User abnormal load recognition method based on stereoscopic image library and dual-channel weighting

Abstract

The application discloses a user abnormal load identification method based on a three-dimensional portrait library and a double-channel weighting, comprising the following steps: collecting load data and industry labels of users, obtaining a load data set after data cleaning, and then dividing training and test data sets; constructing a user load characteristic analysis module, counting the imbalance degree between different users and between different industry categories in the training data set, clustering and extracting typical power consumption curves of users in multiple time scales and multi-dimensional load characteristic indexes; combining the user load characteristic analysis result with the industry label, weighting each industry user, and clustering and extracting industry multi-time scale typical power consumption curves and multi-dimensional load characteristic indexes for a second time to construct a three-dimensional portrait library; and after the load curve-feature double-channel weighting of each industry in the three-dimensional portrait library, identifying the load type and abnormal power consumption behavior of unknown users in the test data set. The application realizes high-accuracy identification of the load type and abnormal power consumption behavior of users.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart meters and big data analysis of electricity consumption, in particular to a user abnormal load identification method based on a stereoscopic portrait library and double-channel weighting. BACKGROUND

[0002] The characteristics and differences of user behavior at different time scales for different load types are very obvious and prominent. For example, during the holiday period, the electricity consumption of hotels, restaurants and bars and other places increases sharply; the electricity consumption behavior of resident charging piles is obviously different between commuting hours and non-commuting hours, and there is great pressure to ensure power supply and electricity safety, so it is necessary to identify the user load type to develop a reasonable power supply strategy; secondly, some users have abnormal electricity consumption behavior, such as residential electricity consumption for commercial use, which poses a huge electricity safety hazard. Therefore, an accurate user abnormal load identification method is of great significance to ensure and maintain a safe and standardized power supply environment.

[0003] Currently, there are the following difficulties in user abnormal load identification:

[0004] 1. The user load level changes significantly with fluctuations in meteorological factors and economic environment, making it difficult to characterize user electricity consumption habits and behavior characteristics; 2. The data between different users and different industry categories is imbalanced, and the identification strategy cannot focus on some users and industries with small sample data, resulting in reduced load identification accuracy; 3. The difference between user load types is small, and the intra-class diversity is high, making clustering difficult, which makes it difficult to effectively extract load features.

[0005] The above problems increase the difficulty of user abnormal load identification to some extent, reduce the accuracy of abnormal load identification, and make the traditional abnormal load identification method not accurate in identifying user load types and abnormal electricity consumption behavior.

[0006] Therefore, the user abnormal load identification method based on a stereoscopic portrait library and double-channel weighting is provided to achieve high-accuracy identification of user load types and illegal electricity consumption behavior. As a link in power distribution network operation and planning, it can better characterize user electricity consumption habits and behavior characteristics, effectively assist operation personnel in developing reasonable power supply strategies, promote the application of advanced data processing technology in power distribution network operation and planning scenarios, support the digitalization and intelligentization of power distribution networks, and help build a new power system. SUMMARY

[0007] The present application aims to overcome the deficiencies of the prior art, and provides a user abnormal load recognition method based on a stereoscopic portrait library and a double-channel weighting, which can realize high-accuracy recognition of user load types and abnormal electricity use behaviors, solves the problem of low recognition accuracy of user load types and abnormal electricity use behaviors in traditional load recognition methods, and as a link in power distribution network operation and planning, will help to ensure power supply and use safety.

[0008] To achieve the above object, the technical scheme provided by the present application is as follows: a user abnormal load recognition method based on a stereoscopic portrait library and a double-channel weighting, comprising the following steps:

[0009] 1) Collecting load data and industry labels of users, performing data cleaning on the load data, including outlier rejection, missing value filling and standardization, the load data after data cleaning forming a load data set, and dividing the load data set into a training data set and a test data set according to a proportion;

[0010] 2) Constructing a user load characteristic analysis module including a statistical module and a clustering module, statistically analyzing the imbalance degree between different users and between different industry categories in the training data set, clustering and extracting user multi-time scale typical electricity use curves, and then calculating user multi-dimensional load characteristic indexes by using the user multi-time scale typical electricity use curves to obtain user load characteristic analysis results;

[0011] 3) Combining the user load characteristic analysis results with the industry labels, calculating user and industry weighting coefficients according to the imbalance degree between different users and between different industry categories, improving the density peak clustering algorithm by weighting, and then using the improved density peak clustering algorithm to perform secondary clustering to obtain industry multi-time scale typical electricity use curves and calculate industry multi-dimensional load characteristic indexes, and using the industry multi-time scale typical electricity use curves and the multi-dimensional load characteristic indexes to construct a stereoscopic portrait library;

[0012] 4) Designing a load curve-feature double-channel weighting recognition strategy, weighting each industry in the stereoscopic portrait library by load curve-feature double channels, calculating the total correlation coefficients between unknown users in the test data set and each industry in the stereoscopic portrait library, and obtaining the recognition results of unknown user load types and abnormal electricity use behaviors.

[0013] Further, the specific operation steps of step 1) are as follows:

[0014] 1.1) Obtaining load data of each user saved in the form of daily load sequence;

[0015] 1.2) Traverse each user's daily load sequence, and judge the missing condition of the user's daily load sequence. If the daily load sequence is continuously missing more than 30% of the sampling data, the daily load sequence is discarded. For the remaining user's daily load sequence, the average value and the standard deviation of the daily load sequence are calculated using the following formula:

[0016]

[0017] In the formula, is the average value of the daily load sequence, x(i) is the load data of the i th sampling time point in the daily load sequence, i = 1, 2, …, n, n is the total sampling point number of the daily load sequence, and s is the standard deviation of the daily load sequence;

[0018] 1.3) Traverse the load data of each sampling time point in the daily load sequence. If the difference between the load data corresponding to the sampling time point and the average value of the daily load sequence is greater than 3 times the standard deviation of the daily load sequence, it is considered as abnormal data and is removed. After removal, the load data corresponding to the sampling time point is treated as a missing value;

[0019] 1.4) Traverse the user's daily load sequence, and judge whether the load data corresponding to the adjacent sampling time point of the missing value exists. If the load data corresponding to the adjacent sampling time points before and after the missing value are both missing values, the missing value is assigned as 0. If the sampling time point corresponding to the missing value is the first sampling time point or the last sampling time point in the daily load sequence, the load data corresponding to the only adjacent sampling time point is used to replace the missing value. If the load data corresponding to the adjacent sampling time points are both normal values, the average value of the load data corresponding to the adjacent sampling time points is used to replace the missing value;

[0020] 1.5) Traverse each sampling time point in the user's daily load sequence, and map the load data corresponding to the sampling time point to the interval [0, 1], and the calculation formula is as follows:

[0021]

[0022] In the formula, L max , L min respectively represent the maximum value and the minimum value of the user's daily load sequence, x i ′ is the standardized load data of the i th sampling time point.

[0023] Further, the specific operation steps of the step 2) are as follows:

[0024] 2.1) Construct a user load characteristic analysis module, which is composed of a statistical module and a clustering module, and is respectively used for statistical and division load data and clustering to obtain a user multi-time scale typical electricity consumption curve;

[0025] 2.2) The statistical module first calculates the imbalance degree between different users and between different industry categories in the training data set. The imbalance degree between different users includes the number of users contained in the user load data and the number of daily load sequences contained in each user. The imbalance degree between different industry categories includes the number of industry categories contained in the user load data and the total number of users of each industry category.

[0026] Then, the user industry label data is counted to generate the user industry label sequence H Y , as shown in the following formula:

[0027] H Y = {H1, H2, …, H H}

[0028] In the formula, H H represents the industry label corresponding to the Hth industry;

[0029] Finally, according to the time index in the user load data, combined with the statutory holiday information, the user load data is divided into four time scales of year, month, day and holiday;

[0030] 2.3) The clustering module first calculates the expectation E x , entropy E n and hyper entropy H e of the daily load sequence of the user in the time scale T, T' ∈ {year, month, day, holiday}, generates the cloud drop Cloud(E X , E n , H e ), which is used to calculate the shape similarity between the daily load sequences, as shown in the following formula:

[0031]

[0032]

[0033]

[0034] In the formula, L k and L j represent the kth and jth daily load sequences, respectively, k = 1, 2, …, N, j = 1, 2, …, N, N is the total number of daily load sequences contained in the user in the time scale T', D(L k , L j ) represents the comprehensive similarity between the cloud drops generated by the kth and jth daily load sequences, En k and En j represent the entropy of the cloud drops generated by the kth and jth daily load sequences, respectively, He k and He jS(E) represents the hyperentropy of cloud droplets generated by the k-th and j-th daily load sequences, respectively. xk -E xj |) represents the area of ​​overlap between the cloud droplet expectation curves generated by the k-th and j-th daily load sequences, and S(0) represents the area of ​​overlap when the distance between the cloud droplet expectation curves generated by the k-th and j-th daily load sequences is 0;

[0035] Density peak clustering algorithm is used for clustering. For user daily load sequences within the same time range, the local density and relative distance of each daily load sequence within its respective time range are calculated. The calculation formula is shown below:

[0036]

[0037]

[0038] In the formula, ρ k and δ k Let ρ represent the local density and relative distance of the k-th daily load sequence, respectively. j ρ i Let d represent the local density of the j-th and i-th daily load sequences, respectively. c X(·) represents the cutoff distance, and X(·) represents a piecewise function that takes the value 0 or 1. If the value inside the parentheses is negative, the function takes the value 1; if the value inside the parentheses is positive, the function takes the value 0.

[0039] After obtaining the local density and relative distance of the daily load sequence, the center value and outlier of the daily load sequence are calculated using the following formula:

[0040] λ k =ρ k 2 +δ k 3

[0041]

[0042] In the formula, λ k and ξ k Let these represent the center value and outlier of the k-th daily load sequence, respectively.

[0043] All daily load sequences within time scale T′ are sorted in descending order of outlier degree, and the first q daily load sequences are excluded as outliers. Then, all the remaining daily load sequences are sorted in descending order of their center values, and the first w daily load sequences are taken as the initial cluster centers to create d initial clusters. The remaining daily load sequences are then assigned to the clusters with the smallest relative distance to obtain the initial clustering results.

[0044] The distance between clusters is calculated by the following formula:

[0045]

[0046] wherein D c a b represents the distance between cluster C a and cluster C b , N a C b C represent the total number of daily load sequences in cluster C a and cluster C b , respectively;

[0047] The two clusters with the smallest distance between them are merged, and the distance between the remaining d-1 clusters is recalculated. If all the calculated distances between clusters are greater than the set threshold D c,max , the merging is stopped, and the clustering result is saved. In the k clusters saved finally, the daily load sequence with the smallest distance to other daily load sequences in the cluster is selected and output as the user multi-time scale typical power consumption curve in the time scale T'.

[0048] 2.4) After obtaining the user multi-time scale typical power consumption curve, the user multi-dimensional load characteristic index is calculated to form a user multi-dimensional load characteristic index sequence, including the daily minimum load rate, daily average load rate, daily peak-valley difference rate, valley electricity coefficient, and off-peak power consumption percentage. The definitions and specific calculation formulas of each index are as follows:

[0049] The daily minimum load rate C1 is the ratio of the minimum value of the daily load sequence to the maximum value of the daily load sequence:

[0050]

[0051] The daily average load rate C2 is the ratio of the average value of the daily load sequence to the maximum value of the daily load sequence:

[0052]

[0053] wherein L a represents the average value of the daily load sequence;

[0054] The daily peak-valley difference rate C3 is the ratio of the peak-valley difference of the daily load sequence to the maximum value of the daily load sequence:

[0055]

[0056] The valley electricity coefficient C4 is the ratio of the off-peak power consumption to the total power consumption of the daily load sequence:

[0057] ​​​

[0058] wherein, L G represents the valley period electricity consumption, L T represents the total daily load sequence electricity consumption;

[0059] The flat period electricity consumption percentage C5 is the ratio of the flat period electricity consumption to the total daily load sequence electricity consumption:

[0060]

[0061] wherein, L p represents the flat period electricity consumption.

[0062] Further, the specific operation steps of the step 3) are as follows:

[0063] 3.1) Combine the user load characteristic analysis result with the industry label, calculate the user and industry weighting coefficients according to the imbalance degree between different users and between different industry categories, and save to the compensation module in the three-dimensional portrait library;

[0064] Wherein, the user weighting coefficient calculation formula is as shown in the following formula:

[0065]

[0066] wherein, v Yy represents the user weighting coefficient of the yth user, Y y represents the number of daily load sequences of the yth user, y=1, 2, …, p z , p z represents the total number of users, Y max and Y min respectively represent the maximum value and the minimum value of the number of user daily load sequences;

[0067] The industry weighting coefficient includes an industry curve weighting coefficient and an industry characteristic weighting coefficient, and the calculation formula is as shown in the following formula:

[0068]

[0069]

[0070] wherein, v Ch and v fh respectively represent the industry curve weighting coefficient and the industry characteristic weighting coefficient of the hth industry, p h represents the total number of users of the hth industry, h=1, 2, …, H, H is the number of industry categories contained in the user load data, p max and p min respectively represent the maximum value and the minimum value of the total number of industry users;

[0071] 3.2) Obtain the user multi-time scale typical power consumption curve of each user in the training data set within the time scale T', divide the users with the same industry label into the same range, and perform secondary clustering using the improved density peak clustering algorithm. When calculating the local density and relative distance of the user multi-time scale typical power consumption curve, the density peak clustering algorithm is improved by using the industry curve weighting coefficient, as shown in the following formula:

[0072]

[0073]

[0074] In the formula, ρ y and δ y represent the local density and relative distance of the typical power consumption curve of the yth user, L y and L z represent the typical power consumption curves of the yth user and the zth user, respectively.

[0075] After secondary clustering using the improved density peak clustering algorithm, the industry multi-time scale typical power consumption curve of all industries within the time scale T' is obtained and saved to the curve module in the stereoscopic portrait library. For each industry, the average correlation coefficient of the user multi-time scale typical power consumption curve and the reference sequence is calculated as the industry curve correlation coefficient threshold, using the industry multi-time scale typical power consumption curve within the daily time scale as the reference sequence, as shown in the following formula:

[0076]

[0077] In the formula, ξ ch represents the curve correlation coefficient threshold of the hth industry, x h0 (i) is the load data of the ith sampling time point in the typical power consumption curve of the hth industry, x hy (i) is the load data of the ith sampling time point in the daily load sequence of the yth user in the hth industry, and ρ is the resolution coefficient. and represent the two-level minimum difference and the two-level maximum difference, respectively.

[0078] 3.3) Obtain the multi-dimensional load characteristic index of each user in the training data set, weight using the user weighting coefficient, calculate the multi-dimensional load characteristic index of each industry, and save it to the feature module in the stereoscopic portrait library as the industry multi-dimensional load characteristic index sequence, as shown in the following formula:

[0079]

[0080] In the formula, C hgCg,h represents the gth multi-dimensional load characteristic index of the hth industry, g = 1, 2, 3, 4, 5, respectively corresponding to the five load characteristic indexes of daily minimum load rate, daily average load rate, daily peak-valley difference rate, valley electricity coefficient and off-peak electricity percentage; yg Cg,y represents the gth multi-dimensional load characteristic index of the yth user;

[0081] For each industry, the correlation coefficient mean value of the user multi-dimensional load characteristic index sequence and the reference sequence is calculated as the industry characteristic correlation coefficient threshold value, as shown in the following formula:

[0082]

[0083] In the formula, ξ fh represents the characteristic correlation coefficient threshold value of the hth industry;

[0084] 3.4) Add the industry curve correlation coefficient threshold value and the industry characteristic correlation coefficient threshold value to obtain the industry correlation coefficient threshold value, as shown in the following formula:

[0085] ξ h = ξ ch + ξ fh

[0086] In the formula, ξ h represents the correlation coefficient threshold value of the hth industry;

[0087] 3.5) According to the user load data change in different time scales and the electricity consumption characteristics of different industries, set the curve weight matrix θ C and the feature weight matrix θ f , wherein the curve weight matrix θ C = (θ C1 , θ C2 , θ C3 , θ C4 ) respectively represents the attention weight of the typical electricity consumption curve in the annual, monthly, daily and holiday time scales of each industry; the feature weight matrix θ f = (θ f1 , θ f2 , θ f3 , θ f4 , θ f5 ) respectively represents the attention weight of the five multi-dimensional load characteristic indexes of daily minimum load rate, daily average load rate, daily peak-valley difference rate, valley electricity coefficient and off-peak electricity percentage, and the two weight matrices are saved to the weight module in the stereoscopic image library.

[0088] Further, the specific operation steps of the step 4) are as follows:

[0089] 4.1) Obtain the unknown user daily load sequence L from the test data set N , calculate the multi-dimensional load characteristic indicators of the unknown user, and form a multi-dimensional load characteristic indicator sequence C N , as shown in the following formula:

[0090] C N ={C N1 ,C N2 ,C N3 ,C N4 ,C N5}

[0091] In the formula, C N1 , C N2 , C N3 , C N4 and C N5 represent the daily minimum load rate, daily flat load rate, daily peak-valley difference rate, valley electricity coefficient, and flat period electricity percentage of the unknown user, respectively;

[0092] 4.2) Extract all industry annual, monthly, daily and holiday typical electricity curves from the stereoscopic image library, as shown in the following formula:

[0093] L Yh ={x Yh (1),x Yh (2),…,x Yh (n)}

[0094] L Mh ={x Mh (1),x Mh (2),…,x Mh (n)}

[0095] L Dh ={x Dh (1),x Dh (2),…,x Dh (n)}

[0096] L Jh ={x Jh (1),x Jh (2),…,x Jh (n)}

[0097] In the formula, L Yh , L Mh , L Dh and L Jh represent the annual, monthly, daily and holiday typical electricity curves of the hth industry, and x Yh (n), x Mh (n), x Dh (n) and x Jh(n) represents the load data of the nth sampling time point in the annual, monthly, daily and holiday typical power consumption curve of the hth industry;

[0098] The unknown user daily load sequence L N is taken as the reference sequence, and the curve correlation coefficient ζ Yh (i) and ξ Mh (i) and ξ Dh (i) and ξ Jh (i) between the reference sequence and the annual, monthly, daily and holiday typical power consumption curve of the hth industry at the ith sampling time point is calculated, as shown in the following formula:

[0099]

[0100]

[0101]

[0102]

[0103] In the formula, and respectively represent the two-level minimum difference and the two-level maximum difference, x Yh (i), x Mh (i), x Dh (i) and x Jh (i) represent the load data of the ith sampling time point in the annual, monthly, daily and holiday typical power consumption curve of the hth industry;

[0104] The curve correlation coefficients of n sampling time points are integrated to obtain the curve correlation coefficient R Yh , R Mh , R Dh and R Jh between the reference sequence and the annual, monthly, daily and holiday typical power consumption curve of the hth industry:

[0105]

[0106]

[0107]

[0108]

[0109] The annual, monthly, daily and holiday typical power consumption curves of all industries in the stereoscopic image library are traversed to obtain a 4×H curve correlation coefficient matrix R1:

[0110]

[0111] In the formula, RYH , R MH , R DH , and R JH represent the curve correlation coefficients of the reference sequence and the typical electricity consumption curve of the hth industry on the year, month, day, and holiday, respectively;

[0112] and the curve weight matrix θ C = (θ C1 , θ C2 , θ C3 , θ C4 ), to obtain the 1xH weighted curve correlation coefficient matrix R C of the unknown user and each industry:

[0113] R c = θ c R1

[0114] 4.3) Extract the multi-dimensional load characteristic index sequence of each industry from the stereoscopic image library:

[0115] C h = {C h1 , C h2 , C h3 , C h4 , C h5}

[0116] In the formula, C h represents the multi-dimensional load characteristic index sequence of the hth industry, C h1 , C h2 , C h3 , C h4 , and C h5 are the daily minimum load rate, daily average load rate, daily peak-valley difference rate, valley electricity coefficient, and off-peak period electricity percentage of the hth industry, respectively;

[0117] Take the multi-dimensional load characteristic index sequence of the unknown user as the reference sequence, and calculate the feature correlation coefficient ζ hg of the reference sequence and the multi-dimensional load characteristic index sequence of the hth industry on the gth multi-dimensional load characteristic index, as shown in the following formula:

[0118]

[0119] In the formula, C Ng represents the gth multi-dimensional load characteristic index of the unknown user;

[0120] Iterate through the feature sequences of each industry to obtain the 5xH feature correlation coefficient matrix R2:

[0121]

[0122] In the formula, ξH1 H2 H3 H4 H5 are the characteristic correlation coefficients of the reference sequence and the multi-dimensional load characteristic index sequence of the Hth industry in terms of daily minimum load rate, daily average load rate, daily peak-valley difference rate, valley electricity coefficient, and off-peak period electricity percentage, respectively;

[0123] The characteristic correlation coefficients are weighted using the characteristic weight matrix θ f =(θ f1 ,θ f2 ,θ f3 ,θ f4 ,θ f5 ) to obtain a 1xH weighted characteristic correlation coefficient matrix R f :

[0124] R f =θ f R2

[0125] 4.4) Read the hx1 industry curve weighting coefficient matrix v C and the industry characteristic weighting coefficient matrix v f from the stereoscopic image library, and perform double-channel weighting on R C and R f , as shown in the following formula:

[0126] v C =(v C1 ,v C2 ,…,v CH ) T

[0127] v f =(v f1 ,v f2 ,…,v fH ) T

[0128] R C =v C R c

[0129] R F =v f R f

[0130] In the formula, R C and R F represent the curve correlation coefficient matrix and the characteristic correlation coefficient matrix after double-channel weighting, respectively;

[0131] ​​​​4.5) Using the weighted curve correlation coefficient matrix and the characteristic correlation coefficient matrix of the dual channel, the total correlation coefficient matrix of the unknown user and each industry in the stereoscopic portrait library is calculated, and the index value corresponding to the maximum value is taken as the index value corresponding to the user load type, as shown in the following formula:

[0132] R = argmax(diag(R C + R F ))

[0133] In the formula, R is the index value corresponding to the unknown user load type, and the argmax(·) function represents the index value corresponding to the maximum value in the matrix;

[0134] According to the index value corresponding to the user load type, the corresponding industry label is read from the industry label sequence to obtain the user load type recognition result;

[0135] 4.6) After obtaining the unknown user load type recognition result, the corresponding industry correlation coefficient threshold is obtained from the constructed stereoscopic portrait library, and if the correlation coefficient of the unknown user and the industry is less than the corresponding industry correlation coefficient threshold, it is considered that the unknown user deviates from the normal power consumption behavior habit of the corresponding industry, and there may be abnormal power consumption behavior, and an abnormal alarm signal is output, which is further used to guide the power consumption hidden danger investigation work.

[0136] Compared with the prior art, the present application has the following advantages and beneficial effects:

[0137] 1、The present application uses the three standard deviation principle to remove outliers and fill in missing values when cleaning user load data, significantly reduces the number of days of missing data, reduces the proportion of bad data, and significantly improves the quality of load data.

[0138] 2、The present application uses the comprehensive similarity between cloud droplets generated by user daily load sequences as the standard to represent the correlation between user daily load sequences, and improves the density peak clustering method using a weighting coefficient, which can more efficiently determine the clustering center, thereby accurately extracting user and industry typical power consumption curves and effectively extracting user load features.

[0139] 3、The present application fully considers the imbalance between different users and different industry categories in load data, introduces user and industry weighting coefficients when constructing the stereoscopic portrait library, compensates for industries with fewer samples, effectively increases the attention to easily misidentified industries, and further improves the accuracy of user load type recognition.

[0140] 4、The application adopts a load curve-feature dual-channel weighting recognition strategy, can more comprehensively and finely measure the correlation degree of unknown users and each industry in the stereoscopic portrait library after the load curve-feature dual-channel weighting of each industry in the stereoscopic portrait library, realizes efficient and accurate recognition of user load types and abnormal electricity use behaviors, helps safe electricity use management, and is more suitable for actual engineering. BRIEF DESCRIPTION OF DRAWINGS

[0141] Figure 1 It is a logical flow diagram of the method of the application.

[0142] Figure 2 It is a typical electricity use curve graph of each industry in the method embodiment of the application in the daily time scale. DETAILED DESCRIPTION

[0143] The application will be further described in detail below in combination with embodiments and drawings, but the embodiments of the application are not limited thereto.

[0144] Table 1 below is a user load data structure of the embodiment of the user abnormal load recognition method of the application based on the stereoscopic portrait library and dual-channel weighting:

[0145] Table 1

[0146]

[0147]

[0148] Table 2 below is a user industry label data structure of the embodiment of the user abnormal load recognition method of the application based on the stereoscopic portrait library and dual-channel weighting:

[0149] Table 2

[0150] User number Industry category 1 Guest house 2 Charging pile 3 Catering restaurant 4 Barbecue 5 Residential electricity 6 Bar

[0151] As shown in the application, the embodiment discloses a user abnormal load recognition method based on a stereoscopic portrait library and dual-channel weighting, and the specific circumstances are as follows: Figure 1

[0152] 1) Collect the load data and industry labels of users, perform data cleaning on the load data, including outlier rejection, missing value filling and standardization, the load data after data cleaning constitutes a load data set, and the load data set is divided into a training data set and a test data set according to a 7:3 ratio, and the specific steps are as follows:

[0153] 1.1) Obtain the load data of each user saved in the form of a daily load sequence;

[0154] ​1.2) Traverse each user's daily load sequence one by one, make a missing condition judgment on the user's daily load sequence, if the daily load sequence is continuously missing more than 30% of the sampling data, then discard the daily load sequence; for the remaining user's daily load sequence, calculate the average value and standard deviation of the daily load sequence using the following formula:

[0155]

[0156]

[0157] In the formula, is the average value of the daily load sequence, x(i) is the load data of the i-th sampling time point in the daily load sequence, i = 1, 2, …, n, n is the total number of sampling points of the daily load sequence, n = 1440 in the embodiment, s is the standard deviation of the daily load sequence.

[0158] 1.3) Traverse the load data of each sampling time point in the daily load sequence, if the difference between the load data corresponding to the sampling time point and the average value of the daily load sequence is greater than 3 times the standard deviation of the daily load sequence, then it is considered as abnormal data and is removed, and the load data corresponding to the sampling time point after removal is treated as a missing value.

[0159] 1.4) Traverse the user's daily load sequence, judge whether the load data corresponding to the adjacent sampling time point of the missing value exists, if the load data corresponding to the adjacent sampling time points before and after the missing value are both missing values, then the missing value is assigned as 0; if the sampling time point corresponding to the missing value is the first sampling time point or the last sampling time point in the daily load sequence, then the load data corresponding to its only adjacent sampling time point is used to replace the missing value; if the load data corresponding to its adjacent sampling time points are both normal values, then the average value of the load data corresponding to the adjacent sampling time points before and after it is used to replace the missing value;

[0160] 1.5) Traverse each sampling time point in the user's daily load sequence, map the load data corresponding to it to the interval [0, 1], the calculation formula is as shown in the following formula:

[0161]

[0162] In the formula, L max , L min respectively represent the maximum value and the minimum value of the user's daily load sequence, x i ' is the standardized load data of the i-th sampling time point.

[0163] 2) Construct a user load characteristic analysis module including a statistical module and a clustering module, and statistically analyze the imbalance degree between different users and between different industry categories in the training data set, cluster and extract the user multi-time scale typical power consumption curve, and then calculate the user multi-dimensional load characteristic index by using the user multi-time scale typical power consumption curve to obtain the user load characteristic analysis result. The specific steps are as follows:

[0164] 2.1) Construct a user load characteristic analysis module, which is composed of a statistical module and a clustering module, and is respectively used for statistical analysis and division of load data and clustering to obtain a user multi-time scale typical power consumption curve.

[0165] 2.2) The statistical module first statistically analyzes the imbalance degree between different users and between different industry categories in the training data set. The imbalance degree between different users includes the number of users contained in the user load data and the number of daily load sequences contained in each user. The imbalance degree between different industry categories includes the number of industry categories contained in the user load data and the total number of users of each industry.

[0166] Then, the user industry label data is statistically analyzed to generate a user industry label sequence H Y , as shown in the following formula:

[0167] H Y ={H1,H2,…,H H}

[0168] In the formula, H H represents the industry label corresponding to the Hth industry.

[0169] Finally, according to the time index in the user load data and in combination with the statutory holiday information, the user load data is divided into four time scales of year, month, day and holiday;

[0170] 2.3) The clustering module first calculates the expectation E x , the entropy E n and the hyper entropy H e of the daily load sequence of the user in the time scale T, T' ∈ {year, month, day, holiday}, generates a cloud drop Cloud(E X , E n , H e ), and is used for calculating the shape similarity between the daily load sequences, as shown in the following formula:

[0171]

[0172]

[0173]

[0174]

[0175] wherein L k and L j denote the kth and jth daily load sequence respectively, k = 1, 2, …, N, j = 1, 2, …, N, N is the total number of daily load sequences contained in the time scale T' of the user, D(L k , L j ) denotes the comprehensive similarity between the cloud drops generated by the kth and jth daily load sequences, En k and En j denote the entropy of the cloud drops generated by the kth and jth daily load sequences respectively, He k and He j denote the hyper entropy of the cloud drops generated by the kth and jth daily load sequences respectively, S(E xk -E xj ) denotes the overlapping area between the expected curves of the cloud drops generated by the kth and jth daily load sequences, S(0) denotes the overlapping area when the distance between the expected curves of the cloud drops generated by the kth and jth daily load sequences is 0.

[0176] The density peak clustering algorithm is used for clustering, and for the daily load sequences of the user in the same time range, the local density and relative distance of each daily load sequence in the time range to which it belongs are calculated, and the calculation formula is as follows:

[0177]

[0178]

[0179] wherein ρ k and δ k denote the local density and relative distance of the kth daily load sequence respectively, ρ j , ρ i denote the local density of the jth and ith daily load sequences respectively, d c denotes the cut-off distance, which is taken as 0.8 in the embodiment, and χ(·) denotes a piecewise function with a value of 0 or 1, which takes a value of 1 when the value in the bracket is negative, and takes a value of 0 when the value in the bracket is positive.

[0180] After obtaining the local density and relative distance of the daily load sequence, the center value and outlying degree of the daily load sequence are calculated, and the calculation formula is as follows:

[0181] λ k = ρ k 2 + δ k 3

[0182]

[0183] where λkand ξkrepresent the center value and the outlying degree of the kth daily load sequence, respectively. k and ξ k represent the center value and the outlying degree of the kth daily load sequence, respectively.

[0184] All daily load sequences in the time scale T' are arranged in descending order of the outlying degree, and the first q daily load sequences are excluded as the outlying daily load sequences. Then, all the remaining daily load sequences are arranged in descending order of the center value, and the first w daily load sequences are taken as the initial clustering centers to create d initial clusters. The remaining daily load sequences are sequentially attributed to the cluster to which the clustering center with the minimum relative distance belongs, to obtain the initial clustering result. In the embodiment, the values of q and w are 10 and 3, respectively.

[0185] The inter-cluster distance is calculated by using the following formula by traversing the clusters in the initial clustering result:

[0186]

[0187] where D c (C a ,C b ) represents the distance between the cluster C a and the cluster C b , and represent the total number of daily load sequences in the cluster C a and the cluster C b , respectively.

[0188] The two clusters with the minimum inter-cluster distance are merged, the inter-cluster distance is recalculated by traversing the remaining d-1 clusters, and if all the calculated inter-cluster distances are greater than the set threshold D c,max , the merging is stopped, and the clustering result is saved. In the finally saved k clusters, the daily load sequence with the minimum distance to other daily load sequences in the cluster is selected and output as the user multi-time-scale typical power consumption curve in the time scale T'. In the embodiment, D c,max is 1.4.

[0189] 2.4) After obtaining the user multi-time-scale typical power consumption curve, the user multi-dimensional load characteristic index is calculated to form a user multi-dimensional load characteristic index sequence, including the daily minimum load rate, the daily average load rate, the daily peak-valley difference rate, the valley electricity coefficient, and the off-peak period power consumption percentage. The definitions and specific calculation formulas of the indexes are as follows:

[0190] The daily minimum load rate C1 is the ratio of the minimum value of the daily load sequence to the maximum value of the daily load sequence:

[0191]

[0192] The daily average load rate C2 is the ratio of the average value of the daily load sequence to the maximum value of the daily load sequence:

[0193]

[0194] wherein L a represents the average value of daily load sequence.

[0195] The daily peak-valley difference rate C3 is the ratio of the daily load sequence peak-valley difference to the daily load sequence maximum value:

[0196]

[0197] The valley electricity coefficient C4 is the ratio of the valley period electricity consumption to the total daily load sequence electricity consumption:

[0198]

[0199] wherein L G represents the valley period electricity consumption, L T represents the total daily load sequence electricity consumption.

[0200] The flat period electricity percentage C5 is the ratio of the flat period electricity consumption to the total daily load sequence electricity consumption:

[0201]

[0202] wherein L p represents the flat period electricity consumption.

[0203] 3) The user load characteristic analysis result is combined with the industry label, the user and industry weighting coefficients are calculated according to the imbalance degree between different users and different industry categories, the density peak value clustering algorithm is improved by weighting, the industry multi-time scale typical electricity consumption curve is obtained by using the improved density peak value clustering algorithm for secondary clustering, the multi-dimensional load characteristic index of the industry is calculated, the multi-time scale typical electricity consumption curve (see Figure 2 ) of each industry and the multi-dimensional load characteristic index are used to construct a three-dimensional portrait library, and the specific steps are as follows:

[0204] 3.1) The user load characteristic analysis result is combined with the industry label, the user and industry weighting coefficients are calculated according to the imbalance degree between different users and different industry categories, and are saved to the compensation module in the three-dimensional portrait library;

[0205] The user weighting coefficient calculation formula is as shown in the following formula:

[0206]

[0207] wherein v Yy represents the user weighting coefficient of the yth user, Y y represents the number of daily load sequences of the yth user, y = 1, 2, …, pz , p z represents the total number of users, Y max and Y min represent the maximum and minimum values of the number of user daily load sequences, respectively.

[0208] The industry weighting coefficient includes an industry curve weighting coefficient and an industry feature weighting coefficient, and the calculation formula is as shown in the following formula:

[0209]

[0210]

[0211] In the formula, v Ch and v fh represent the industry curve weighting coefficient and the industry feature weighting coefficient of the hth industry, respectively, p h represents the total number of users of the hth industry, h = 1, 2, …, H, and H is the number of industry types included in the user load data, p max and p min represent the maximum and minimum values of the total number of users of the industry, respectively.

[0212] 3.2) Obtain the user multi-time scale typical power consumption curve of each user in the training data set within the time scale T', divide the users with the same industry label into the same range, and perform secondary clustering using the improved density peak clustering algorithm, wherein when calculating the local density and relative distance of the user multi-time scale typical power consumption curve, the density peak clustering algorithm is improved using the industry curve weighting coefficient, as shown in the following formula:

[0213]

[0214]

[0215] In the formula, p' y and d' y represent the local density and relative distance of the typical power consumption curve of the yth user, L' y and L' z represent the typical power consumption curves of the yth user and the zth user, respectively.

[0216] After secondary clustering using the improved density peak clustering algorithm, the industry multi-time scale typical power consumption curve of all industries within the time scale T' is obtained and saved to the curve module of the stereoscopic image library.

[0217] Figure 2Figures of typical electricity consumption curves of each industry in the multi-time scale in the daily time scale in the embodiment of the method of the present application, wherein the curves with darker colors are the typical electricity consumption curves of each industry in the daily time scale after secondary clustering, and the curve set with lighter colors is the typical electricity consumption curve set of all users of the corresponding industry in the daily time scale.

[0218] For each industry, the average correlation coefficient of the multi-time scale typical electricity consumption curve of the user and the reference sequence of the industry multi-time scale typical electricity consumption curve in the daily time scale is calculated as the industry curve correlation coefficient threshold value, as shown in the following formula:

[0219]

[0220] In the formula, ξ ch represents the curve correlation coefficient threshold value of the hth industry, x h0 (i) is the load data of the ith sampling time point in the typical electricity consumption curve of the hth industry, x hy (i) is the load data of the ith sampling time point in the daily load sequence of the yth user in the hth industry, and ρ is the resolution coefficient. and respectively represent the two-level minimum difference and the two-level maximum difference.

[0221] 3.3) Obtain the multi-dimensional load characteristic indicators of each user in the training data set, weight using the user weighting coefficient, calculate the multi-dimensional load characteristic indicators of each industry, and save the industry multi-dimensional load characteristic indicator sequence to the feature module in the stereoscopic portrait library as the industry multi-dimensional load characteristic indicator sequence, as shown in the following formula:

[0222]

[0223] In the formula, C hg represents the gth multi-dimensional load characteristic indicator of the hth industry, g = 1, 2, 3, 4, 5, respectively corresponding to the daily minimum load rate, the daily average load rate, the daily peak-valley difference rate, the valley electricity coefficient, and the electricity percentage in the flat period; C yg represents the gth multi-dimensional load characteristic indicator of the yth user.

[0224] For each industry, the average correlation coefficient of the multi-dimensional load characteristic indicator sequence of the user and the reference sequence of the industry multi-dimensional load characteristic indicator sequence is calculated as the industry feature correlation coefficient threshold value, as shown in the following formula:

[0225]

[0226] In the formula, ξ fh represents the feature correlation coefficient threshold value of the hth industry.

[0227] 3.4) Add the industry curve correlation coefficient threshold value and the industry characteristic correlation coefficient threshold value to obtain the industry correlation coefficient threshold value, as shown in the following formula:

[0228] ξ h = ξ ch + ξ fh

[0229] In the formula, ξ h represents the correlation coefficient threshold value of the hth industry.

[0230] 3.5) According to the user load data changes in different time scales and the power consumption characteristics of different industries, set the curve weight matrix θ C and the characteristic weight matrix θ f , wherein the curve weight matrix θ C = (θ C1 , θ C2 , θ C3 , θ C4 ) respectively represents the attention weight of the typical power consumption curve in the annual, monthly, daily and holiday time scales of each industry; the characteristic weight matrix θ f = (θ f1 , θ f2 , θ f3 , θ f4 , θ f5 ) respectively represents the attention weight of the five multi-dimensional load characteristic indicators of the daily minimum load rate, the daily average load rate, the daily peak-valley difference rate, the valley power coefficient and the electricity consumption percentage during the flat period, and the two weight matrices are saved to the weight module in the stereoscopic image library. In the embodiment, θ C = (4, 3, 1, 2), θ f = (0.2, 0.3, 0.4, 0.1, 0.1).

[0231] 4) Design a load curve-feature dual-channel weighted recognition strategy. After the load curve-feature dual-channel weighting of each industry in the stereoscopic image library, calculate the total correlation coefficient of the unknown user in the test data set and each industry in the stereoscopic image library to obtain the recognition result of the unknown user load type and abnormal power consumption behavior. The specific steps are as follows:

[0232] 4.1) Obtain the daily load sequence L N of the unknown user from the test data set, calculate the multi-dimensional load characteristic indicators of the unknown user, and form the multi-dimensional load characteristic indicator sequence C N , as shown in the following formula:

[0233] C N = {C N1 , C N2 , C N3 , C N4 , C N5}

[0234] C N1 , C N2 , C N3 , C N4 and C N5 respectively represent the daily minimum load rate, daily flat load rate, daily peak-valley difference rate, valley electricity coefficient, and flat period electricity percentage of the unknown user.

[0235] 4.2) Extract the annual, monthly, daily, and holiday typical electricity curves of all industries from the stereoscopic image library, as shown in the following formula:

[0236] L Yh = {x Yh (1), x Yh (2), …, x Yh (n)}

[0237] L Mh = {x Mh (1), x Mh (2), …, x Mh (n)}

[0238] L Dh = {x Dh (1), x Dh (2), …, x Dh (n)}

[0239] L Jh = {x Jh (1), x Jh (2), …, x Jh (n)}

[0240] In the formula, L Yh , L Mh , L Dh , and L Jh respectively represent the annual, monthly, daily, and holiday typical electricity curves of the hth industry, x Yh (n), x Mh (n), x Dh (n), and x Jh (n) respectively represent the load data at the nth sampling time point in the annual, monthly, daily, and holiday typical electricity curves of the hth industry.

[0241] With the unknown user daily load sequence L N as the reference sequence, the curve correlation coefficients ζ Yh (i), ξ Mh (i), ξ Dh (i), and ξ Jh (i) of the reference sequence and the annual, monthly, daily, and holiday typical electricity curves of the hth industry at the ith sampling time point are calculated.(i), as shown in the following formula:

[0242]

[0243]

[0244]

[0245]

[0246] In the formula, p is the resolution coefficient, and respectively represent two-stage minimum difference and two-stage maximum difference, x Yh (i), x Mh (i), x Dh (i), and x Jh (i) respectively represent the load data of the annual, monthly, daily, and holiday typical power consumption curve of the hth industry at the ith sampling time point.

[0247] The curve correlation coefficients of the reference sequence and the annual, monthly, daily, and holiday typical power consumption curves of the hth industry are obtained by synthesizing the curve correlation coefficients of n sampling time points, and the curve correlation coefficients R Yh , R Mh , R Dh , and R Jh are obtained:

[0248]

[0249]

[0250]

[0251]

[0252] The curve correlation coefficient matrix R1 of 4xH is obtained by traversing the annual, monthly, daily, and holiday typical power consumption curves of all industries in the stereoscopic image library:

[0253]

[0254] In the formula, R YH , R MH , R DH , and R JH respectively represent the curve correlation coefficients of the reference sequence and the annual, monthly, daily, and holiday typical power consumption curves of the Hth industry.

[0255] The curve weight matrix θ C =(θ C1 , θ C2 , θ C3 , θ C4) is multiplied to obtain the 1xH unknown user and each industry weighted curve correlation coefficient matrix R C :

[0256] R c = θ c R1

[0257] 4.3) Extract the multi-dimensional load characteristic index sequence of each industry from the stereoscopic image library:

[0258] C h = {C h1 , C h2 , C h3 , C h4 , C h5}

[0259] In the formula, C h represents the multi-dimensional load characteristic index sequence of the hth industry, C h1 , C h2 , C h3 , C h4 , and C h5 are the daily minimum load rate, the daily average load rate, the daily peak-valley difference rate, the valley electricity coefficient, and the off-peak period electricity percentage of the hth industry, respectively.

[0260] The multi-dimensional load characteristic index sequence of the unknown user is taken as the reference sequence, and the feature correlation coefficient ζ hg of the reference sequence and the multi-dimensional load characteristic index sequence of the hth industry at the gth multi-dimensional load characteristic index is calculated as follows:

[0261]

[0262] In the formula, C Ng represents the gth multi-dimensional load characteristic index of the unknown user.

[0263] The feature sequence of each industry is traversed to obtain the 5xH feature correlation coefficient matrix R2:

[0264]

[0265] In the formula, ξ H1 , ξ H2 , ξ H3 , ξ H4 , and ξ H5 are the feature correlation coefficients of the reference sequence and the multi-dimensional load characteristic index sequence of the hth industry at the daily minimum load rate, the daily average load rate, the daily peak-valley difference rate, the valley electricity coefficient, and the off-peak period electricity percentage, respectively.

[0266] The feature weight matrix θ f = (θ f1 , θf2 ,θ f3 ,θ f4 ,θ f5 The feature correlation coefficients are weighted to obtain a 1×H weighted feature correlation coefficient matrix R. f :

[0267] R f =θ f R2

[0268] 4.4) Read the h×1 industry curve weighting coefficient matrix v from the stereoscopic image database. C Industry-specific weighted coefficient matrix v f , for R C and R f Perform dual-channel weighting as shown in the following formula:

[0269] v C =(v C1 ,v C2 ,…,v CH ) T

[0270] v f =(v f1 ,ν f2 ,…,ν fH ) T

[0271] R C =ν C R c

[0272] R F =ν f R f

[0273] In the formula, R C and R F These represent the curve correlation coefficient matrix and the characteristic correlation coefficient matrix after dual-channel weighting, respectively.

[0274] 4.5) Using the weighted curve correlation coefficient matrix and feature correlation coefficient matrix, calculate the total correlation coefficient matrix between unknown users and various industries in the 3D user profile database, and take the index value corresponding to the maximum value as the index value corresponding to the user load type, as shown in the following formula:

[0275] R = argmax(diag(R) C +R F ))

[0276] In the formula, R is the index value corresponding to the unknown user load type, and the argmax(·) function represents finding the index value corresponding to the maximum value in the matrix.

[0277] According to the index value corresponding to the user load type, the corresponding industry label is read from the industry label sequence to obtain a user load type recognition result.

[0278] 4.6) After obtaining the unknown user load type recognition result, the corresponding industry correlation coefficient threshold is obtained from the constructed stereoscopic portrait library. If the correlation coefficient of the unknown user and the industry is less than the corresponding industry correlation coefficient threshold, it is considered that the unknown user's electricity use behavior deviates from the normal electricity use behavior habit of the corresponding industry, and there may be abnormal electricity use behavior. An abnormal alarm signal is output, which is further used to guide the electricity use hazard investigation work.

[0279] In order to verify the effectiveness of the user abnormal load recognition method based on the stereoscopic portrait library and the double-channel weighting of the application, the load data of 1500 users in a certain tourist area is collected as a load data set for experiment. According to the ratio of 7:3, the load data set is divided into training data set and test data set. The abnormal load recognition is performed on 450 users in the test data set. The calculation formula of the load type recognition accuracy and the abnormal electricity use behavior recognition accuracy is as follows:

[0280]

[0281]

[0282] In the formula, α h and β h respectively represent the industry recognition accuracy and the abnormal recognition accuracy of the hth industry, n h and n Y respectively represent the number of users whose load type recognition and abnormal electricity use behavior recognition are correct in the hth industry, and n z represents the total number of users in the test data set.

[0283] The following Table 3 is the recognition result accuracy rate of the embodiment of the user abnormal load recognition method based on the stereoscopic portrait library and the double-channel weighting of the application:

[0284] Table 3

[0285]

[0286]

[0287] The above embodiments are the preferred embodiments of the application, but the embodiments of the application are not limited by the above embodiments. Any changes, modifications, substitutions, combinations and simplifications made without departing from the spirit and principles of the application shall be equivalent replacement modes, and shall be included in the protection scope of the application.

Claims

1. A method for user abnormal load recognition based on a stereoscopic image library and a two-channel weighting, characterized by, The method comprises the following steps: 1) Collecting load data of users and industry labels, performing data cleaning on the load data, including outlier rejection, missing value filling and standardization, the load data after data cleaning forming a load data set, dividing the load data set into a training data set and a test data set according to a proportion; 2) Building a user load characteristic analysis module comprising a statistical module and a clustering module, statistically analyzing the imbalance degree between different users and between different industry categories in the training data set, clustering and extracting a user multi-time scale typical power consumption curve, and then calculating a user multi-dimensional load characteristic index by using the user multi-time scale typical power consumption curve, to obtain a user load characteristic analysis result; 3) Combining the user load characteristic analysis result with the industry labels, calculating user and industry weighting coefficients according to the imbalance degree between different users and between different industry categories, performing weighted improvement on a density peak value clustering algorithm, and then using the weighted improved density peak value clustering algorithm to perform secondary clustering to obtain an industry multi-time scale typical power consumption curve and calculate an industry multi-dimensional load characteristic index, and using the industry multi-time scale typical power consumption curve and the multi-dimensional load characteristic index to build a three-dimensional portrait library; the specific operation steps are as follows: 3.1) Combining the user load characteristic analysis result with the industry labels, calculating user and industry weighting coefficients according to the imbalance degree between different users and between different industry categories, and saving the user and industry weighting coefficients to a compensation module in the three-dimensional portrait library; 3.2) Obtaining the user multi-time scale typical power consumption curve of each user in a time scale T' in the training data set, dividing users with the same industry labels into the same range, and performing secondary clustering by using the improved density peak value clustering algorithm; 3.3) Obtaining the multi-dimensional load characteristic index of each user in the training data set, weighting by using the user weighting coefficient, calculating the multi-dimensional load characteristic index of each industry, and saving the multi-dimensional load characteristic index of each industry as an industry multi-dimensional load characteristic index sequence to a feature module in the three-dimensional portrait library; 3.4) Adding an industry curve correlation coefficient threshold value to an industry feature correlation coefficient threshold value to obtain an industry correlation coefficient threshold value; 3.5) According to the changes of user load data in different time scales and the power consumption characteristics of different industries, set the curve weight matrix θ C and feature weight matrix θ f , wherein the curve weight matrix θ C =(θ C1 ,θ C2 ,θ C3 ,θ C4 ) respectively represents the attention weight of the typical power consumption curve in the annual, monthly, daily and holiday time scales of each industry; the feature weight matrix θ f =(θ f1 ,θ f2 ,θ f3 ,θ f4 ,θ f5 ) respectively represents the attention weight of the five multi-dimensional load characteristic indicators of the daily minimum load rate, the daily average load rate, the daily peak-valley difference rate, the valley electricity coefficient and the electricity percentage during the flat period, and the two weight matrices are saved to the weight module in the stereoscopic image library; 4) Designing a load curve-feature dual-channel weighted recognition strategy, performing load curve-feature dual-channel weighting on each industry in the three-dimensional portrait library, calculating the total correlation coefficient between an unknown user in a test data set and each industry in the three-dimensional portrait library, and obtaining a recognition result of the load type and abnormal power consumption behavior of the unknown user.

2. The user abnormal load recognition method based on a stereoscopic image library and a dual-channel weighting according to claim 1, characterized by, The specific operation steps of step 1) are as follows: 1.1) Obtaining load data of each user in the form of a daily load sequence; 1.2) Iterating through each daily load sequence of each user, judging the missing condition of the daily load sequence of the user, if more than 30% of the sampling data of the daily load sequence is continuously missing, discarding the daily load sequence; for the remaining daily load sequence of the user, calculating the average value and the standard deviation of the daily load sequence by using the following formula: In the formula, is the average value of the daily load sequence, x(i) is the load data at the i th sampling time point in the daily load sequence, i = 1, 2, …, n, n is the total number of sampling points of the daily load sequence, and s is the standard deviation of the daily load sequence. 1.3) Traverse the load data of each sampling time point in the daily load sequence, if the difference between the load data corresponding to the sampling time point and the average value of the daily load sequence is greater than 3 times the standard deviation of the daily load sequence, it is considered as abnormal data and is excluded, and the load data corresponding to the sampling time point after exclusion is treated as a missing value; 1.4) Traverse the daily load sequence of the user, judge whether the load data corresponding to the adjacent sampling time point of the missing value exists, if the load data corresponding to the adjacent sampling time points before and after the missing value are both missing values, the missing value is assigned as 0; if the sampling time point corresponding to the missing value is the first or last sampling time point in the daily load sequence, the load data corresponding to the only adjacent sampling time point is used to replace the missing value; if the load data corresponding to the adjacent sampling time points are both normal values, the average value of the load data corresponding to the adjacent sampling time points before and after is used to replace the missing value; 1.5) Traverse each sampling time point in the daily load sequence of the user, map the load data corresponding to the sampling time point to the interval [0, 1], and the calculation formula is as follows: where L max , L min denote the maximum and minimum values of the user daily load sequence, respectively, and x i is the normalized load data at the i-th sampling time point. 3.The user abnormal load recognition method based on a stereoscopic image library and a two-channel weighting according to claim 2, characterized in that, The specific operation steps of the step 2) are as follows: 2.1) Construct a user load characteristic analysis module, which is composed of a statistical module and a clustering module, respectively used for statistical and division of load data and clustering to obtain a user multi-time scale typical electricity consumption curve; 2.2) The statistical module first calculates the imbalance degree between different users and between different industry categories in the training data set, the imbalance degree between different users includes the number of users contained in the user load data and the number of daily load sequences contained by each user; the imbalance degree between different industry categories includes the number of industry categories contained in the user load data and the total number of users of each industry; Then, the user industry label data is counted to generate a user industry label sequence H Y As shown in the following formula: H Y = {H1, H2,..., H H} In the formula, H H represents the industry label corresponding to the Hth industry; Finally, according to the time index in the user load data, combined with the legal holiday information, the user load data is divided into four time scales of year, month, day and holiday; 2.3) The clustering module first computes the expectation E x , the entropy E n , and the hyperentropy H e of the daily load sequence of a user in the time scale T, T' e {year, month, day, holiday}, generates a cloud Cloud(E X , E n , H e ) for computing the shape similarity between daily load sequences, as shown in the following equation: wherein L k and L j represent the kth and jth daily load sequence, respectively, k = 1, 2, …, N, j = 1, 2, …, N, N is the total number of daily load sequences contained in the user time scale T', D(L k , L j ) represents the comprehensive similarity between the cloud drops generated by the kth and jth daily load sequences, En k and En j represent the entropy of the cloud drops generated by the kth and jth daily load sequences, respectively, He k and He j represent the hyperentropy of the cloud drops generated by the kth and jth daily load sequences, respectively, S(|E xk -E xj |) represents the overlapping area between the expected curves of the cloud drops generated by the kth and jth daily load sequences, and S(0) represents the overlapping area when the distance between the expected curves of the cloud drops generated by the kth and jth daily load sequences is 0. The density peak clustering algorithm is used for clustering, and for the user daily load sequence in the same time range, the local density and relative distance of each daily load sequence in the time range are calculated, and the calculation formula is as follows: where ρ k and δ k denote the local density and relative distance of the kth daily load sequence, respectively, ρ j and ρ i denote the local density of the jth and ith daily load sequence, respectively, d c denotes the cut-off distance, and χ(·) denotes a piecewise function with value 0 or 1, which takes value 1 if the value in the bracket is negative, and takes value 0 if the value in the bracket is positive. After obtaining the local density and relative distance of the daily load sequence, the center value and the outlying degree of the daily load sequence are calculated, and the calculation formula is as follows: λ k = p k 2 + δ k 3 where λkand ξkrepresent the center value and the outlying degree of the kth daily load sequence, respectively. k and ξ k represent the center value and the outlying degree of the kth daily load sequence, respectively. Arrange all the daily load sequences in the time scale T' in descending order of outlying degree, and exclude the first q daily load sequences as outlying daily load sequences; then arrange all the remaining daily load sequences in descending order of center value, take the first w daily load sequences as the initial clustering center, create d initial clusters, and attribute the remaining daily load sequences to the cluster with the smallest relative distance to the clustering center, to obtain the initial clustering result; Traverse the clusters in the initial clustering result, and calculate the distance between the clusters by the following formula: where D c (C a ,C b ) represents the distance between clusters C a and C b , and represent the total number of daily load sequences within clusters C a and C b , respectively. Merge the two clusters with the minimum inter-cluster distance, traverse the remaining d-1 clusters, and recalculate the inter-cluster distance; if all the calculated inter-cluster distances are greater than the set threshold D c,max , stop merging, and save the clustering result; in the finally saved k clusters, select the daily load sequence with the minimum distance to other daily load sequences in the cluster as the output, as the user multi-time scale typical power consumption curve in the time scale T 2.4) After obtaining the user multi-time scale typical electricity consumption curve, calculate the user multi-dimensional load characteristic index to form a user multi-dimensional load characteristic index sequence, including daily minimum load rate, daily average load rate, daily peak-valley difference rate, valley electricity coefficient and normal period electricity percentage, and the definition and specific calculation formula of each index are as follows: The daily minimum load rate C1 is the ratio of the minimum value of the daily load sequence to the maximum value of the daily load sequence: The daily average load rate C2 is the ratio of the average value of the daily load sequence to the maximum value of the daily load sequence: In the formula, L a denotes the average value of the daily load sequence; The daily peak-valley difference rate C3 is the ratio of the peak-valley difference of the daily load sequence to the maximum value of the daily load sequence: The valley electricity coefficient C4 is the ratio of the electricity consumption in the valley period to the total electricity consumption of the daily load sequence: In the formula, L G represents the low valley period electricity consumption, L T represents the total daily load sequence electricity consumption; The electricity consumption percentage in the flat period C5 is the ratio of the electricity consumption in the flat period to the total electricity consumption of the daily load sequence: In the formula, L p represents the amount of electricity used during the off-peak period.

4. The user abnormal load recognition method based on a stereoscopic image library and two-channel weighting according to claim 3, characterized by, In step 3), the user weighting coefficient calculation formula is as follows: where v Yy represents the user weighting coefficient of the yth user, Y y represents the number of daily load sequences of the yth user, y = 1, 2, …, p z , p z represents the total number of users, Y max and Y min represent the maximum and minimum values of the number of daily load sequences of users, respectively; The industry weighting coefficient includes an industry curve weighting coefficient and an industry characteristic weighting coefficient, and the calculation formula is as follows: wherein v Ch and v fh represent the industry curve weighting factor and the industry characteristic weighting factor of the hth industry, respectively, p h represents the total number of users of the hth industry, h = 1, 2, …, H, H being the number of industry types contained in the user load data, p max and p min represent the maximum and minimum values of the total number of users of the industry, respectively. When calculating the local density and relative distance of the user multi-time scale typical electricity consumption curve, the industry curve weighting coefficient is used to improve the density peak value clustering algorithm, as shown in the following formula: wherein ρ' and δ' y y respectively represent the local density and the relative distance of the typical power consumption curve of the yth user, L' y z respectively represent the typical power consumption curves of the yth user and the zth user.​​ After secondary clustering using the improved density peak value clustering algorithm, the industry multi-time scale typical electricity consumption curve of all industries in the time scale T' is obtained, and the curve module in the stereoscopic portrait library is saved. For each industry, the average correlation coefficient of the user multi-time scale typical electricity consumption curve and the reference sequence is calculated, which is used as the industry curve correlation coefficient threshold, as shown in the following formula: wherein ξ ch represents the curve correlation threshold value of the hth industry, x h0 (i) is the load data of the ith sampling time point in the typical load curve of the hth industry, x hy (i) is the load data of the ith sampling time point in the daily load sequence of the yth user in the hth industry, and ρ is a resolution coefficient, and respectively represent two-level minimum difference and two-level maximum difference; The multi-dimensional load characteristic index of each industry is calculated, as shown in the following formula: In the formula, C hg denotes the gth multi-dimension load characteristic index of the hth industry, g = 1, 2, 3, 4, 5, respectively corresponding to the five load characteristic indexes of daily minimum load rate, daily average load rate, daily peak-valley difference rate, valley electricity coefficient and off-peak period electricity percentage; C yg denotes the gth multi-dimension load characteristic index of the yth user. For each industry, the average correlation coefficient of the user multi-dimensional load characteristic index sequence and the reference sequence is calculated, which is used as the industry characteristic correlation coefficient threshold, as shown in the following formula: wherein ξ fh denotes the characteristic correlation coefficient threshold value for the hth industry; The calculation of the industry correlation coefficient threshold is as follows: ξ h = ξ ch + ξ fh In the formula, ξ h represents the correlation coefficient threshold value of the hth industry. 5.The user abnormal load recognition method based on a stereoscopic image library and two-channel weighting according to claim 4, wherein, The specific operation steps of step 4) are as follows: 4.1) Obtain the unknown user daily load sequence L from the test data set N , calculate the multi-dimensional load characteristic index of the unknown user, form a multi-dimensional load characteristic index sequence C N As shown in the following formula: C N = {C N1 , C N2 , C N3 , C N4 , C N5} In the formula, C N1 , C N2 , C N3 , C N4 , and C N5 respectively represent the daily minimum load rate, daily flat load rate, daily peak-valley difference rate, valley power coefficient, and flat period electricity percentage of the unknown user. 4.2) Extract the annual, monthly, daily, and holiday typical electricity consumption curves of all industries from the stereoscopic portrait library, as shown in the following formula: L Yh = {x Yh (1), x Yh (2),..., x Yh (n)} L Mh = {x Mh (1), x Mh (2),..., x Mh (n)} L Dh = {x Dh (1), x Dh (2),..., x Dh (n)} L Jh = {x Jh (1), x Jh (2),..., x Jh (n)} wherein L Yh , L Mh , L Dh , and L Jh represent the annual, monthly, daily, and holiday typical load curves of the hth industry, respectively, x Yh (n), x Mh (n), x Dh (n), and x Jh (n) represent the load data at the nth sampling time point of the annual, monthly, daily, and holiday typical load curves of the hth industry, respectively. With the unknown user daily load sequence L N As a reference sequence, the curve correlation ζ of the reference sequence and the hth industry annual, monthly, daily and holiday typical power consumption curve at the ith sampling time point is calculated Yh (i), ξ Mh (i), ξ Dh (i) and ξ Jh (i), as shown in the following formula: wherein and respectively represent two-stage minimum difference and two-stage maximum difference, x Yh (i), x Mh (i), x Dh (i) and x Jh (i) respectively represent the load data of the annual, monthly, daily and holiday typical electricity consumption curve of the hth industry at the i th sampling time point. The curve correlation coefficients of the reference sequence and the typical power consumption curves of the hth industry year, month, day and holiday are obtained by integrating the curve correlation coefficients of n sampling time points Yh , R Mh , R Dh and R Jh : Traverse the annual, monthly, daily, and holiday typical electricity consumption curves of all industries in the stereoscopic portrait library to obtain a 4×H curve correlation coefficient matrix R1: wherein R YH , R MH , R DH and R JH represent the curve correlation coefficient of the reference sequence with the typical power consumption curve of the Hth industry on the year, month, day and holiday, respectively; The curve weight matrix θ C = (θ C1 , θ C2 , θ C3 , θ C4 ) is multiplied by the 1 x H unknown user and each industry weighted curve correlation coefficient matrix R C : R c = θ c R1 4.3) Extract the multi-dimensional load characteristic index sequence of each industry from the stereoscopic portrait library: C h = {C h1 , C h2 , C h3 , C h4 , C h5} In the formula, C h represents the multi-dimensional load characteristic index sequence of the hth industry, C h1 , C h2 , C h3 , C h4 and C h5 are the daily minimum load rate, the daily average load rate, the daily peak-valley difference rate, the valley electricity coefficient and the off-peak period electricity percentage of the hth industry, respectively. With the unknown user multi-dimension load characteristic index sequence as a reference sequence, a characteristic correlation coefficient ζ of the reference sequence and the multi-dimension load characteristic index sequence of the hth industry at the gth multi-dimension load characteristic index is calculated hg As shown in the following formula: In the formula, C Ng represents the gth multi-dimension load characteristic index of an unknown user; Traverse the characteristic sequence of each industry to obtain a 5×H characteristic correlation coefficient matrix R2: In the formula, ξ H1 , ξ H2 , ζ H3 , ξ H4 and ζ H5 are the characteristic correlation coefficients of the reference sequence and the multi-dimensional load characteristic index sequence of the Hth industry in terms of daily minimum load rate, daily average load rate, daily peak-valley difference rate, valley electricity coefficient and off-peak period electricity percentage; The feature weight matrix θ is utilized f = (θ f1 , θ f2 , θ f3 , θ f4 , θ f5 ) is used to weight the feature correlation coefficient, and a 1xH weighted feature correlation coefficient matrix R is obtained f : R f = θ f R2 4.4) Read the hxl industry profile weighting coefficient matrix v from the stereoscopic image library C and the industry characteristic weighting coefficient matrix v f , R C and R f are double channel weighted as shown in the following equations: v C = (v C1 , v C2 ,..., v CH ) T v f = (v f1 , v f2 ,..., v fH ) T R C = v C R c R F = v f R f wherein R C and R F respectively represent the double-channel weighted curve correlation matrix and the feature correlation matrix. 4.5) Use the double-channel weighted curve correlation coefficient matrix and the characteristic correlation coefficient matrix to calculate the total correlation coefficient matrix of the unknown user and each industry in the stereoscopic portrait library, and take the index value corresponding to the maximum value as the index value corresponding to the user load type, as shown in the following formula: R = argmax(diag(R C + R F )) In the formula, R is the index value corresponding to the user load type, and the argmax(·) function represents the index value corresponding to the maximum value in the matrix; According to the index value corresponding to the user load type, the corresponding industry label is read from the industry label sequence to obtain the user load type recognition result; 4.6) After obtaining the unknown user load type recognition result, the corresponding industry correlation coefficient threshold is obtained from the constructed stereoscopic portrait library. If the correlation coefficient of the unknown user and the industry is less than the corresponding industry correlation coefficient threshold, it is considered that the unknown user deviates from the normal electricity consumption behavior habit of the corresponding industry, and there may be abnormal electricity consumption behavior. An abnormal alarm signal is output, which is further used to guide the electricity hidden danger investigation work.

Images