient and sustainable power data management.

Description

Technical Field

[0001] The present invention belongs to the technical field of distribution network data quality management, and in particular relates to a quality assessment method for multivariate time series data of a distribution network. Background Art

[0002] The modernization and intelligence of distribution networks are becoming an important trend in the development of the power industry. The introduction of technologies such as the Internet of Things and cloud computing has effectively reduced the cost of power transmission and significantly optimized the scheduling and management of resources. Through real-time monitoring and data analysis, power companies can respond more flexibly to changes in electricity demand and improve service quality and efficiency. With the rapid expansion of distribution networks and the application of large-scale sensors, the amount of data is growing exponentially. Although this surge in data provides more information support for power management, it also brings serious data quality problems. For example, problems such as missing data, redundancy, and outliers frequently occur, seriously affecting the reliability and availability of data. These quality problems not only increase the burden on enterprises in data processing, but also pose a threat to the scientific nature of decision-making and the effectiveness of management.

[0003] In the process of distribution network construction, the importance of ensuring data quality has become increasingly prominent. The quality of data is directly related to the decision-making ability of enterprises in operation and management, especially when making investment decisions, resource allocation and risk assessment, accurate data analysis is particularly critical. However, the existing data quality evaluation system mainly focuses on the safety and power quality of the power system, and lacks in-depth discussion on evaluation dimensions such as data accuracy and completeness. The data quality evaluation frameworks and models used in existing studies mostly make qualitative analysis of data quality issues, fail to fully consider the correlation between evaluation dimensions, and lack systematic quantitative analysis capabilities, which makes it impossible to provide effective evaluation and improvement solutions when facing complex multivariate time series data.

[0004] Therefore, there is an urgent need for a quality assessment method for multivariate time series data of distribution networks and to establish a reasonable data quality evaluation system, so as to optimize the enterprise decision-making process, promote the intelligent development of distribution networks, and achieve more efficient and sustainable power management. Summary of the invention

[0005] The object of the present invention is to provide a quality assessment method for multivariate time series data of a distribution network, characterized in that it comprises the following steps:

[0006] Step S101: Based on the data characteristics and distribution features of the distribution network, select the quality evaluation dimension of the multivariate time series data of the distribution network;

[0007] Step S102: designing an evaluation method for the distribution network multivariate time series data quality evaluation dimension, including: dividing each dimension into sub-dimensions, and calculating the quality score of each dimension;

[0008] Step S103: designing a weight distribution scheme for each dimension and determining the weight of each dimension; the designing a weight distribution scheme for each dimension includes: distributing sub-dimension weights and distributing combined weights of each dimension;

[0009] Step S104: Calculate the total quality score of the data to be tested according to the quality score of each dimension and the weight of each dimension, and evaluate the data quality level;

[0010] The distribution network multivariate time series data quality evaluation dimensions include: completeness, standardization, consistency, timeliness, uniqueness and accuracy.

[0011] The step S102 of dividing each dimension into sub-dimensions and calculating the quality score of each dimension includes:

[0012] The integrity dimension is divided into three sub-dimensions, including: attribute integrity C com1 , Record Integrity C com2 and numerical integrity C com3 ;

[0013] The attribute integrity C com1 for:

[0014]

[0015] The record integrity C com2 for:

[0016]

[0017] The numerical integrity of C com3 for:

[0018]

[0019] Where D 11 is the number of non-missing attributes in the given data set, and n is the number of attributes in the complete data set; D 12 is the number of records in the actual data set, m is the total number of records to be collected based on the start and end time of data collection and the sampling frequency; Q(y ij ) indicates whether the data yij in the test data set is missing:

[0020]

[0021] The step S102 of dividing each dimension into sub-dimensions and calculating the quality score of each dimension includes:

[0022] The normative dimension is divided into two sub-dimensions, including: Type normativeness C N1 and accuracy specification CN2 ;

[0023] The type specification C N1 for:

[0024]

[0025] The accuracy specification C N2 for:

[0026]

[0027] In the formula, M(y ij ) is used to determine the data value y ij Is the format correct:

[0028] The step S102 of dividing each dimension into sub-dimensions and calculating the quality score of each dimension includes:

[0029] The consistency dimension is divided into two sub-dimensions, including: sequential consistency C con1 and logical consistency C con2 ; The sequential consistency C con1 for:

[0030]

[0031] The logical consistency C con2 for;

[0032]

[0033] Where D 31 T con The number of elements in T con is the largest subset of the time attribute T that meets the timing requirements, and the time attribute T = [t 1 t 2 …t m ] T ;D 32 A data pair that does not conform to the logical relationship; D 3 The total number of attributes of the logical relationship.

[0034] The step S102 of dividing each dimension into sub-dimensions and calculating the quality score of each dimension includes:

[0035] The timeliness C T for:

[0036]

[0037] Where D 4 is the total number of data that meets the timeliness requirement; in the largest subset T that meets the timing requirementcon Take any two adjacent elements in and denote them as t i-1 and t i , if 0≤t i -t i-1 ≤τ max , then it is regarded as element t i Meet timeliness; τ max The maximum time difference between adjacent records.

[0038] The step S102 of dividing each dimension into sub-dimensions and calculating the quality score of each dimension includes:

[0039] The uniqueness dimension is divided into two sub-dimensions, including: attribute uniqueness C U1 and record uniqueness C U2 ;

[0040] The attribute uniqueness C U1 for:

[0041]

[0042] The uniqueness of the record C U2 for:

[0043]

[0044] Where D 51 Represents attribute a j Whether the value is repeated: D 52 Record the number of redundant entries in the data set.

[0045] The step S102 of dividing each dimension into sub-dimensions and calculating the quality score of each dimension includes:

[0046] The accuracy dimension is divided into two sub-dimensions, including: range accuracy C A1 and numerical accuracy C A2 ;

[0047] The range accuracy C A1 for:

[0048]

[0049] The numerical accuracy C A2 for:

[0050]

[0051] Where D 61 For attribute a j The number of elements that meet the range requirements, D 62Indicates the number of data that do not meet the numerical accuracy requirements for each attribute.

[0052] The allocation of sub-dimension weights in step S103 includes:

[0053] Step S1031: Determine the sub-dimension weight distribution formula S of integrity com for:

[0054] S com =S 11 ×(W 12 ×S 12 +W 13 ×S 13 )

[0055] In the formula, S 11 is the attribute integrity, S 12 and W 12 To record the completeness score and its weight, S 13 and W 13 is the numerical completeness score and its weight, W 12 +W 13 =1 and W 12 >W 13 , allocate W according to actual application requirements 12 , W 13 Specific value of

[0056] Step S1032: Determine the sub-dimension weight distribution formula S for standardization, consistency, uniqueness, and accuracy else for:

[0057] S else =W 1 ×S 1 +W 2 ×S 2

[0058] Where W 1 is the first subdimension, W 2 is the second sub-dimension, the first sub-dimension W 1 and the second subdimension W 2 They correspond to the two sub-dimensions of standardization, consistency, uniqueness, and accuracy, and satisfy W 1 +W 2 =1;

[0059] For normativeness, the first sub-dimension W 1 The second sub-dimension W is type normativeness. 2 For accuracy specification, satisfy W 1 <W 2 ;

[0060] For consistency, the first sub-dimension W1 For sequential consistency, the second sub-dimension W 2 For logical consistency, satisfy W 1 >W 2 ; If there is no logical relationship between the attributes, then W 1 =1,W 2 =0;

[0061] For uniqueness and accuracy, W 1 =0.5,W 2 =0.5;

[0062] Step S1033: Determine timeliness without sub-dimension weight allocation.

[0063] The allocation of combined weights of each dimension in step S103 includes:

[0064] The subjective weights were determined using fuzzy analytic hierarchy process;

[0065] The objective weights were determined using the CRITIC-entropy weight method;

[0066] Use the inverse tangent function least squares method to combine weights;

[0067] Determining the subjective weight using the fuzzy analytic hierarchy process includes:

[0068] According to the specific situation of the problem, the importance of each factor in the same criterion layer is compared two by two, and the 0.1-0.9 scaling method is used to establish the fuzzy judgment matrix B between the two levels. ij ) m1×m1 (i,j=1,2,…,m1), m1 is the number of factors in the lower layer corresponding to the judgment matrix, that is, the number of evaluation indicators;

[0069] The element b in the fuzzy judgment matrix B ij Take values ​​in the range of 0 to 1 and calculate the fuzzy consistency matrix F;

[0070] F=(f ij ) m×m (i,j=1,2,…,m):

[0071]

[0072] The elements in the fuzzy consistent matrix F satisfy f ij =f ik -f jk +0.5, the weight calculation formula of fuzzy analytic hierarchy process is as follows:

[0073]

[0074] Where β is the difference factor, and the difference in weights is most reasonable when taking the value in the formula; w mi is the weight corresponding to the i-th indicator, then the weight vector w obtained by the fuzzy hierarchical analysis method is m for:

[0075] w m =[w m1 ,w m2 ,…,w mm ]

[0076] Determining the objective weight using the CRITIC-entropy weight method includes:

[0077] First, the distribution network multivariate time series data set is dimensionless processed:

[0078]

[0079] In the formula, a ij is the value of the jth evaluation index of the i-th object to be evaluated; a i ' j is the normalized value;

[0080] Calculate the probability matrix P = (p ij ) m×n , that is, the proportion of each quality indicator:

[0081]

[0082] Calculate the entropy value e of each indicator j :

[0083]

[0084] in When p ij =0 or p ij =1, p ij ln(p ij )=0.

[0085] Calculate the standard deviation S j :

[0086]

[0087] Calculate the correlation coefficient R j :

[0088]

[0089] In the formula, a i ' j For the object i to be evaluated, the jth *The normalized values ​​of the j-th and j-th evaluation indicators; For the jth * The average value of the index of item and item j; For the jth * The correlation coefficient between the item and the j-th indicator; find the objective weight value w of the j-th evaluation indicator j :

[0090]

[0091] The method of combining weights using the inverse tangent function least squares method includes:

[0092] Assume that there are m data samples, each sample has n evaluation indicators, and its data matrix is ​​X = (x ij ) m×n ; For all indicators of the sample to be evaluated, the difference between the final score of each indicator under the optimal weight and the score under the subjective and objective weight scheme should be as small as possible; the square of the deviation between the subjective and objective weight and the combined weight is calculated by comparing the sample data x ij The product and minimum of is the optimization goal. First, a combined weighting model is established. Assuming that the weight vectors determined by the fuzzy analytic hierarchy process and the CRITIC-entropy weight method are α and β respectively, the combined weight vector to be obtained is w, and a combined weight model is established:

[0093]

[0094] The optimal solution is obtained by using the Lagrange multiplier method, and the function L(ω,λ) is constructed, where λ is the Lagrange multiplier:

[0095]

[0096] From the extreme value existence conditions, we can know that:

[0097]

[0098] Solve the above equation and transpose it to get the combined weight of each dimension w = [w 1 ,w 2 ,…,w n ].

[0099] The calculation of the total quality score D of the data to be tested in step S104 includes:

[0100] D=wC i

[0101] In the formula, w is the combined weight of each dimension, C i is the quality score of each dimension;

[0102] The evaluation data quality levels include:

[0103] According to the total score D of the data quality to be tested, the data quality is divided into five levels, including: "fail",;

[0104] If the total quality score of the data to be tested is D∈(0,a], the data quality level of the data set to be tested is evaluated as "failed", "passed", "medium", "good" or "excellent";

[0105] If the total quality score of the data to be tested is D∈(a,b], the data quality level of the data set to be tested is evaluated as "passing";

[0106] If the total quality score of the data to be tested is D∈(b,c], the data quality level of the data set to be tested is assessed as “medium”;

[0107] If the total quality score of the data to be tested is D∈(c,d], the data quality level of the data set to be tested is evaluated as “good”;

[0108] If the total quality score of the data to be tested is D∈(d,e], the data quality level of the data set to be tested is evaluated as “excellent”;

[0109] Where a, b, c, d, and e are preset grade division intervals.

[0110] The beneficial effects of the present invention are:

[0111] The present invention proposes a quality assessment method for multivariate time series data of distribution network, which fully considers multiple key factors such as data integrity, redundancy and timeliness, and conducts in-depth analysis around six core dimensions, namely integrity, standardization, consistency, timeliness, uniqueness and accuracy. Several sub-dimensions are further subdivided under each dimension, and the correlation between each evaluation dimension is fully considered to form a hierarchical evaluation method. In the weight distribution scheme, a subjective and objective combined weighting method is adopted to calculate the weight of each evaluation dimension, which integrates the subjective judgment of experts and objective data analysis, avoids the one-sidedness and limitations of traditional evaluation methods, overcomes the data quality evaluation framework and model used in the prior art, which only makes qualitative analysis of data quality problems, cannot fully consider the correlation between evaluation dimensions, lacks systematic quantitative analysis capabilities, and cannot provide effective evaluation and improvement solutions when facing complex multivariate time series data. Applying a quality assessment method for multivariate time series data of distribution network disclosed by the present invention, a reasonable data quality evaluation system is established, so that the results can more truly reflect the data quality, thereby optimizing the enterprise decision-making process, promoting the intelligent development of distribution network, and realizing more efficient and sustainable power management. BRIEF DESCRIPTION OF THE DRAWINGS

[0112] Figure 1It is a flow chart of a quality assessment method for multivariate time series data of a distribution network according to the present invention;

[0113] Figure 2 This is a diagram showing the dimensional division of evaluation indicators in an embodiment of the present invention;

[0114] Figure 3 A weight distribution comparison diagram of different weighting methods according to an embodiment of the present invention; DETAILED DESCRIPTION

[0115] The present invention provides a quality assessment method for multivariate time series data of a distribution network, which is further described in detail below in conjunction with the accompanying drawings.

[0116] like Figure 1 The embodiment of the present invention shown discloses a quality assessment method for multivariate time series data of a distribution network, comprising:

[0117] Step S101: Based on the data characteristics and distribution features of the distribution network, select the quality evaluation dimension of the multivariate time series data of the distribution network;

[0118] Step S102: designing an evaluation method for the distribution network multivariate time series data quality evaluation dimension, including: dividing each dimension into sub-dimensions, and calculating the quality score of each dimension;

[0119] Step S103: designing a weight distribution scheme for each dimension and determining the weight of each dimension; the designing a weight distribution scheme for each dimension includes: distributing sub-dimension weights and distributing combined weights of each dimension;

[0120] Step S104: Calculate the total quality score of the data to be tested according to the quality score of each dimension and the weight of each dimension, and evaluate the data quality level;

[0121] The distribution network multivariate time series data quality evaluation dimensions include: completeness, standardization, consistency, timeliness, uniqueness and accuracy.

[0122] The specific implementation process of each step is described in detail below.

[0123] Step S101: Based on the data characteristics and distribution features of the distribution network, select the quality evaluation dimension of the multivariate time series data of the distribution network;

[0124] In this embodiment, data quality problem analysis and evaluation dimension selection are performed:

[0125] With the continuous improvement of the intelligence level of distribution network construction, power companies have generated complex and diverse data, the amount of data has increased exponentially, and data quality issues exist in multiple links such as data collection, transmission, and storage.

[0126] Among them, during the data collection process, if the collection system fails, the collected data will be empty, resulting in data missing and data anomalies; during the data transmission process, if the communication equipment fails, the collected data will not be transmitted in a timely manner or even cannot be transmitted, resulting in data missing; during the power data storage process, if the data storage equipment fails or human errors occur, it will lead to data loss, data redundancy, etc.

[0127] Based on the structural characteristics of the multivariate time series data of the distribution network, in view of the current situation of huge data volume and complex structure of the distribution network, as well as the widespread problems of data missing, redundancy and outliers, this paper takes demand as the guide and comprehensively considers multiple key factors such as data completeness, redundancy and update timeliness, and selects six quality evaluation dimensions, namely: completeness, standardization, consistency, timeliness, uniqueness and accuracy.

[0128] The multivariate time series data structure of the distribution network is shown in Table 1. In the table, Y is the data matrix to be evaluated, where r i =[y i1 y i2 …y in ] represents the i-th record, each element is an attribute value of the record, a j =[y 1j y 2j …y mj ] T Represents the jth attribute, where each element is a record value of the attribute, a 1 =[y 11 y 21 …y m1 ] T It is the time attribute.

[0129] Table 1 Summary of the multivariate time series data structure of distribution network

[0130] Y <![CDATA[a 1 ]]> <![CDATA[a 2 ]]> … <![CDATA[a n ]]> r1 y11 y12 … <![CDATA[y1 n ]]> r2 y21 y22 … <![CDATA[y2 n ]]> … … … … … <![CDATA[r m ]]> <![CDATA[y m1 ]]> <![CDATA[y m2 ]]> … <![CDATA[y mn ]]>

[0131] Step S102: designing an evaluation method for the distribution network multivariate time series data quality evaluation dimension, including: dividing each dimension into sub-dimensions, and calculating the quality score of each dimension;

[0132] In this embodiment, the evaluation sub-dimensions are divided and the calculation method is designed:

[0133] Each data quality dimension to be evaluated is divided into multiple sub-dimensions that reflect different data quality information. Through different degrees of subdivision, the quality issues of the data set to be evaluated can be more comprehensively presented in a hierarchical manner. The specific division method is as follows: Figure 2 shown.

[0134] (1) In modern data management, data integrity is one of the key factors to ensure data quality and reliability. This paper divides data integrity into three sub-dimensions: attribute integrity, record integrity, and value integrity, which helps to reflect the situation of missing data in the data set in detail and provide a scientific basis for data governance and decision-making.

[0135] ① Attribute completeness mainly measures the degree of missingness of each attribute in the data set. If all values ​​of an attribute are empty, the attribute is considered missing. The evaluation of attribute completeness is not only related to the availability of data, but also directly affects the accuracy of subsequent analysis. 11 is the number of non-missing attributes in a given data set, n is the number of attributes in the complete data set, and the calculation formula for attribute completeness is:

[0136]

[0137] ②Record integrity reflects the degree of missing of the entire record in the data set. Due to communication errors, system failures or data transmission problems, a certain record may be missing. The total number of records to be collected can be obtained based on the start and end time of data collection and the sampling frequency. The actual number of records in the data set is D 12 . The calculation formula for record completeness is:

[0138]

[0139] ③ Numerical integrity evaluates the degree of missingness of each element in the data set, which is expressed as the ratio of the existing elements in the data set to the total number of elements to be collected. The calculation formula for numerical integrity is:

[0140]

[0141] Among them, Q(y ij ) indicates whether the data yij in the test data set is missing:

[0142]

[0143] (2) In the field of data management and analysis, data standardization is the basis for ensuring the validity and availability of information, and indicates the degree to which the data content expression complies with relevant norms and standards. This invention divides it into type standardization and accuracy standardization.

[0144] ① Type standardization reflects the correctness of the data type and ensures that the data meets the predetermined type standard during use. In the application scenario of multivariate time series data in the distribution network, the data type is generally numeric. If the collected data type cannot be converted into a numeric type, it will be regarded as type error data. The calculation formula for type standardization is:

[0145]

[0146] Among them, M(y ij ) is used to determine the data value y ij Is the format correct:

[0147]

[0148] ② Accuracy is used to evaluate whether the data value meets the specific decimal point requirements. In application scenarios such as scientific computing and financial reporting, the accuracy of the data is directly related to the reliability of the analysis results. 2 ,l 3 ,…,l n} is the number of digits of precision required for each attribute of the test data set, D 22 is the attribute a in the data set j Medium precision is the required precision l j The calculation formula for the number of elements and accuracy specification is:

[0149]

[0150] (3) Consistency refers to the coordination of the relationship between data and other data. The consistency of data includes their temporal coherence and the logic between data in a specific scenario. The consistency dimension is divided into two sub-dimensions, including: sequential consistency C con1 and logical consistency C con2 ;

[0151] ① Sequential consistency mainly reflects the time order between data records. In many cases, the data transmission process may be affected by factors such as delays or equipment failures, resulting in deviations in the order of records. In order to ensure sequential consistency, it is necessary to identify the maximum monotonically non-decreasing subsequence in time in the data record. For the time attribute T = [t 1 t 2 ...t m ] T , remember T con is the largest subset that meets the timing requirements, denoted by T con The number of elements in is D 31 The calculation formula for sequential consistency is:

[0152]

[0153] ② Logical consistency focuses on the logical relationship between different attributes. In some application scenarios, there should be clear logical connections between attributes, but conflicts between attribute values ​​may occur due to sensor failure or data collection errors. The total number of attributes with logical relationships is D. 3 , D 32A data pair that does not conform to the logical relationship. Considering the correlation between different sub-dimensions, if there is an integrity error in the data pair or the type specification is not met, the data pair is judged to violate the logical relationship. The calculation formula for logical consistency is:

[0154]

[0155] (4) Timeliness is a key indicator for measuring data quality. Its definition is to evaluate whether the time when data arrives meets the predetermined requirements. Factors that affect data timeliness mainly include the amount of data that exceeds the time limit. The more overdue data means the worse the timeliness of the overall data set. In addition, the order in which the data is arranged will also change the evaluation of the time difference. Therefore, the evaluation of timeliness requires that the data meet the sequential consistency.

[0156] The largest subset T that meets the timing requirements con Take any two adjacent elements in and denote them as t i-1 and t i , if 0≤t i -t i-1 ≤τ max , then it is regarded as element t i Meet the timeliness, remember D 4 is the total number of data that meets timeliness requirements. max The maximum time difference between adjacent records. The calculation formula for timeliness is:

[0157]

[0158] (5) Uniqueness is used to measure the degree of duplication of data in a data set. The uniqueness dimension is divided into two sub-dimensions, including: attribute uniqueness C U1 and record uniqueness C U2 ;

[0159] ① Attribute uniqueness reflects whether there is repeated sampling for a specific attribute. Due to human error or improper data input, there are often multiple identical attribute values ​​in the data set. This redundancy not only makes the data confusing, but also affects the accuracy of data analysis. The calculation formula for attribute uniqueness is:

[0160]

[0161] Among them, D 51 Represents attribute a j Whether the value is repeated:

[0162]

[0163] ② Record uniqueness reflects whether a record in the data set is redundant. For the complex environment of distribution network operation scenarios, each data record must be able to uniquely identify its source and content. 52 The number of redundant records in the data set. The calculation formula for record uniqueness is:

[0164]

[0165] (6) Accuracy is generally considered to be the most important indicator in data quality assessment. It measures the extent to which the data measurement results can correctly represent the real distribution network data. The accuracy dimension is divided into two sub-dimensions, including: range accuracy C A1 and numerical accuracy C A2 Considering the correlation between different evaluation dimensions, for multivariate time series data, the data must meet sequential consistency and timeliness before evaluating accuracy. If the data set to be evaluated does not meet sequential consistency, it needs to be rearranged in ascending time order. Data that does not meet completeness and type norms will also be considered as not meeting accuracy.

[0166] ① Range accuracy focuses on whether the data value is within the predetermined required range. In the distribution network application scenario, parameters such as current, voltage, and power have their specific value ranges. Any data outside this range is considered inaccurate. In order to define the effective range, an acceptable data interval is usually set, and the range accuracy of the data is evaluated by calculating the number of elements that meet the range requirements. Definition and Respectively represent the attribute a j The minimum and maximum values ​​allowed are denoted by D 61 For attribute a j The number of elements that meet the range requirements, the calculation formula for range accuracy is:

[0167]

[0168] ② Numerical accuracy focuses on measuring the difference between data and the true value. The present invention uses Z-score standardization to describe numerical accuracy. Z-score standardization is a commonly used method in data preprocessing. It is used to convert data with different characteristics into the same standard normal distribution. By calculating the difference between each data point and the mean and dividing it by the standard deviation, the original data is converted into a new data set with zero mean and unit variance. Z-score is a numerical calculation that describes the distance between the value x of a set of data and the mean value μ of the set of data. The Z-score of a value x in a set of data is defined as follows:

[0169]

[0170] Among them, x is the original data point, μ is the mean of the original data set, and σ is the standard deviation of the original data set. Through Z-score standardization, outliers can be identified. Usually, if the Z-score of a data point is greater than 3 or less than -3, it can be considered an outlier.

[0171] For attribute a j , remember D 62 Indicates the number of data that do not meet the numerical accuracy requirements for each attribute. The calculation formula for numerical accuracy is:

[0172]

[0173] Step S103: designing a weight distribution scheme for each dimension and determining the weight of each dimension; the designing a weight distribution scheme for each dimension includes: distributing sub-dimension weights and distributing combined weights of each dimension;

[0174] In this embodiment, the evaluation dimension weights are allocated:

[0175] (1) Sub-dimension weight distribution

[0176] In step S102, the present invention selects six data quality evaluation dimensions and divides them into sub-dimensions. In the weight distribution scheme, considering that the different importance of different sub-dimensions under the same evaluation dimension will affect the evaluation results of data quality, the present invention first determines the weight distribution scheme of different sub-dimensions.

[0177] First, weights are assigned to the sub-dimensions of integrity. The present invention divides integrity into three dimensions, namely attribute integrity, record integrity, and numerical integrity. In the application scenario of the distribution network, the present invention regards the attribute integrity of the data set as the most important sub-dimension. The assessment of attribute integrity is not only related to the availability of the data, but also directly affects the accuracy of subsequent analysis, and can often reflect more value. With regard to record integrity and numerical integrity, the present invention believes that for the multivariate time series data set of the distribution network, there are usually multiple attribute value data in one record. Therefore, it is believed that the importance of record integrity should be higher than that of numerical integrity. In summary, the weight assignment formula for the sub-dimensions of integrity is given;

[0178] Step S1031: Determine the sub-dimension weight distribution formula S of integrity com for:

[0179] S com =S 11 ×(W 12 ×S 12 +W 13 ×S 13 )

[0180] Among them, S 11 is the attribute integrity, S12 and W 12 To record the completeness score and its weight, S 13 and W 13 is the numerical completeness score and its weight, W 12 +W 13 =1 and W 12 >W 13 According to the importance of the two sub-dimensions, W is allocated according to the actual application needs. 12 , W 13 The specific value of .

[0181] Standardization, consistency, uniqueness, and accuracy all contain two sub-dimensions, so the same calculation method is used to determine the quality score.

[0182] Step S1032: Determine the sub-dimension weight distribution formula S for standardization, consistency, uniqueness, and accuracy else for:

[0183] S else =W 1 ×S 1 +W 2 ×S 2

[0184] Where W 1 is the first subdimension, W 2 is the second sub-dimension, the first sub-dimension W 1 and the second subdimension W 2 They correspond to the two sub-dimensions of standardization, consistency, uniqueness, and accuracy, and satisfy W 1 +W 2 =1;

[0185] For normativeness, the first sub-dimension W 1 The second sub-dimension W is type normativeness. 2 For accuracy specification, satisfy W 1 <W 2 ;

[0186] For consistency, the first sub-dimension W 1 For sequential consistency, the second sub-dimension W 2 For logical consistency, satisfy W 1 >W 2 ; If there is no logical relationship between the attributes, then W 1 =1,W 2 =0;

[0187] For uniqueness and accuracy, W 1 =0.5,W 2 =0.5;

[0188] Selse =W 1 ×S 1 +W 2 ×S 2

[0189] In this embodiment, W 1 +W 2 =1. For standardization, incorrect types will lead to confusion in data parsing and analysis. In the power business scenario, the data type is usually numeric, and type errors generally do not occur except for human factors. However, there are often strict requirements for the accuracy of different attribute data. Therefore, the weight distribution of the two sub-dimensions of standardization should meet W 1 <W 2 ; For consistency, the distribution network multivariate time series data set must contain time attributes and have strict requirements on the time order, while logical consistency can usually be detected by accuracy or normative evaluation. Therefore, for the weight allocation of the two sub-dimensions of consistency, W 1 >W 2 , In addition, if there is no logical relationship between the attributes, then W 1 =1,W 2 =0; For uniqueness and accuracy, the present invention considers that each of their sub-dimensions is equally important;

[0190] Step S1033: Determine timeliness without sub-dimension weight allocation.

[0191] For timeliness, since it is not divided into sub-dimensions, there is no need to assign weights.

[0192] In this embodiment, only a general method for weight allocation schemes for different sub-dimensions is provided, and the weight allocation scheme for each sub-dimension can be customized according to the needs of different actual application scenarios.

[0193] (2) Weight allocation of each dimension combination:

[0194] The allocation of combined weights of each dimension in step S103 includes:

[0195] ·Use fuzzy analytic hierarchy process to determine subjective weights;

[0196] ·Use CRITIC-entropy weight method to determine objective weights;

[0197] ·Use the inverse tangent function least squares method to combine weights;

[0198] In this embodiment, the evaluation dimension weights directly affect the distribution network data quality evaluation results and are the key link in the entire evaluation process. The present invention adopts fuzzy hierarchical analysis method and CRITIC-entropy weight method to respectively determine the subjective and objective weights of each dimension, and adopts the inverse tangent function least squares method to comprehensively determine the combined weight of each evaluation dimension.

[0199] ① Determine subjective weights using fuzzy analytic hierarchy process

[0200] Fuzzy analytic hierarchy process (FAHP) introduces fuzzy theory on the basis of analytic hierarchy process to improve the shortcomings of analytic hierarchy process, such as strong subjectivity and weak scientificity, and to solve the problem that the consistency of judgment matrix is ​​difficult to satisfy.

[0201] Determining the subjective weight using the fuzzy analytic hierarchy process includes:

[0202] The fuzzy hierarchical analysis method hierarchizes the complex evaluation problems. According to the specific situation of the problem, the importance of each factor in the same criterion layer is compared two by two, and the scale method of 0.1-0.9 is adopted. The specific comparison principle and scale method are shown in Table 2. Based on this, the fuzzy judgment matrix B between the two levels is established. ij ) m1×m1 (i,j=1,2,…,m1), m1 is the number of factors in the lower layer corresponding to the judgment matrix, that is, the number of evaluation indicators.

[0203] Table 2 Summary of comparison principles and scaling methods

[0204]

[0205] The element b in the fuzzy judgment matrix B ij The fuzzy consistency matrix F is calculated according to the following formula: ij ) m×m (i,j=1,2,…,m):

[0206]

[0207] The elements in the fuzzy consistent matrix F satisfy f ij =f ik -f jk +0.5, the weight calculation formula of fuzzy analytic hierarchy process is as follows:

[0208]

[0209] Where β is the difference factor, and the difference in weights is most reasonable when taking the value in the formula; w mi is the weight corresponding to the i-th indicator, then the weight vector w obtained by the fuzzy hierarchical analysis method is m for:

[0210] w m =[w m1 ,w m2 ,…,w mm ]

[0211] ②CRITIC-Entropy Weight Method to Determine Objective Weights

[0212] The entropy weight method and the CRITIC method are common objective weighting methods. The entropy weight method mainly uses entropy to calculate the weight of indicators, which can well reflect the degree of dispersion between indicators, but does not take into account the correlation and contrast between indicators. On the contrary, the CRITIC method calculates the weight of indicators through the amount of information, which can better reflect the correlation and contrast between indicators. Therefore, the entropy weight method is combined with the CRITIC method to comprehensively consider the degree of dispersion, correlation and contrast between indicators, making the objective weighting more reasonable.

[0213] Determining the objective weight using the CRITIC-entropy weight method includes:

[0214] First, the distribution network multivariate time series data set is dimensionless processed:

[0215]

[0216] Among them, a ij is the value of the jth evaluation index of the i-th object to be evaluated; a i ' j is the normalized value.

[0217] Calculate the probability matrix P = (p ij ) m×n , that is, the proportion of each quality indicator:

[0218]

[0219] Calculate the entropy value e of each indicator j :

[0220]

[0221] in When p ij =0 or p ij =1, p ij ln(p ij )=0.

[0222] Calculate the standard deviation S j :

[0223]

[0224] Calculate the correlation coefficient R j :

[0225]

[0226] in, a i ' j For the object i to be evaluated, the jth * The normalized values ​​of the j-th and j-th evaluation indicators; For the jth * The average value of the index of item and item j; For the jth * The correlation coefficient between the item and the j-th indicator.

[0227] Obtain the objective weight value w of the jth evaluation index j :

[0228]

[0229] ③ Combination weighting by the least squares method of inverse tangent function

[0230] The present invention adopts the inverse tangent function least squares combined weighting model combined with fuzzy analytic hierarchy process and CRITIC-entropy weight method to calculate the subjective and objective weights. The inverse tangent function has unique mathematical characteristics. It takes into account the impact of data oscillation on the development trend of the data sequence. In order to weaken the impact of extreme index evaluation data, the inverse tangent function is introduced on the basis of the least squares method. It has a better combination effect for the case of large weight fluctuations. The combined weight calculation process is as follows:

[0231] The method of combining weights using the inverse tangent function least squares method includes:

[0232] Assume that there are m data samples, each sample has n evaluation indicators, and its data matrix is ​​X = (x ij ) m×n For all indicators of the sample to be evaluated, the difference between the final score of each indicator under the optimal weight and the score under the subjective and objective weight scheme should be as small as possible. ij The product and minimum of is the optimization goal. First, a combined weighting model is established. Assuming that the weight vectors determined by the fuzzy analytic hierarchy process and the CRITIC-entropy weight method are α and β respectively, the combined weight vector to be obtained is w, and a combined weight model is established:

[0233]

[0234] The optimal solution is obtained by using the Lagrange multiplier method, and the function L(ω,λ) is constructed, where λ is the Lagrange multiplier:

[0235]

[0236] From the extreme value existence conditions, we can know that:

[0237]

[0238] Solving the above formula, we can get the combined weight of each dimension w = [w 1 ,w 2 ,…,w n ].

[0239] Step S104: Calculate the total quality score of the data to be tested according to the quality score of each dimension and the weight of each dimension, and evaluate the data quality level;

[0240] In this embodiment, the total score of data quality is calculated and the level is evaluated:

[0241] According to the evaluation dimension score calculation method and corresponding weights obtained in S102 and S103, the total data quality score of the test data set is calculated:

[0242] The calculation of the total quality score D of the data to be tested in step S104 includes:

[0243] D=wC i

[0244] In the formula, w is the combined weight of each dimension, C i is the quality score of each dimension;

[0245] The evaluation data quality levels include:

[0246] According to the total score D of the data quality to be tested, the data quality is divided into five levels, including: "fail",;

[0247] If the total quality score of the data to be tested is D∈(0,a], the data quality level of the data set to be tested is evaluated as "failed", "passed", "medium", "good" or "excellent";

[0248] If the total quality score of the data to be tested is D∈(a,b], the data quality level of the data set to be tested is evaluated as "passing";

[0249] If the total quality score of the data to be tested is D∈(b,c], the data quality level of the data set to be tested is assessed as “medium”;

[0250] If the total quality score of the data to be tested is D∈(c,d], the data quality level of the data set to be tested is evaluated as “good”;

[0251] If the total quality score of the data to be tested is D∈(d,e], the data quality level of the data set to be tested is evaluated as “excellent”;

[0252] Where a, b, c, d, and e are preset grade division intervals.

[0253] The present invention refers to the general score of the multivariate time series data quality of the distribution network and divides the data quality into five levels. If the quality score D∈(0,a], the data quality level of the data set to be tested is evaluated as "failed"; if D∈(a,b], the data quality level of the data set to be tested is evaluated as "passed"; if D∈(b,c], the data quality level of the data set to be tested is evaluated as "medium"; if D∈(c,d], the data quality level of the data set to be tested is evaluated as "good"; if D∈(d,e], the data quality level of the data set to be tested is evaluated as "excellent". The present invention only proposes a grading method for the application scenario of the distribution network. The specific interval can be set according to business needs, such as setting a=60, b=70, c=80, d=90, e=100. At this point, the data quality evaluation of the data set to be tested is completed.

[0254] The distribution transformer data of a regional distribution network with a voltage level of 10kV and a rated capacity of 200kVA is selected as the research object. According to the relevant national standards, the voltage range U∈[198,235.4], the current I∈[0,288.68], and the power factor A ctor ∈[0,1], active power P∈[0,200], reactive power Q∈[0,120]. The original data set meets the requirements and does not have quality problems such as data vacancies, redundancy, and duplication. In order to verify the practicality of the method proposed in this invention, some data are modified on the basis of the original data set to reduce its score. The distribution transformer data set to be evaluated in a certain area is shown in Table 3.

[0255] Table 3 Summary of the distribution transformer evaluation data set in a certain region

[0256]

[0257] Based on this, the scores of each indicator are calculated as follows:

[0258] There are 90 data in the table, including 9 attributes and 10 time records. According to the method proposed in step S102, after integrity detection, it can be known that attribute I a All data of D are missing. 11 = 8, the attribute integrity score is 88.89, and the record at time 2:45 is U a Data missing, numerical completeness is 87.78; change the A with time stamp 3:16 ctorThe data type is changed to text type. After the normative test, it is found that it does not meet the type normativeness. After the consistency test, it is found that the record at the time stamp of 2:45 does not meet the monotonicity of the time sequence, and the order consistency score is 90. The time difference threshold is set to 2 minutes, and the area data is updated every 15 minutes. After the timeliness test, it is found that the transmission of the record at the time stamp of 1:33 has timed out. After the uniqueness test, it is found that the data of the data set at the time stamp of 2:30 has record redundancy. 52 =1; According to the relevant national standard range, the I c The value does not meet the range accuracy. According to the calculation method of numerical accuracy, the I at the time scale of 2:45 b Value, time mark 3:00 I b The value and Q value do not meet the numerical accuracy, and the present invention regards null values ​​and data with incorrect types as inaccurate data, then the numerical accuracy score is 82.22.

[0259] In the weight distribution scheme, the present invention takes into account that the different importance of different sub-dimensions under the same evaluation dimension will affect the evaluation results of data quality. According to the discussion in step S103, the weight distribution schemes of different sub-dimensions are determined. The specific scores of each sub-dimension and their weight distribution results are shown in Table 4.

[0260] Table 4 Summary of scores of each dimension and weight distribution of sub-dimensions

[0261]

[0262] For the weight distribution scheme of the six evaluation dimensions, firstly, the evaluation factors are ranked according to the fuzzy analytic hierarchy process and the opinions of experts: accuracy > completeness > standardization > consistency > timeliness > uniqueness, forming a fuzzy judgment matrix as follows:

[0263]

[0264] According to the fuzzy hierarchical analysis method, the subjective weight of each evaluation dimension can be obtained:

[0265] w m =[0.2026,0.1667,0.1667,0.1308,0.0948,0.2385]

[0266] When obtaining the objective weight, the present invention adopts a method combining CRITIC and entropy weight method, and obtains the entropy value e of each evaluation dimension as shown in step S103. j , standard deviation S j And the correlation coefficient R j , in order:

[0267] e j=[0.7686,0.8164,0.8244,0.4662,0.8598,0.6679]

[0268] S j =[0.4023,0.3731,0.4012,0.4194,0.4149,0.3996]

[0269] R j =[6.4384,6.2521,5.7276,5.9305,5.6059,5.4719]

[0270] Finally, the objective weights of each dimension are obtained as follows:

[0271] w j =[0.1874,0.1848,0.1745,0.1305,0.1776,0.1452]

[0272] The present invention adopts the inverse tangent least squares method to determine the final combination weights, and introduces the inverse tangent function on the basis of the least squares method, which has a better combination effect for the case where the weights fluctuate greatly. The gap between the final score of each indicator under the optimal weight and the score under the subjective and objective weight scheme should be as small as possible, and this is used as the optimization goal. In order to accelerate convergence during the optimization process, the SLSQP (Sequential Least Squares Quadratic Programming) algorithm is adopted. SLSQP is a sequential quadratic programming method, which is particularly suitable for nonlinear constrained optimization problems. According to the steps described in S103, the optimal weights can be obtained:

[0273] w=[0.1931,0.1797,0.1726,0.1326,0.1339,0.1881]

[0274] like Figure 3 As shown, in order to verify the rationality of the method proposed in the present invention, the weight distribution obtained by organically combining subjective and objective weights using the inverse tangent least squares method proposed in the present invention is compared with the single fuzzy hierarchical analysis method, the single CRITIC-entropy weight method, the single CRITIC method, the single entropy weight method and the weight distribution obtained by organically combining subjective and objective weights using the inverse tangent least squares method proposed in the present invention. It can be intuitively seen from the figure that the combined weighting method proposed in the present invention is both reasonable and can effectively reduce the impact of extreme indicators on the overall data quality assessment, and has high objectivity and practicality.

[0275] According to the above-mentioned scoring calculation method and weight distribution scheme for each evaluation dimension, the final score of the test data set under the optimal weight distribution is calculated to be D=88.61. If the grade interval set in step S104 is followed, the quality grade of this data set is "good".

Claims

1. A quality assessment method for multivariate time series data of distribution network, characterized in that: include: Step S101: Based on the data characteristics and distribution features of the distribution network, select the quality evaluation dimension of the multivariate time series data of the distribution network; Step S102: designing an evaluation method for the distribution network multivariate time series data quality evaluation dimension, including: dividing each dimension into sub-dimensions, and calculating the quality score of each dimension; Step S103: designing a weight distribution scheme for each dimension and determining the weight of each dimension; the designing a weight distribution scheme for each dimension includes: distributing sub-dimension weights and distributing combined weights of each dimension; Step S104: Calculate the total quality score of the data to be tested according to the quality score of each dimension and the weight of each dimension, and evaluate the data quality level; The distribution network multivariate time series data quality evaluation dimensions include: completeness, standardization, consistency, timeliness, uniqueness and accuracy.

2. A quality assessment method for multivariate time series data of a distribution network according to claim 1, characterized in that: The step S102 of dividing each dimension into sub-dimensions and calculating the quality score of each dimension includes: The integrity dimension is divided into three sub-dimensions, including: attribute integrity C com1 , Record Integrity C com2 and numerical integrity C com3 ; The attribute integrity C com1 for: The record integrity C com2 for: The numerical integrity of C com3 for: Where D 11 is the number of non-missing attributes in the given data set, and n is the number of attributes in the complete data set; D 12 is the number of records in the actual data set, m is the total number of records to be collected based on the start and end time of data collection and the sampling frequency; Q(y ij ) represents the data y in the test data set ij Is it lost:

3. The quality assessment method for multivariate time series data of a distribution network according to claim 1 is characterized in that: The step S102 of dividing each dimension into sub-dimensions and calculating the quality score of each dimension includes: The normative dimension is divided into two sub-dimensions, including: Type normativeness C N1 and accuracy specification C N2 ; The type specification C N1 for: The accuracy specification C N2 for: In the formula, M(y ij ) is used to determine the data value y ij Is the format correct:

4. The quality assessment method for multivariate time series data of a distribution network according to claim 1 is characterized in that: The step S102 of dividing each dimension into sub-dimensions and calculating the quality score of each dimension includes: The consistency dimension is divided into two sub-dimensions, including: sequential consistency C con1 and logical consistency C con2 ; The sequential consistency C con1 for: The logical consistency C con2 for; Where D 31 T con The number of elements in T con is the largest subset of the time attribute T that meets the timing requirements, and the time attribute T = [t1 t2...t m ] T ;D 32 is a data pair that does not conform to the logical relationship; D3 is the total number of attributes in the logical relationship.

5. The quality assessment method for multivariate time series data of a distribution network according to claim 1 is characterized in that: The step S102 of dividing each dimension into sub-dimensions and calculating the quality score of each dimension includes: The timeliness C T for: Where D4 is the total number of data that meets the timeliness requirement; con Take any two adjacent elements in and denote them as t i-1 and t i , if 0≤t i -t i-1 ≤τ max , then it is regarded as element t i Meet timeliness; τ max The maximum time difference between adjacent records.

6. A quality assessment method for multivariate time series data of a distribution network according to claim 1, characterized in that: The step S102 of dividing each dimension into sub-dimensions and calculating the quality score of each dimension includes: The uniqueness dimension is divided into two sub-dimensions, including: attribute uniqueness C U1 and record uniqueness C U2 ; The attribute uniqueness C U1 for: The uniqueness of the record C U2 for: Where D 51 Represents attribute a j Whether the value is repeated: D 52 Record the number of redundant entries in the data set.

7. The quality assessment method for multivariate time series data of a distribution network according to claim 1 is characterized in that: The step S102 of dividing each dimension into sub-dimensions and calculating the quality score of each dimension includes: The accuracy dimension is divided into two sub-dimensions, including: range accuracy C A1 and numerical accuracy C A2 ; The range accuracy C A1 for: The numerical accuracy C A2 for: Where D 61 For attribute a j The number of elements that meet the range requirements, D 62 Indicates the number of data that do not meet the numerical accuracy requirements for each attribute.

8. The quality assessment method for multivariate time series data of a distribution network according to claim 1 is characterized in that: The allocation of sub-dimension weights in step S103 includes: Step S1031: Determine the sub-dimension weight distribution formula S of integrity com for: S com =S 11 ×(W 12 ×S 12 +W 13 ×S 13 ) In the formula, S 11 is the attribute integrity, S 12 and W 12 To record the completeness score and its weight, S 13 and W 13 is the numerical completeness score and its weight, W 12 +W 13 =1 and W 12 >W 13 , allocate W according to actual application requirements 12 , W 13 Specific value of Step S1032: Determine the sub-dimension weight distribution formula S for standardization, consistency, uniqueness, and accuracy else for: S else =W1×S1+W2×S2 In the formula, W1 is the first sub-dimension, W2 is the second sub-dimension, the first sub-dimension W1 and the second sub-dimension W2 correspond to the two sub-dimensions of standardization, consistency, uniqueness, and accuracy respectively, and satisfy W1+W2=1; For standardization, the first sub-dimension W1 is type standardization, and the second sub-dimension W2 is accuracy standardization, satisfying W1<W2; For consistency, the first sub-dimension W1 is sequential consistency, and the second sub-dimension W2 is logical consistency, satisfying W1>W2; if there is no logical relationship between the attributes, then W1=1, W2=0; For uniqueness and accuracy, W1 = 0.5, W2 = 0.5; Step S1033: Determine timeliness without sub-dimension weight allocation.

9. The quality assessment method for multivariate time series data of a distribution network according to claim 1, characterized in that: The allocation of combined weights of each dimension in step S103 includes: The subjective weights were determined using fuzzy analytic hierarchy process; The objective weights were determined using the CRITIC-entropy weight method; Use the inverse tangent function least squares method to combine weights; Determining the subjective weight using the fuzzy analytic hierarchy process includes: According to the specific situation of the problem, the importance of each factor in the same criterion layer is compared two by two, and the 0.1-0.9 scaling method is used to establish the fuzzy judgment matrix B between the two levels. ij ) m1×m1 (i,j=1,2,…,m1), m1 is the number of factors in the lower layer corresponding to the judgment matrix, that is, the number of evaluation indicators; The element b in the fuzzy judgment matrix B ij Take values ​​in the range of 0 to 1 and calculate the fuzzy consistency matrix F; F=(f ij ) m×m (i,j=1,2,…,m): The elements in the fuzzy consistent matrix F satisfy f ij =f ik -f jk +0.5, the weight calculation formula of fuzzy analytic hierarchy process is as follows: Where β is the difference factor, and the difference in weights is most reasonable when taking the value in the formula; w mi is the weight corresponding to the i-th indicator, then the weight vector w obtained by the fuzzy hierarchical analysis method is m for: In m =[in m1 ,In m2 ,…,In mm ] Determining the objective weight using the CRITIC-entropy weight method includes: First, the distribution network multivariate time series data set is dimensionless processed: In the formula, a ij a′ is the value of the jth evaluation index of the i-th object to be evaluated; ij is the normalized value; Calculate the probability matrix P = (p ij ) m×n , that is, the proportion of each quality indicator: Calculate the entropy value e of each indicator j : in When p ij =0 or p ij =1, p ij ln(p ij )=0; Calculate the standard deviation S j : Calculate the correlation coefficient R j : In the formula, a′ ij* , a′ ij For the object i to be evaluated, the jth * The normalized values ​​of the j-th and j-th evaluation indicators; For the jth * The average value of the index of item and item j; r j*j For the jth * The correlation coefficient between the item and the j-th indicator; Obtain the objective weight value w of the jth evaluation index j : The method of combining weights using the inverse tangent function least squares method includes: Assume that there are m data samples, each sample has n evaluation indicators, and its data matrix is ​​X = (x ij ) m×n ; For all indicators of the sample to be evaluated, the difference between the final score of each indicator under the optimal weight and the score under the subjective and objective weight scheme should be as small as possible; the square of the deviation between the subjective and objective weight and the combined weight is calculated by comparing the sample data x ij The product and minimum of is the optimization goal. First, a combined weighting model is established. Assuming that the weight vectors determined by the fuzzy analytic hierarchy process and the CRITIC-entropy weight method are α and β respectively, the combined weight vector to be obtained is w, and a combined weight model is established: The optimal solution is obtained by using the Lagrange multiplier method, and the function L(ω,λ) is constructed, where λ is the Lagrange multiplier: From the extreme value existence conditions, we can know that: Solving the above equation and transposing it, we can get the combined weights of each dimension w=[w1,w2,…,w n ].

10. The quality assessment method for multivariate time series data of a distribution network according to claim 1, characterized in that: The calculation of the total quality score D of the data to be tested in step S104 includes: D=wC i In the formula, w is the combined weight of each dimension, C i is the quality score of each dimension; The evaluation data quality levels include: According to the total score D of the data quality to be tested, the data quality is divided into five levels, including: "fail", If the total quality score of the data to be tested is D∈(0,a], the data quality level of the data set to be tested is evaluated as "failed", "passed", "medium", "good" and "excellent"; If the total quality score of the data to be tested is D∈(a,b], the data quality level of the data set to be tested is evaluated as "passing"; If the total quality score of the data to be tested is D∈(b,c], the data quality level of the data set to be tested is assessed as "medium"; If the total quality score of the data to be tested is D∈(c,d], the data quality level of the data set to be tested is evaluated as "good"; If the total quality score of the data to be tested is D∈(d,e], the data quality level of the data set to be tested is evaluated as "excellent"; Where a, b, c, d, and e are preset grade division intervals.

Images