Fuzzy density electricity consumption anomaly detection method based on granular ball calculation

Abstract

The invention discloses a fuzzy density electricity consumption anomaly detection method based on granular ball calculation, and belongs to the field of electric power data analysis, and the method comprises the steps: carrying out the normalization processing of electricity consumption data with mixed attributes, and obtaining standard data; dividing the hybrid power utilization data into pellets according to the data distribution to obtain a pellet set; constructing a particle-ball fuzzy information particle set based on the particle-ball center; calculating particle ball fuzzy density and attribute importance; according to the particle-ball fuzzy density and the attribute importance degree, constructing a particle-ball anomaly score; obtaining an abnormal score of the sample through abnormal score mapping; and finally, judging abnormal conditions of all sample points according to a set threshold value. According to the method, an existing anomaly detection model based on the finest granularity sample is changed, a particle ball generation method of the mixed data is provided, the particle ball fuzzy density is constructed by using the fuzzy information granularity theory so as to describe the polymerization degree of the particle balls, and unsupervised anomaly detection of the mixed power consumption data is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of power data analysis, and mainly relates to realizing a fuzzy density power consumption anomaly detection method based on granular sphere calculation. Background Art

[0002] In the smart grid industry, the detection of abnormal power consumption patterns is of great importance, as it is not only related to the stable operation of the power grid but also directly affects the economic well-being of power companies. In addition, improper power consumption behavior may trigger safety hazards, posing a threat to the power grid infrastructure and user safety. In view of this, developing an effective abnormal power consumption detection mechanism has significant research value and practical application importance for ensuring the security of energy supply, the stable operation of the power grid, and protecting the economic interests of power enterprises. Although current machine learning technologies have played an important role in the analysis and mining of power big data, there are still some challenges in dealing with abnormal power consumption data detection. Existing methods mostly focus on the analysis of single-type data, and the processing of mixed-type data often requires format conversion, which may lead to the loss of key information; most existing methods rely on a point-to-point calculation method, which not only has a large amount of calculation but also has a high sensitivity to noise and is easily interfered by it; most existing methods rely on a large amount of manually labeled data, which not only limits the improvement of algorithm performance but also increases the implementation cost. Summary of the Invention

[0003] In view of the above deficiencies in the prior art, a fuzzy density power consumption anomaly detection method based on granular sphere calculation provided by the present invention not only improves the efficiency of processing large-scale data sets but also solves the problem of information loss that may occur during data format conversion. Through this method, the present invention can effectively perform unsupervised anomaly detection on a mixed power data set containing numerical and categorical data.

[0004] To achieve the above invention objective, the technical solution adopted by the present invention is: a fuzzy density power consumption anomaly detection method based on granular sphere calculation, including the following steps:

[0005] S1. Normalize the power consumption data with mixed attributes to obtain the data after standard processing;

[0006] S2. Divide the mixed power consumption data into granular spheres according to the data distribution to obtain a granular sphere set;

[0007] S3. Construct a granular sphere fuzzy information granule set according to the granular sphere center set;

[0008] S4. Calculate the granular sphere fuzzy density and attribute importance according to the granular sphere fuzzy information granule;

[0009] S5. Construct the granule sphere anomaly score according to the granule sphere fuzzy density and the attribute importance degree;

[0010] S6. Map the granule sphere anomaly score to obtain the anomaly score of the sample;

[0011] S7. Judging one by one whether the scores of all sample points exceed the set threshold according to the set threshold. If so, judge it as an abnormal sample; otherwise, judge it as a normal sample. Repeat step S8 until all samples are judged.

[0012] The beneficial effects of the present invention are as follows: The granule sphere calculation technology is used to perform adaptive granule sphere representation on the data set, effectively simplifying the data processing flow; by constructing a fuzzy relationship matrix, the information loss that may be brought about by data format conversion is cleverly avoided, ensuring the integrity and accuracy of the data; the granule sphere fuzzy density is used to describe the uncertainty and outlier characteristics of the granule sphere fuzzy information granule, and then the fuzzy entropy is used to calculate the importance degree of the attribute to weighted calculate the outlier degree of the granule sphere and the outlier degree of the sample; the present invention does not require labeled data for model training and can effectively realize unsupervised anomaly detection of hybrid power consumption data.

[0013] Further, the specific expression of the normalization process in step S1 is:

[0014]

[0015] where F(·) is the maximum-minimum normalization process; a(o i ) is the value of the power consumption data sample o i on the numerical attribute a; max(a) is the maximum value of the data on the attribute a; min(a) is the minimum value of the data on the attribute a;

[0016] The beneficial effect of the above further solution is: Normalize the numerical data in the power consumption data, eliminate the gap between different feature dimensions and numerical ranges, and ensure comparability.

[0017] Further, step S2 is specifically as follows:

[0018] S21. Initialize the entire data set as a granule sphere;

[0019] S22. Calculate the center c k and radius r k of each granule sphere GB k in the granule sphere set:

[0020]

[0021] r k = max(dis(c k , o l ))

[0022] Among them, a(c k ) is the value of the granule center c k under the attribute a; n k is the number of samples in the granule GB k ; mode(a) is the mode of the values of the samples in the granule GB k under the attribute a; δ(a j (c k ), a j (o l )) is an indicator function, which is 1 when a j (c k ) = a j (o l ), and 0 otherwise;

[0023] S23. Calculate the mass of the granule GB k . Assume that the granule GB k splits, and calculate the masses of the two sub - granules:

[0024]

[0025]

[0026] Among them, DM k is the mass calculation function of the granule GB k ; is the calculation function of the weighted density sum of the two sub - granules and ; n k is the number of sample points in the granule GB k ; and are respectively the number of sample points in the two sub - granules and ; is the sum of the distances from all sample points in all granules to the granule center; is the weighted mass of the two sub - granules;

[0027] S24. For all granules in the granule set, if the weighted mass of the sub - granule is higher, the granule splits; otherwise, the granule does not split;

[0028] S25. Repeat step S24 until the number of granules in the granule set no longer increases.

[0029] S26. Calculate the average radius r mean and the radius median t median of the granule set. For the granules in the granule set with a radius greater than 2×max(rmean , r median The granulocytes of ) divide one more time.

[0030] The beneficial effects of the above further solution are as follows: By using granulocyte computing technology to perform adaptive granulocyte representation on the data set, the data processing flow is effectively simplified.

[0031] Further, the specific steps of step S3 are as follows:

[0032] S31. Calculate the fuzzy relationship generated by each attribute according to the attribute set:

[0033]

[0034] Among them, R a (x, y) is the similarity measure between sample x and sample y; a(x) is the value of sample x under attribute a; R a is the fuzzy relationship generated by attribute a; σ is an adjustable parameter;

[0035] S32. According to the fuzzy relationship R a , the specific expression of the granulocyte fuzzy granule structure guided by R a is as follows:

[0036] G GB (R a ) = {[c 1 a , [c 2 a , …, [c k a}

[0037] Among them, k is the number of granulocytes in the granulocyte set; c i is the center of the granulocyte GB i ; [c i a = (R a (c i , c 1 ), R a (c i , c 1 ), …, R a (c i , c k )) is the granulocyte fuzzy granule generated by c i under the fuzzy relationship R a ; G GB (R a ) is the granulocyte fuzzy granule family generated under the fuzzy relationship R a ; is the cardinality of [c i a . ​​​​​

[0038] The beneficial effects of the above further solution are as follows: By constructing a fuzzy relation matrix, the information loss that may be caused by data format conversion is skillfully avoided, ensuring the integrity and accuracy of the data.

[0039] Furthermore, the specific steps of step S4 are as follows:

[0040] S41. According to the attribute subset A in the attribute set AT, the granular ball fuzzy density is specifically calculated as follows:

[0041]

[0042] where FD A (c) is the calculation function of the granular ball fuzzy density under the attribute subset A; c is the center of the granular ball GB; Dens a (c) = ∑ x∈C [c] a (x) / |C|; C is the set of granular ball centers; |C| is its cardinality;

[0043] S42. According to each granular ball in the granular ball set, calculate the granular ball fuzzy density FD a (c) for each attribute one by one;

[0044] S43. According to the attribute subset A in the attribute set, the granular ball fuzzy entropy is specifically calculated as follows:

[0045]

[0046] where FE GB (A) is the calculation function of the granular ball fuzzy entropy of the granular ball set under the attribute subset A; [c] A is the granular ball fuzzy granule generated under the fuzzy relation R A ; is the cardinality of [c] A ;

[0047] S44. According to the granular ball fuzzy entropy, the importance degree of attribute a is defined as:

[0048]

[0049] where W(a) is the calculation function of the importance degree of attribute a; FE GB (a) is the granular ball fuzzy entropy of attribute a; AT is the attribute set; |AT| is the cardinality of the attribute set AT, that is, the number of attributes in the set;

[0050] S45. According to each attribute in AT, calculate the attribute importance degree W(a) of each attribute one by one.

[0051] The beneficial effects of the above further solution are as follows: The uncertainty of information granules is characterized by the fuzzy density of granule balls, and the importance is calculated for each attribute according to the granule ball fuzzy entropy, reflecting the variation degree and information content of the attribute in the dataset.

[0052] Further, the expression of the abnormal score of the granule ball fuzzy density in step S5 is as follows:

[0053]

[0054] where OS GB (c) is the abnormal score calculation function of the granule ball; |GB| is the number of sample points in the granule ball; c is the center of the granule ball; AT is the attribute set; W(a) is the importance of attribute a; FD a (c) is the granule ball fuzzy density under attribute a.

[0055] The beneficial effects of the above further solution are as follows: By weighted fusion of the fuzzy densities of multiple attributes, comprehensive recognition of abnormal patterns is achieved. This fusion strategy not only considers the abnormality of a single attribute, but also enhances the overall accuracy and reliability of anomaly detection through the importance weights between attributes.

[0056] Further, the expression of the abnormal score of the sample in step S6 is as follows:

[0057] OS(o) = OS GB (c k ), s.t. o ∈ GB k

[0058] where o is any sample; OS(o) is the abnormal score of sample o; GB k is the granule ball to which sample o belongs; c k is the center of the granule ball GB k ; OS GB (c k ) is the abnormal score of the granule ball GB k .

[0059] The beneficial effects of the above further solution are as follows: The abnormality of the sample point is determined by mapping through the abnormal score of its corresponding granule ball. This mapping mechanism not only significantly reduces the computational complexity, but also ensures the stability and reliability of the detection process. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 is a flowchart of the anomaly detection method of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0061] To further clarify the present invention, the following content will describe the present invention in detail with reference to the accompanying drawings. It should be noted that the provided embodiments are intended to illustrate rather than limit the scope of protection of the present invention. The examples are intended to enable those skilled in the art to fully understand the present invention and be able to implement and apply it according to the basic idea of the present invention. On the basis of not violating the essence and core principles of the present invention, various substitutions and changes made according to the needs of technological development and practical applications should be covered within the scope of protection of the present invention.

[0062] As Figure 1 shown, in one embodiment of the present invention, a fuzzy density-based electricity anomaly detection method using granular computing includes the following steps:

[0063] S1. Normalize the electricity consumption data with mixed attributes to obtain the data after canonical processing;

[0064] S2. Divide the mixed electricity consumption data into granular balls according to the data distribution to obtain a granular ball set;

[0065] S3. Construct a granular ball fuzzy information granule set according to the granular ball center set;

[0066] S4. Calculate the granular ball fuzzy density and attribute importance according to the granular ball fuzzy information granules;

[0067] S5. Construct a granular ball anomaly score according to the granular ball fuzzy density and attribute importance;

[0068] S6. Map the granular ball anomaly score to obtain the anomaly score of the sample;

[0069] S7. Judging one by one whether the scores of all sample points exceed the set threshold according to the set threshold. If so, judge it as an abnormal sample; otherwise, judge it as a normal sample. Repeat step S8 until all samples are judged.

[0070] In this embodiment, in the specific application of the fuzzy rough set, the electricity consumption data of the power grid is integrated and input into a comprehensive information system (or information table), which organizes the data in tabular form. Among them, each row represents an independent sample or electricity consumption entity, and each column represents different attributes or features, and these attributes together constitute the analysis dimension. The attributes can be continuous numerical data, including voltage, power, etc.; they can be discrete nominal data, including equipment status, user mode, etc. Such an information system can be represented as (U, A), where U is the set of all samples and A is the set of all attributes.

[0071] The specific content of step S1 is as follows:

[0072] S11. Perform min-max normalization on the electricity consumption data with mixed attributes, and its specific expression is as follows:

[0073]

[0074] Among them, F(·) is the maximum-minimum normalization process; a(o i ) is the value of the electricity consumption data sample o i on the numerical attribute a; max(a) is the maximum value of the data on the attribute a; min(a) is the minimum value of the data on the attribute a;

[0075] S12. Normalize the numerical attribute data in all electricity consumption data, and do not normalize the nominal data, to obtain the normalized electricity consumption data.

[0076] In this example, the electricity consumption data information system of the power grid and the normalization results are shown in the following table:

[0077] Table 1 Electricity Consumption Data Information System and Normalization Results

[0078]

[0079] In Table 1, a 1 is the nominal attribute, a 2 and a 3 are numerical attributes.

[0080] The specific steps of S2 are as follows:

[0081] S21. Initialize the entire electricity consumption data set as a granule ball;

[0082] S22. Calculate the center c k and radius r k of each granule ball GB k in the granule ball set:

[0083]

[0084] r k = max(dis(c k , o l ))

[0085] Among them, a(c k ) is the value of the granule ball center c k under the attribute a; n k is the number of samples in the granule ball GB k ; mode(a) is the mode of the values of the samples in the granule ball GB k under the attribute a; δ(a j (c k ), a j (o l)) is an indicator function that is 1 when a j (c k ) = a j (o l ) and 0 otherwise;

[0086] S23. Calculate the mass of granule GB k . Assuming granule GB k splits, calculate the masses of the two sub - granules:

[0087]

[0088] where DM k is the mass calculation function of granule GB k ; is the calculation function of the weighted density sum of the two sub - granules and ; n k is the number of sample points in granule GB k ; and are the numbers of sample points in the two sub - granules and respectively; is the sum of the distances from all sample points in all granules to the granule center; is the weighted mass of the two sub - granules;

[0089] S24. For all granules in the granule set, if the weighted mass of the sub - granule is higher, the granule splits; otherwise, the granule does not split.

[0090] S25. Repeat step S24 until the number of granules in the granule set no longer increases.

[0091] S26. Calculate the average radius r mean and the median radius r median of the granule set. For granules in the granule set with a radius greater than 2×max(r mean , r median ), split them one more time.

[0092] In this embodiment, first, the entire data set is used as a granule GB = {o 1 , o 2 , o 3 , o 4 , o 5 , o 6}. It is split into two sub - granules {o 1 , o 4 , o 5} and {o 2 , o 3 , o6}. The mass of the granule is approximately 0.763, the weighted mass of the granule is approximately 0.469, and the overall mass is higher after splitting, so it is classified. Repeat this process to obtain the granule set {{o 1}, {o 4 , o 5}, {o 2 , o 3 , o 6}}. The radii in the granule set are 0, 0.273, 0.31 respectively, and the average radius r mean = 0.194, and the median radius r median = 0.273. Therefore, no further splitting is required, and the granule set {GB 1 , GB 2 , GB 3} is obtained.

[0093] Step S3 is specifically as follows:

[0094] S31. Calculate the fuzzy relationship generated by each attribute according to the attribute set:

[0095]

[0096] Among them, R a (x, y) is the similarity measure between sample x and sample y; a(x) is the value of sample x under attribute a; R a is the fuzzy relationship generated by attribute a; σ is an adjustable parameter;

[0097] S32. According to the fuzzy relationship R a , the specific expression of the granule fuzzy granule structure guided by R a is:

[0098] G GB (R a ) = {[c 1 a , [c 2 a , …, [c k a}

[0099] Among them, k is the number of granules in the granule set; c i is the center of granule GB i ; [c i a = (R a (c i , c 1 ), R a (c i , c 1 ), …, R a (c​​​​i , c k )) is c i The granular ball fuzzy granule generated under the fuzzy relation R a ; G GB (R a ) is the granular ball fuzzy granule family generated under the fuzzy relation R a ; is [c i a 's cardinality.

[0100] In this embodiment, let the adjustable parameter σ = 1, and the calculated fuzzy relation matrix is:

[0101]

[0102] The specific steps of step S4 are as follows:

[0103] S41. According to the attribute subset A in the attribute set AT, the granular ball fuzzy density is specifically calculated as:

[0104]

[0105] where FD A (c) is the calculation function of the granular ball fuzzy density under the attribute subset A; c is the center of the granular ball GB; Dens a (c) = ∑ x∈C [c] a (x) / |C|; C is the set of granular ball centers; |C| is its cardinality;

[0106] S42. According to each granular ball in the granular ball set, calculate the granular ball fuzzy density FD a (c) for each attribute one by one;

[0107] S43. According to the attribute subset A in the attribute set, the granular ball fuzzy entropy is specifically calculated as:

[0108]

[0109] where FE GB (A) is the granular ball fuzzy entropy calculation function of the granular ball set under the attribute subset A; [c] A is the granular ball fuzzy granule generated under the fuzzy relation R A ; is [c] A 's cardinality;

[0110] S44. According to the granular ball fuzzy entropy, the importance degree of attribute a is defined as:

[0111]

[0112] ​Among them, W(a) is the calculation function of the importance degree of attribute a; FE GB (a) is the granular ball fuzzy entropy of attribute a; AT is the attribute set; |AT| is the cardinality of the attribute set AT, that is, the number of attributes in the set;

[0113] S45. According to each attribute in AT, calculate the attribute importance degree W(a) of each attribute one by one.

[0114] In this embodiment, calculate the granular ball fuzzy density under each attribute one by one, which are respectively:

[0115]

[0116] Subsequently, calculate the importance degree of each attribute one by one. For attribute a 1 , its granular ball fuzzy entropy is calculated as Similarly, the granular ball fuzzy entropies of attribute a 2 and attribute a 3 are 0.679 and 0.593 respectively. According to the granular ball fuzzy entropies of each attribute, the attribute importance degrees of each attribute can be obtained as: W(a 1 ) = 0.554, W(a 2 ) = 0.238, W(a 3 ) = 0.208.

[0117] Furthermore, the expression of the granular ball fuzzy density anomaly score in step S5 is:

[0118]

[0119] Among them, OS GB (c) is the calculation function of the anomaly score of the granular ball; |GB| is the number of sample points in the granular ball; c is the center of the granular ball; AT is the attribute set; W(a) is the importance degree of attribute a; FD a (c) is the granular ball fuzzy density under attribute a.

[0120] In this embodiment, by weighted fusion of the fuzzy densities of multiple attributes, a comprehensive recognition of the abnormal pattern is achieved. This fusion strategy not only considers the abnormality of a single attribute, but also enhances the overall accuracy and reliability of anomaly detection through the importance weights between attributes. Calculate the anomaly score of c 1 Similarly, OS GB (c 2 ) = 0.254, OS GB (c 3 ) = 0.177.

[0121] ​Further, the expression for the anomaly score of the electricity consumption data sample in step S6 is as follows:

[0122] OS(o) = OS GB (c k ), s.t. o ∈ GB k

[0123] where o is an arbitrary electricity consumption data sample; OS(o) is the anomaly score of sample o; GB k is the granule to which sample o belongs; c k is the center of granule GB k ; OS GB (c k ) is the anomaly score of granule GB k .

[0124] In this embodiment, the anomaly of the electricity consumption data sample is mapped and determined by the corresponding granule anomaly score. OS(o 1 ) = OS GB (c 1 ) = 0.552. Similarly, it can be obtained that: OS(o 2 ) = 0.177, OS(o 3 ) = 0.177, OS(o 4 ) = 0.254, OS(o 5 ) = 0.254, OS(o 6 ) = 0.177.

[0125] Further, step S7 is specifically as follows:

[0126] For a set outlier threshold μ, if the outlier degree OS(o i ) of the electricity consumption data sample o i is greater than the given outlier threshold μ, then o i is determined as an outlier. By comparing the outlier degrees of all samples with the threshold one by one, the abnormal electricity consumption samples in the system can be calculated.

[0127] In this embodiment, by comparing the outlier degrees of all electricity consumption data samples, obviously, the outlier degree of the electricity consumption data sample o 1 is significantly greater than that of the other samples. The outlier threshold is set to μ = 0.4. By comparing the outlier degrees of all electricity consumption data samples with the threshold μ one by one, it can be calculated that the electricity consumption data sample o 1 is an outlier.

Claims

1. A fuzzy density power consumption anomaly detection method based on granular sphere calculation, characterized in that: The following steps are involved: S1. Normalize the electricity consumption data with mixed attributes to obtain standardized data; S2, dividing the mixed electricity consumption data into spheres according to data distribution, thereby obtaining a sphere set; S3, constructing a set of granular fuzzy information granules according to the granular sphere center set; S4, calculating the particle sphere fuzzy density and attribute importance according to the particle sphere fuzzy information particles; S5, constructing a particle anomaly score based on the particle fuzzy density and attribute importance; S6, obtaining the sample anomaly score according to the particle-sphere anomaly score mapping; S7. Determine whether the scores of all sample points exceed the threshold one by one according to the set threshold. If so, determine them as abnormal samples. Otherwise, determine them as normal samples. Repeat step S8 until all samples are determined.

2. According to claim 1, the fuzzy density power consumption anomaly detection method based on granular sphere calculation is characterized in that: The step S1 is specifically as follows: S11. Perform minimum-maximum normalization processing on the electricity consumption data of mixed attributes. The specific expression is as follows: Among them, F(·) is the maximum-minimum normalization process; a(o i ) is the electricity consumption data sample o i The value on the numerical attribute a; max(a) is the maximum value of the data on attribute a; min(a) is the minimum value of the data on attribute a; S12. Normalize all numerical attribute data, but do not normalize the nominal data, to obtain standardized data.

3. The fuzzy density power consumption anomaly detection method based on granular sphere calculation according to claim 1 is characterized in that: The step S2 is specifically as follows: S21, initialize the entire data set as a sphere; S22. Calculate each ball GB in the ball set k Center of k and radius r k : r k =max(dis(c k ,o l )) Among them, a(c k ) is the center of the sphere c k The value under attribute a; n k GB k The number of samples in mode(a) is the granular sphere GB k The mode of the values ​​of the samples under attribute a; δ(a j (c k ), a j (o l )) is the indicator function, when a j (c k )=a j (o l ) is 1, otherwise it is 0; S23. Calculate the particle sphere GB k The mass of the sphere GB k Split and calculate the masses of the two sub-balls: Among them, DM k GB k The mass calculation function of There are two balls and The calculation function of the weighted density sum; n k GB k The number of sample points in and There are two balls respectively. and The number of sample points in ; It is the sum of the distances from all sample points in all spheres to the center of the sphere; is the weighted mass of the two sub-balls; S24. For all the spheres in the sphere set, if the number of sample points in the sphere is greater than 2 and the weighted quality of the sub-sphere is higher, the sphere is split, otherwise the sphere is not split; S25, repeat step S24 until the number of balls in the ball set no longer increases; S26. Calculate the average radius r of the sphere set mean and the median radius r median , for the spheres with a radius greater than 2×max(r mean , r median ) of the ball and split it again.

4. The fuzzy density power consumption anomaly detection method based on granular sphere calculation according to claim 1 is characterized in that: The step S3 is specifically as follows: S31. Calculate the fuzzy relationship generated by each attribute according to the attribute set: Among them, R a (x, y) is the similarity measure between sample x and sample y; a(x) is the value of sample x under attribute a; R a is the fuzzy relationship generated by attribute a; σ is an adjustable parameter; S32, according to the fuzzy relationship R a , by R a The specific expression of the derived particle-sphere fuzzy particle structure is: G GB (R a )={[c1] a ,[c2] a ,…,[c k ] a } Where k is the number of balls in the ball set; c i GB i The center of i ] a =(R a (c i , c1), R a (c i , c1),…,R a (c i , c k )) is c i In the fuzzy relation R a The spherical fuzzy particles generated under G GB (R a ) is the fuzzy relation R a The fuzzy particle clusters of the spheres generated below; For [c i ] a The cardinality of .

5. The fuzzy density power consumption anomaly detection method based on granular sphere calculation according to claim 1 is characterized in that: The step S4 is specifically as follows: S41. According to the attribute subset A in the attribute set AT, the particle sphere fuzzy density is specifically calculated as: Among them, F DA (c) is the calculation function of the particle sphere fuzzy density under the attribute subset A; c is the center of the particle sphere GB; Dens a (c) = ∑ x∈C [c] a (x) / |C|; C is the set of particle sphere centers; |C| is its cardinality; S42, according to each particle in the particle set, calculate the particle fuzzy density FD under each attribute one by one a (c); S43, according to the attribute subset A in the attribute set, the granular fuzzy entropy is specifically calculated as: Among them, FE GB (A) is the spherical fuzzy entropy calculation function of the spherical set under the attribute subset A; [c] A In the fuzzy relation R A The spherical fuzzy particles produced below; Yes [c] A The cardinality of S44. According to the granular fuzzy entropy, the importance of attribute a is defined as: Among them, W(a) is the importance calculation function of attribute a; FE GB (a) is the granular fuzzy entropy of attribute a; AT is the attribute set; |AT| is the cardinality of the attribute set AT, that is, the number of attributes in the set; S45. Calculate the attribute importance W(a) of each attribute in AT one by one.

6. The fuzzy density power consumption anomaly detection method based on granular sphere calculation according to claim 1 is characterized in that: The expression of the particle-ball abnormality score in step S5 is: Among them, OS GB (c) is the calculation function of the anomaly score of the sphere; |GB| is the number of sample points in the sphere; c is the center of the sphere; AT is the attribute set; W(a) is the importance of attribute a; FD a (c) is the granular blur density under attribute a.

7. The fuzzy density power consumption anomaly detection method based on granular sphere calculation according to claim 1 is characterized in that: The expression of the sample anomaly score in step S6 is: OS(o)=OS GB (c k ),hundred∈GB k Where o is any sample; OS(o) is the abnormal score of sample o; GB k is the sphere to which sample o belongs; c k GB k Center of OS GB (c k ) is the particle ball GB k The anomaly score.

Images