A method for identifying wrong households in the special transformer industry based on electricity consumption characteristics and outlier detection
Through PCA dimensionality reduction and LOF outlier detection algorithm, the abnormality of power consumption characteristics of special-change users was identified, which solved the problem that traditional methods could not identify information errors in special-change industries, and achieved accurate identification and energy management of special-change users. promote.
Patent Information
- Application Number
- CN202010539733.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2020-06-12
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2040-06-12
AI Technical Summary
The existing technology cannot effectively identify users with incorrect industry information, resulting in poor energy use management, traditional classification methods cannot be updated in time, and accurate identification of industry information.
Using PCA dimensionality reduction algorithm and LOF outlier detection algorithm, normalize and reduce the electricity consumption data of special variable users, calculate the LOF coefficient, and use the elbow method to determine the threshold to identify users whose LOF coefficient is greater than the threshold. As a suspected user of industry classification errors to be investigated.
It realizes the rapid and accurate identification of users with information errors in the special-change industry, improves the accuracy of energy management and the optimization of the power consumption structure, and provides a basis for the special-change user profile management and load prediction.
Smart Images

Figure CN111861785B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of electric power operation and maintenance, and in particular to a method for identifying wrong households in a special transformer industry based on power consumption characteristics and outlier detection. Background Art
[0002] The electricity consumption characteristics of power users in different industries are not the same. If we do not distinguish between industries and only analyze the load characteristics of power users as a whole, it will be difficult to accurately grasp the electricity consumption characteristics of each user and the accuracy will be low. In order to ultimately achieve the optimization of the electricity consumption structure and the efficient management of electricity consumption behavior, it is necessary to start from each industry and analyze the load characteristics and electricity consumption habits of each industry.
[0003] The traditional classification method of special transformer industry usually divides the types of special transformers according to the business scope of the industry. For example, the "Classification of Electricity Consumption in National Economic Industries" is used as a reference for a detailed and accurate classification catalog of special transformer power users' loads. In the electricity information management system, the industry archive information of special transformer users is generally determined by the business of the enterprise to which the user corresponds. However, due to the rapid changes in the business of enterprises in some regions and industries, when changes occur, the archive information in the marketing management system often cannot be updated in time, resulting in the inability to carry out energy management of special transformers well.
[0004] The existing research on load feature classification of power users of special transformers is essentially a manifestation of clustering methods in data mining, which classifies users with the same or similar power consumption features into one category, and ultimately achieves similar power consumption features for users of the same category, while there are significant differences in power consumption features for users of different categories. This classification is usually carried out based on the industry information of power users of special transformers, without identifying the authenticity of the industry information. In other words, the traditional special transformer industry classification method and load feature mining method cannot effectively identify users with incorrect information in the special transformer industry. Summary of the invention
[0005] The technical problem to be solved and the technical task proposed by the present invention are to improve and perfect the existing technical solutions, and provide a method for identifying wrong users in the special transformer industry based on power consumption characteristics and outlier detection, so as to effectively identify users with wrong information in the special transformer industry. To this end, the present invention adopts the following technical solutions.
[0006] A method for identifying wrong households in the specialized transformer industry based on electricity consumption characteristics and outlier detection, characterized by comprising the steps of:
[0007] 1) Collect the daily electricity consumption data of the dedicated transformer users in a certain industry within the region for 365 days a year and normalize their daily electricity consumption;
[0008] 2) Input the daily electricity consumption data of 365 days into the PCA dimensionality reduction algorithm, sort them according to the size of the principal component eigenvalues, and select the principal component vector with the largest eigenvalue representing the electricity consumption characteristics as the feature data after dimensionality reduction;
[0009] 3) Calculate the LOF coefficient of the feature data of the special variable user after dimensionality reduction in the new feature space;
[0010] 4) Using the elbow method to determine the LOF coefficient threshold;
[0011] 5) Select users with LOF coefficients greater than the threshold as suspected users of industry classification errors to be investigated.
[0012] This technical solution adopts the PCA dimensionality reduction algorithm and the LOF outlier detection algorithm. First, the daily electricity consumption data of the special transformer users in an industry within the regional scope for 365 days a year is collected and normalized. Due to the high dimension, the analysis results may be disturbed by repeated and useless information. Therefore, the principal component analysis (PCA) is used to reduce the dimension of the original high-dimensional data, and the principal component with a larger eigenvalue representing the power consumption characteristics is retained as the input of the local outlier factor algorithm (LOF); the local density coefficient is calculated according to the LOF algorithm, and the elbow rule is used to determine the potential special transformer users with incorrect industry file classification. In this way, it is possible to quickly and accurately identify the special transformer users with incorrect file classification in an industry based on the typical power consumption information of different industries.
[0013] As a preferred technical means: in step 1), according to the industry classification information of the special transformer in the electricity consumption information collection system, the daily electricity consumption data of the special transformer of a certain industry in the region for 365 days a year is used as input, and its daily electricity consumption is normalized. The normalization formula is:
[0014]
[0015] Where: X* is the normalized daily electricity consumption for one year; X is the original daily electricity consumption data; X max is the maximum daily electricity consumption in a year. The normalization method can scale the electricity consumption curves of all special transformers to [0,1], and retain the information of power outage, peak-to-valley ratio, and load rate.
[0016] As a preferred technical means: in step 2) including the steps of:
[0017] 201) Input the normalized daily power consumption matrix Z of the industry n,365 =X*:
[0018]
[0019] Where: z i,j represents the normalized power consumption of the i-th dedicated transformer user on the j-th day; n is the number of dedicated transformer users;
[0020] 202) Calculate Z n,365 The center translation matrix That is, the daily power consumption minus the average power consumption of all special transformers on that day:
[0021]
[0022] In the formula is the average power consumption of all special transformers on day j; is the center translation matrix Elements of
[0023] 203) Calculation The covariance matrix C between the electricity consumption vectors of each day 365×365 :
[0024]
[0025] Where C i,j is the covariance between the total power consumption vector of the dedicated transformer on the i-th day and the total power consumption vector of the dedicated transformer on the j-th day; for The mean of
[0026] 204) Calculate the covariance matrix C 365×365 The 365 eigenvalues of 365 and the eigenvector E 365 :
[0027] Λ 365 =diag[λ1,λ2,…,λ 365 ]
[0028] E 365 =(e1,e2,…,e 365 )
[0029] Where: i is the i-th eigenvalue; e i is i The corresponding eigenvector;
[0030] 205) Sort the eigenvalues and select the eigenvalues Λ of some principal component vectors q =diag[λ1,λ2,…,λ q ], where q is determined by
[0031]
[0032] Where H d is the dimension reduction feature threshold. In order to retain the important feature components in the data, reduce the feature dimension, improve the computational efficiency and discard the noise in the data set, the value is 0.8;
[0033] 206) Calculate the retained q-dimensional feature matrix M after dimensionality reduction n,q :
[0034] M n,q =Z n,q ×E q
[0035] The PCA dimensionality reduction algorithm is used to reduce the dimension of normalized electricity consumption, and to screen the daily electricity consumption data and features that characterize the differences in electricity consumption between users, so as to improve the efficiency of identifying industry errors and abnormal electricity consumption.
[0036] As a preferred technical means: In step 3), the local outlier factor LOF algorithm is used to calculate the feature data M after the dimension reduction of the special variable user n,q In the new feature space R n×q The LOF coefficient in is used to determine whether there is atypical power consumption behavior in each special transformer. The calculation method is:
[0037] 301) Calculate the Euclidean distance d(i,j) between all special variant sample points and other special variants in the feature space. The calculation formula is:
[0038]
[0039] Where M i,m and M j,m is the feature data M n,q The elements in; the kth distance to the pth special variation sample is d k (p)=d(p,M k (p)), where M k (p) is the special variation sample that is k-th farthest from the special variation sample p;
[0040] 302) Calculate the kth distance neighborhood of sample point p, which is the kth distance d of p k All special variant samples within (p), that is, N k (p) = {M j ∈M n,q |d(p,j)≤d k (p)};
[0041] 303) Calculate the kth reachable distance from other samples o to sample p:
[0042] reach-dist k (p,o)=max{d k(o),d(p,o)}
[0043] This means that if sample p is in the kth distance neighborhood of sample o, then the kth reachable distance to sample o is equal to d k (o), otherwise d(p,o);
[0044] 304) Calculate the local reachable density of sample point p as follows:
[0045]
[0046] Local reachable density: The reciprocal of the average reachable distance from the remaining sample points in the kth neighborhood of the sample point p to p. This value represents a density. The higher the density, the more likely it is that they belong to the same cluster. The lower the density, the more likely it is that they are outliers.
[0047] 305) Calculate the local outlier factor LOF of sample point p:
[0048]
[0049] LOF represents the neighborhood point N of the sample point p k The average ratio of the local reachable density of (p) to the local reachable density of sample point p. The larger the value, the more likely the sample point p is an outlier.
[0050] Considering that the special transformer with wrong industry information has a production method that is inconsistent with the industry, its electricity consumption characteristics are quite different from the overall electricity consumption of the industry. Therefore, the local outlier factor (LOF) algorithm is adopted to calculate the characteristic data M after dimensionality reduction of the special transformer user. n,q In the new feature space R n×q The LOF coefficient in the data is used to detect the abnormal power consumption curve and determine whether there are industry information errors and abnormal power consumption in each special transformer.
[0051] As a preferred technical means: in step 4), after determining the LOF coefficients of all special variant samples in the feature space, the LOF coefficient threshold is determined by the elbow method according to the LOF coefficient sequence, and the calculation steps are as follows:
[0052] 401) Let L = {l1,l2,…,l n}={LOF(1),LOF(2),…,LOF(n)}, where n is the number of dedicated transformer users; i is the power consumption anomaly coefficient of dedicated transformer user i;
[0053] 402) Sort L from largest to smallest to form a new LOF sequence L * ;
[0054] 403) Calculate the sequence L *The adjacent fractional difference sequence ΔL of :
[0055]
[0056] In the formula is the LOF sequence L * The power consumption anomaly coefficient of the i-th in ΔL. Returns the sequence number i corresponding to the maximum value in ΔL max , then it is believed that is the LOF sequence L * elbow.
[0057] After determining the LOF coefficient of the special transformer sample as an abnormal indicator of its industry electricity consumption, in order to further determine the user list for the power supply company to conduct inspections, it is necessary to use the corresponding threshold optimization algorithm to determine the LOF coefficient threshold.
[0058] As a preferred technical means: in step 5), select the LOF outlier coefficient greater than the threshold The special-purpose users are treated as suspected users of incorrect industry classification to be investigated.
[0059] Beneficial effects: This technical solution adopts the PCA dimensionality reduction algorithm and the LOF outlier detection algorithm. First, the daily electricity consumption data of the special transformer users in an industry within the regional scope for 365 days a year is collected and normalized. Due to the high dimension, the analysis results may be disturbed by repeated and useless information. Therefore, the principal component analysis (PCA) is used to reduce the dimension of the original high-dimensional data, and the principal component with a larger eigenvalue representing the electricity consumption characteristics is retained as the input of the local outlier factor algorithm (LOF); the local density coefficient is calculated according to the LOF algorithm, and the elbow rule is used to determine the potential special transformer users with incorrect industry file classification. In this way, it is possible to quickly and accurately identify the special transformer users with incorrect file classification in an industry based on the typical electricity consumption information of different industries. Effective identification of users with incorrect information in the special transformer industry provides a basis for file management, load forecasting, and energy optimization of special transformer users. BRIEF DESCRIPTION OF THE DRAWINGS
[0060] Figure 1 It is a flow chart of the present invention.
[0061] Figure 2 It is the LOF outlier coefficient of the specialized user sample in the Internet access and related service industries.
[0062] Figure 3 It is the LOF outlier coefficient of the specialized user sample in the metal processing machinery manufacturing industry. DETAILED DESCRIPTION
[0063] The technical solution of the present invention is further described in detail below in conjunction with the accompanying drawings.
[0064] like Figure 1 As shown, Figure 1 The calculation process of the present invention is as follows: first, the daily electricity consumption data of the special transformer users in a certain industry in the regional scope for 365 days a year is collected, and the daily electricity consumption is normalized; then the 365-dimensional daily electricity consumption data is input into the PCA dimensionality reduction algorithm, sorted according to the size of the principal component eigenvalues, and the principal component vectors with larger eigenvalues representing the electricity consumption characteristics are selected as the feature data after dimensionality reduction; then the LOF coefficient of the feature data of the special transformer users after dimensionality reduction is calculated in the new feature space, and the elbow method is used to determine the LOF coefficient threshold; finally, the special transformer users whose LOF coefficient is greater than the threshold are selected as the suspected users of industry classification errors to be investigated. The specific steps are:
[0065] Step 1. According to the industry classification information of the special transformer in the electricity consumption information collection system, the daily electricity consumption data of the special transformer of a certain industry in the region for 365 days a year is taken as input, and its daily electricity consumption is normalized. The normalization formula is:
[0066]
[0067] Where: X* is the normalized daily electricity consumption for one year; X is the original daily electricity consumption data; X max is the maximum value of daily electricity consumption in a year. The normalization method can scale the electricity consumption curves of all special transformers to [0,1], and retain information such as power outage, peak-to-valley ratio, and load rate.
[0068] Step 2. Based on the normalized annual daily electricity consumption data of the industry's dedicated transformer users, the 365-dimensional daily electricity consumption data is input into the PCA dimensionality reduction algorithm, sorted according to the size of the principal component eigenvalues, and the principal component vectors with larger eigenvalues representing the electricity consumption characteristics are selected as the feature data after dimensionality reduction. The specific calculation steps are as follows:
[0069] 201) Input the normalized daily power consumption matrix Z of the industry n,365 =X*:
[0070]
[0071] Where: z i,j represents the normalized power consumption of the i-th dedicated transformer user on the j-th day; n is the number of dedicated transformer users;
[0072] 202) Calculate the center translation matrix of Zn,365 That is, the daily power consumption minus the average power consumption of all special transformers on that day:
[0073]
[0074] In the formula is the average power consumption of all special transformers on day j;
[0075] 203) Calculation The covariance matrix C between the electricity consumption vectors of each day 365×365 :
[0076]
[0077] Where Ci,j is the covariance between the total power consumption vector of the dedicated transformer on the i-th day and the total power consumption vector of the dedicated transformer on the j-th day;
[0078] 204) Calculate the covariance matrix C 365×365 The 365 eigenvalues of 365 And the eigenvector E365:
[0079] Λ 365 =diag[λ1,λ2,…,λ 365 ]
[0080] E 365 =(e1,e2,…,e 365 )
[0081] 205) Sort the eigenvalues and select the eigenvalues Λ of some principal component vectors q =diag[λ1,λ2,…,λ q ], where q is determined as follows:
[0082]
[0083] Where H d is the dimension reduction feature threshold. In order to retain the important feature components in the data, reduce the feature dimension, improve the computational efficiency and discard the noise in the data set, the value is 0.8;
[0084] 206) Calculate the retained q-dimensional feature matrix M after dimensionality reduction n,q :
[0085] M n,q =Z n,q ×E q .
[0086] Step 3: Use the local outlier factor (LOF) algorithm to calculate the feature data M after dimension reduction of the special variable user n,q In the new feature space R n×q The LOF coefficient in is used to determine whether there is atypical power consumption behavior in each special transformer. The calculation method is:
[0087] 301) Calculate the Euclidean distance d(i,j) between all special variant sample points and other special variants in the feature space. The calculation formula is:
[0088]
[0089] Then the kth distance to the pth special variation sample is d k (p)=d(p,M k (p)), where M k (p) is the special variation sample that is k-th farthest from the special variation sample p;
[0090] 302) Calculate the kth distance neighborhood of sample point p, which is the kth distance d of p k All special variant samples within (p), that is, N k (p) = {M j ∈M n,q |d(p,j)≤d k (p)};
[0091] 303) Calculate the kth reachable distance from other samples o to sample p:
[0092] reach-dist k (p,o)=max{d k (o),d(p,o)}
[0093] This means that if sample p is in the kth distance neighborhood of sample o, then the kth reachable distance to sample o is equal to d k (o), otherwise d(p,o);
[0094] 304) Calculate the local reachable density of sample point p as follows:
[0095]
[0096] Local reachable density: The reciprocal of the average reachable distance from the remaining sample points in the kth neighborhood of the sample point p to p. This value represents a density. The higher the density, the more likely it is that they belong to the same cluster. The lower the density, the more likely it is that they are outliers.
[0097] 305) Calculate the local outlier factor LOF of sample point p:
[0098]
[0099] LOF represents the neighborhood point N of the sample point p k The average ratio of the local reachable density of (p) to the local reachable density of sample point p. The larger the value, the more likely the sample point p is an outlier.
[0100] Step 4. After determining the LOF coefficients of all variant samples in the feature space, the LOF coefficient threshold is determined using the elbow method according to the LOF coefficient sequence. The calculation method is:
[0101] 401) Let L = {l1,l2,…,l n}={LOF(1),LOF(2),…,LOF(n)}, where n is the number of dedicated transformer users;
[0102] 402) Sort L from largest to smallest to form a new LOF sequence L * ;
[0103] 403) Calculate the adjacent fractional difference sequence ΔL of the sequence L*:
[0104]
[0105] In the formula is the LOF sequence L * The power consumption anomaly coefficient of the i-th in ΔL. Returns the sequence number i corresponding to the maximum value in ΔL max , then it is believed that is the LOF sequence L * The elbow (inflection point).
[0106] Step 5. Select LOF outlier coefficient greater than the threshold The special-purpose users are treated as suspected users of incorrect industry classification to be investigated.
[0107] The present invention will be further described below with specific examples:
[0108] 1. Data Source
[0109] The data mainly comes from the State Grid's electricity consumption information collection system, including 365-day daily electricity consumption data of special transformers in two industries, namely Internet access and related services and metal processing machinery manufacturing in a certain province in 2018. 100 special transformer users were selected from each industry, totaling more than 70,000 data items.
[0110] 2. Data Preprocessing
[0111] The annual electricity consumption data of each special transformer user were screened and cleaned, and user samples with a collection rate below 80% and obvious abnormal data were excluded. The daily electricity consumption data of 83 and 51 special transformer users in the remaining Internet access and related services, and metal processing machinery manufacturing industries in 2018 were analyzed, and their daily electricity consumption was normalized.
[0112] 3. Data Dimensionality Reduction and Outlier Analysis
[0113] The 365-dimensional daily electricity consumption data is input into the PCA dimensionality reduction algorithm, sorted according to the size of the principal component eigenvalues, and the principal component vectors that represent the electricity consumption characteristics are selected as the feature data after dimensionality reduction. The LOF coefficient of the feature data of the special transformer user after dimensionality reduction in the new feature space is calculated, and the result is as follows: Figure 2 , 3 shown.
[0114] The elbow rule is further used to identify and exclude suspected users with incorrect industry classification. Figure 2 , 3 It can be obtained that there are 5 and 6 special transformer users in the Internet access and related services and metal processing machinery manufacturing industries respectively as suspected users of misclassification to be investigated.
[0115] IV. Investigation results and accuracy
[0116] After actual investigation and proofreading, the actual industry classification results of these users are shown in Table 1. Table 1 Identification and investigation results of two suspected users with industry classification errors
[0117]
[0118] It can be seen from Table 1 that the identification precision rates of misclassified suspected users in the two industries of Internet access and related services and metal processing machinery manufacturing are 80% and 50% respectively, which shows that the present invention can effectively detect misclassified suspected users in various industries. The use of the elbow rule can make the identification precision rate higher. For industries such as Internet access and related services with more information errors, the scope of investigation can be appropriately expanded to improve its recall rate.
[0119] above Figure 1 The method for identifying wrong households in the specialized transformer industry based on electricity consumption characteristics and outlier detection shown is a specific embodiment of the present invention, which has embodied the substantial characteristics and progress of the present invention. According to actual use needs and under the guidance of the present invention, equivalent modifications in shape, structure, etc. can be made to the method, which are all within the protection scope of this scheme.
Claims
1. A method for identifying wrong households in the specialized transformer industry based on electricity consumption characteristics and outlier detection, characterized in that Includes steps: 1) Collect the daily electricity consumption data of the dedicated transformer users in a certain industry within the region for 365 days a year and normalize their daily electricity consumption; 2) Input the daily electricity consumption data of 365 days into the PCA dimensionality reduction algorithm, sort them according to the size of the principal component eigenvalues, and select the principal component vectors with larger eigenvalues representing the electricity consumption characteristics as the feature data after dimensionality reduction; 3) Calculate the LOF coefficient of the feature data of the special user after dimensionality reduction in the new feature space; 4) Using the elbow method to determine the LOF coefficient threshold; 5) Select users with LOF coefficient greater than the threshold as suspected users of industry classification errors to be investigated; In step 3), the local outlier factor LOF algorithm is used to calculate the feature data of the special variable user after dimensionality reduction M n,q In the new feature space The LOF coefficient in is used to determine whether there is atypical power consumption behavior in each special transformer. The calculation method is: 301) Calculate the Euclidean distance between all special variant sample points and other special variants in the feature space d ( i , j ), the calculation formula is: In the formula and For feature data M n,q Elements in For p The first k Distance is d k ( p ) =d ( p , M k ( p )),in M k ( p ) is the sample of the variation p No. k Far away special variation samples; 302) Calculate sample points p No. k The distance neighborhood is p No. k distance d k ( p ) within the specified range, that is, ; 303) Calculate other samples o To the sample p No. k Reachable distance: This means that if the sample p In the sample o No. k If the distance is within the neighborhood, then to the sample o No. k The reachable distance is equal to d k ( o ), otherwise d ( p , o ); 304) Calculate sample points p The local reachable density is: Locally reachable density sample points p No. k The remaining sample points in the neighborhood p The reciprocal of the average reachable distance. This value represents a density. The higher the density, the more likely it is that they belong to the same cluster. The lower the density, the more likely it is an outlier. 305) Calculate sample points p The local outlier factor LOF is: LOF represents the sample point p Neighborhood points N k ( p )’s local reachability density and sample points p The larger the value, the greater the average value of the ratio of the local reachable density of the sample point. p The more likely it is an outlier; In step 4), after determining the LOF coefficients of all variant samples in the feature space, the LOF coefficient threshold is determined using the elbow method according to the LOF coefficient sequence. The calculation steps are as follows: 401) Order L ={ l 1, l 2,…, l n }={ LOF(1), LOF(2),…, LOF( n )},in n is the number of dedicated users, l i For dedicated users i Abnormal power consumption coefficient; 402) Yes L New LOF sequence sorted from largest to smallest L * ; 403) Calculate sequence L * The sequence of adjacent fractional differences : In the formula LOF sequence L * Middle i The power consumption abnormality coefficient of The serial number corresponding to the maximum value i max , then it is believed that LOF sequence L * elbow.
2. According to claim 1, a method for identifying wrong households in the specialized transformer industry based on power consumption characteristics and outlier detection is characterized by: In step 1), according to the industry classification information of the special transformer in the electricity consumption information collection system, the daily electricity consumption data of the special transformer of a certain industry in the region for 365 days a year is taken as input, and its daily electricity consumption is normalized. The normalization formula is: Where: is the normalized daily electricity consumption for one year; X is the original daily electricity consumption data; X max is the maximum daily electricity consumption in a year. The normalization method can scale the electricity consumption curves of all special transformers to [0,1], and retain the information of power outage, peak-to-valley ratio, and load rate.
3. According to claim 2, a method for identifying wrong households in the specialized transformer industry based on power consumption characteristics and outlier detection is characterized by: In step 2) include: 201) Input the normalized daily power consumption matrix of the industry Z n,365 = : Where: z i,j Indicates i The user of the special variable is in the j Normalized electricity consumption per day; n The number of dedicated users; 202) Calculation Z n,365 The center translation matrix , that is, the daily power consumption minus the average power consumption of all special transformers on that day: Where: For the j The average daily power consumption of all special transformers, is the center translation matrix Elements of 203) Calculation The covariance matrix between the electricity consumption vectors for each day in : Where: C i,j For the i The total daily power consumption vector of the special transformer and the j The covariance between all daily dedicated transformer power consumption vectors; for The mean of 204) Calculate the covariance matrix 365 eigenvalues of and the eigenvector E 365 : Where: λ i For the i eigenvalues; e i for λ i The corresponding eigenvector; 205) Sort the eigenvalues and select the eigenvalues of some principal component vectors ,in q The method to determine is In the formula H d is the dimension reduction feature threshold. In order to retain the important feature components in the data, reduce the feature dimension, improve the computational efficiency and discard the noise in the data set, the value is 0.8; 206) Calculate the retained q Dimensional feature matrix M n,q : 。 4. According to claim 1, a method for identifying wrong households in the specialized transformer industry based on power consumption characteristics and outlier detection is characterized by: In step 5), select the LOF outlier coefficient greater than the threshold The special-purpose users are treated as suspected users of incorrect industry classification to be investigated.
