CN117786564A - Abnormal electricity consumption intelligent detection method - Google Patents
Abnormal electricity consumption intelligent detection method Download PDFInfo
- Publication number
- CN117786564A CN117786564A CN202311575884.9A CN202311575884A CN117786564A CN 117786564 A CN117786564 A CN 117786564A CN 202311575884 A CN202311575884 A CN 202311575884A CN 117786564 A CN117786564 A CN 117786564A
- Authority
- CN
- China
- Prior art keywords
- user
- electricity consumption
- vector
- network
- behavior
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000005611 electricity Effects 0.000 title claims abstract description 143
- 230000002159 abnormal effect Effects 0.000 title claims abstract description 56
- 238000001514 detection method Methods 0.000 title claims abstract description 52
- 239000013598 vector Substances 0.000 claims abstract description 126
- 238000000034 method Methods 0.000 claims abstract description 24
- 238000010606 normalization Methods 0.000 claims abstract description 11
- 238000013528 artificial neural network Methods 0.000 claims abstract description 9
- 230000008569 process Effects 0.000 claims abstract description 8
- 238000012512 characterization method Methods 0.000 claims description 40
- 239000010410 layer Substances 0.000 claims description 34
- 238000011176 pooling Methods 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 239000002356 single layer Substances 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 5
- 230000004927 fusion Effects 0.000 claims description 4
- 230000009849 deactivation Effects 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 2
- 238000009827 uniform distribution Methods 0.000 claims description 2
- 230000002354 daily effect Effects 0.000 abstract 1
- 230000003203 everyday effect Effects 0.000 abstract 1
- 230000006399 behavior Effects 0.000 description 61
- 230000006870 function Effects 0.000 description 12
- 238000010801 machine learning Methods 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 4
- 230000008034 disappearance Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 230000002829 reductive effect Effects 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000007418 data mining Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 230000002779 inactivation Effects 0.000 description 2
- 206010000117 Abnormal behaviour Diseases 0.000 description 1
- 230000005856 abnormality Effects 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000003116 impacting effect Effects 0.000 description 1
- 230000002401 inhibitory effect Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 230000036961 partial effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000003313 weakening effect Effects 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention belongs to the technical field of abnormal electricity consumption detection, and particularly relates to an abnormal electricity consumption intelligent detection method, which comprises the following steps: acquiring power grid user data to be detected; inputting into a abnormal electricity detection model; and obtaining an abnormal electricity utilization detection result of the power grid user. The training process of the abnormal electricity utilization detection model mainly comprises the following steps: obtaining a model training sample; extracting a daily electricity consumption behavior representation vector sequence of a user, extracting an electricity consumption behavior representation vector of the user, and obtaining user category characteristics through clustering; splicing the user electricity consumption behavior representation vector and the continuous characteristic vector, inputting the spliced user electricity consumption behavior representation vector and the continuous characteristic vector into a characteristic crossing network, and inputting the user electricity consumption behavior representation vector sequence every day into a condition level normalization network; performing iterative training through a fully connected neural network; and (5) converging the loss function, and ending the training to obtain the trained abnormal electricity utilization detection model. The method and the device effectively improve the accuracy of abnormal electricity utilization detection by referring to the diversity of users and the periodical influence of electricity utilization habits of the users.
Description
Technical Field
The invention belongs to the technical field of abnormal electricity utilization detection, and particularly relates to an abnormal electricity utilization intelligent detection method.
Background
With the continuous development of power systems, the number of power consumers is also increasing. However, for various reasons, some users may have abnormal electricity consumption behaviors, such as electricity stealing, overload electricity consumption, and the like, and these abnormal electricity consumption behaviors not only affect the normal operation of the power system, but also bring great economic loss to the power company, and also bring unnecessary loss to other users. Therefore, it is important to increase the power management level of the electric company and increase the striking strength of the electricity stealing behavior, and to implement the series of measures, it is first of all to detect and identify the abnormal electricity using behavior of the electric power user, so it is more and more important to detect the abnormal electricity using behavior of the electric power user by using big data.
For abnormal electricity behavior detection tasks of power consumers, common methods include machine learning and data mining based anomaly detection techniques, expert system based anomaly detection techniques, and deep learning and clustering algorithm based anomaly detection techniques. The anomaly detection technology based on machine learning and data mining often depends on a specific environment, a set of system cannot be adaptively applied to power grid anomaly detection in all areas, and meanwhile, due to the fact that the boundaries of normal behaviors and abnormal behaviors are quite fuzzy, a high false alarm rate can be generated by applying a machine learning method; the abnormality detection technology based on the expert system often requires an experienced expert to construct and maintain a feature library, which is different in electricity utilization habit and environmental change of each region for a nationwide power grid system, and has high experience requirements for the expert; based on the anomaly detection technology of the deep learning and clustering algorithm, the internal relation between the data can be effectively searched through the support of big data and the fitting capacity of the deep learning on nonlinearity, and the optimal characteristics do not need to be manually searched.
However, the conventional deep learning-based method does not consider, even completely ignoring, the diversity of users and the periodicity of the electricity consumption habits of the users, and often directly inputs the time sequence data of the electricity consumption of the users into the deep time sequence network to detect abnormal users, which results in poor detection accuracy. In view of the foregoing, there is a need for an abnormal electricity consumption detection method that can sufficiently consider the characteristics of electricity consumption behavior such as the diversity of users and the periodicity of electricity consumption habits of users, so as to improve the accuracy of abnormal electricity consumption detection.
Disclosure of Invention
Aiming at the defects existing in the prior art, the invention provides an abnormal electricity utilization intelligent detection method.
An abnormal electricity consumption intelligent detection method comprises the following steps of;
acquiring power grid user data to be detected;
and inputting the power grid user data to be detected into the trained abnormal electricity utilization detection model, and outputting an abnormal electricity utilization detection result of the power grid user.
The training process of the abnormal electricity utilization detection model comprises the following steps:
s1: obtaining a model training sample, wherein the model training sample is a labeled power grid user data set, and the power grid user data set comprises power consumption recording time, daily power consumption value, week, user resident population and house floor area;
s2: extracting a power consumption behavior characterization vector sequence of each user and a power consumption behavior characterization vector of the user from a power grid user data set;
s3: clustering the user electricity consumption behavior characterization vectors by adopting a clustering algorithm to obtain user category characteristics;
s4: converting discrete data into corresponding continuous feature vectors by adopting an embedded model, wherein the discrete data comprises the user category features, weeks, user resident population and house floor area;
s5: splicing the user electricity consumption behavior characterization vector and the continuous characteristic vector, and inputting the spliced user electricity consumption behavior characterization vector and the continuous characteristic vector into a multi-layer characteristic crossover network to obtain a characteristic crossover vector of the user electricity consumption behavior;
s6: inputting a daily electricity quantity behavior representation vector sequence of a user into a condition level normalization network and a maximum pooling network to obtain a user related behavior representation vector;
s7: and splicing the characteristic cross vector and the user related behavior characterization vector, inputting the characteristic cross vector and the user related behavior characterization vector into a fully-connected neural network for two-classification, obtaining a classification result, calculating model classification loss according to the classification result, adjusting model parameters according to the loss, performing iterative training, ending training when a loss function converges, and obtaining the trained abnormal electricity utilization detection model.
The beneficial effects of the invention are as follows: according to the invention, the influence of electricity consumption caused by resident population, house occupation area and the like can be weakened by a specific clustering algorithm, and the class dug after the influence is weakened is the class which truly classifies the electricity consumption of the user, so that the user class characteristics with more analysis and utilization values are obtained; the problem that certain cross features can be ignored when the traditional machine learning model extracts the features manually is solved by adopting the feature cross network, the continuous features and the category type features are fully interacted through the feature cross network, the false alarm rate of the model is reduced, and the detection accuracy is improved; the condition level normalization network is adopted, and the periodicity and diversity of different users in electricity behavior habit are fully excavated by interacting time sequence type features on each time step, so that the most effective feature vector is provided for the prediction of a final model, and the detection accuracy is improved. The method provided by the invention fully considers the characteristic influence of the electricity consumption behavior such as the diversity of users and the periodicity of electricity consumption habits of users, obviously improves the detection accuracy of abnormal electricity consumption, and has the advantages of high precision, high timeliness, high reliability, high automation degree and the like.
Drawings
FIG. 1 is a flow chart of the steps of the abnormal electrical detection model training process of the present invention;
FIG. 2 is a schematic diagram of a model structure and a data processing flow of an abnormal electricity utilization detection model according to the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The embodiment of the invention provides an abnormal electricity consumption intelligent detection method, which comprises the following steps:
acquiring power grid user data to be detected;
and inputting the power grid user data to be detected into the trained abnormal electricity utilization detection model, and outputting an abnormal electricity utilization detection result of the power grid user.
As shown in fig. 1, the training process of the abnormal electricity consumption detection model includes:
s1: a model training sample is obtained, wherein the model training sample is a labeled power grid user data set, and the power grid user data set comprises power consumption recording time, daily power consumption value, week, user resident population and house floor area. The power grid users are real-name registered users, and information such as resident population numbers of the users and occupied area of houses of the users can be investigated during registration. Along with the popularization of the intelligent electric meter, the information of the intelligent electric meter can be fed back in time through a network, and the information such as the electricity consumption recording time, the daily electricity consumption value, the week, the real-time electricity consumption value, the accumulated electricity consumption value and the like of the power grid user can be obtained in real time.
Preferably, the power grid user data set is manually marked, mark 1 indicates that the user is a user with abnormal electricity, mark 0 indicates that the user is a user with normal electricity, and the power grid user data set with the label is obtained.
S2: and extracting a user daily electricity consumption behavior characterization vector sequence and a user electricity consumption behavior characterization vector from the power grid user data set.
Preferably, the step S2 specifically includes the following steps:
s21: and acquiring a daily electricity consumption value in the power grid user data set, constructing a daily electricity consumption representation vector with the length of 24 by taking an hour as a unit, and normalizing the daily electricity consumption representation vector to obtain a normalized daily electricity consumption representation vector. The daily electricity consumption value is a data set obtained by processing the historical electricity consumption value of the power grid user. The influence of the huge difference of the night electricity consumption and the daytime electricity consumption on the numerical value on the model is reduced by standardization. The standardized formula is as follows:
wherein N is v Representing normalized feature vectors, v i Representing the value of the i-th dimension in the vector.
S22: and splicing the standardized daily electricity quantity representation vectors of 7 adjacent days in the time dimension, inputting an LSTM model, outputting a hidden layer obtained through the LSTM model as a daily electricity quantity behavior representation vector sequence of a user, and finally outputting a final output obtained through the LSTM model as a user electricity quantity behavior representation vector.
The LSTM model has the advantages of capturing long-term dependence of time sequence data, inhibiting gradient disappearance and the like, the hidden layer output obtained through the LSTM model is used as a user daily electric quantity behavior representation vector sequence, and each vector in the user daily electric quantity behavior representation vector sequence at the moment contains related information of each daily electric quantity representation vector before the vector; the final output obtained through the LSTM model is used as a user electricity consumption behavior characterization vector, and the vector has the relevant information of all the electricity consumption characterization vectors of 7 adjacent days (namely one week), so that the electricity consumption condition of the following day can be predicted conveniently.
The dimension of the user per day power behavior characterization vector sequence is (batch_size, 7, inner_dim), and the dimension of the user power behavior characterization vector is (batch_size, inner_dim).
S3: and clustering the user electricity consumption behavior characterization vector by adopting a clustering algorithm to obtain user category characteristics.
In the existing abnormal electricity utilization detection technology, abnormal electricity utilization is often monitored only by referring to conventional electricity indexes such as electricity utilization quantity, voltage, current and the like. But the power grid users have diversity and different electricity utilization habits. The invention fully refers to the influence factors such as the electricity consumption value of the power grid user, the resident population number of the user, the occupied area of the house and the like to form a data set, wherein the resident population number and the occupied area generally have great influence on the electricity consumption, the data are processed, and the class mined after weakening the influence is the class really classifying the electricity consumption behavior of the user through a specific clustering algorithm, so that the class characteristics of the user with more analysis and utilization values are obtained.
Preferably, the power consumption behavior characterization vector of the user is input into a focal-means clustering algorithm model, the number of clustering centers is determined through an elbow method, the number of the clustering centers is used as the classification number of the user categories, the distance between each user and the clustering centers is calculated through a distance formula of the focal-means clustering algorithm, and the category corresponding to the clustering centers is determined according to the distance to be used as the category characteristic of the user. In general, the distance between each user and the cluster center is calculated by a distance formula, and the category corresponding to the cluster center with the smallest distance is used as the category characteristic of the user.
The distance calculation formula of the focal-means clustering algorithm is as follows:
wherein x is i Representing the ith dimension, c, in the x user power usage behavior characterization vector i Represents the ith dimension in the cluster center token vector c,representing the average number of household population of all users belonging to the c cluster center, initializing the average number of household population of all users, μ representing the number of household population of x users, A a Representing the house floor space of x users +.>And (3) representing the average value of the occupied areas of the houses of all the users belonging to the c cluster center, wherein the average value of the occupied areas of the houses of all the users is used in the initialization process. Through the specific clustering algorithm, the influence of electricity consumption caused by resident population, house occupation area and the like can be weakened when the clustering is carried out, so that some characteristics of the user on electricity consumption behaviors are deeply mined, and the clustering is carried out according to the characteristics.
The specific method for determining the number of the clustering centers by the elbow method comprises the following steps: the number of clusters is enumerated from small to large, meanwhile, the sum of squares of errors under the current number of clusters is calculated, the number of clusters is taken as an abscissa, the sum of squares of errors corresponding to the number of the current clusters is taken as an ordinate, the slope between each point and the previous point is calculated, when the change amount of the slope is maximum, the corresponding number of clusters is the number of clusters needed by people, and the formula of the sum of squares of errors is:
wherein SSE represents the sum of squares of errors, K represents the number of clusters, C i The i-th cluster center vector is represented, and x represents the user electricity behavior characterization vector.
S4: an embedded model is used to convert discrete data, including user category characteristics, week, user resident population, and house floor area, to corresponding continuous feature vectors.
Preferably, in step S4, the embedded model is formed by two fully connected layers, and the initialization weight parameters of the fully connected neural network satisfy the uniform distribution of xavier.
Specifically, the embedded model is composed of two full-connection layers and is connected through a residual error network, the input characteristic dimension of the first full-connection network is the sum of products of each discrete data and the corresponding class number, and the output dimension is inner_dim//3; the second fully connected neural network has an input characteristic dimension of inner_dim//3 and an output dimension of inner_dim.
S5: and (3) splicing the user electricity consumption behavior characterization vector obtained in the step (S2) with the continuous feature vector obtained in the step (S4) and inputting the spliced continuous feature vector into a multi-layer feature crossover network to obtain the feature crossover vector of the user electricity consumption behavior.
Preferably, step S5 includes:
and S51, splicing the user electricity consumption behavior characterization vector obtained in the step S2 and the continuous feature vector obtained in the step S4 in the last dimension to obtain a fusion feature vector.
S52, inputting the fusion feature vector into a multi-layer feature crossover network to obtain the feature crossover vector of the electricity consumption behavior of the user.
Preferably, each layer of the multi-layer feature cross network adopts a single-layer feature cross network, the single-layer feature cross network comprises Elu activation functions, and the specific formula of the single-layer feature cross network is as follows:
wherein v is l+1 Representing a (l+1) -th layer feature cross vector, v, in a multi-layer feature cross network l Representing layer I feature cross vector, v 0 Representation ofFusing feature vectors, w l The weight parameter at the first layer is represented, D (x) represents the random deactivation function, and a (x) represents the Elu activation function.
The existing feature crossover network is not added with a random deactivation function and an activation function, and the model is subjected to the phenomenon of over-fitting along with the increase of the number of layers of the feature crossover network. Because not every feature intersection is effective, the ineffective intersection features will become noise features in the model, negatively impacting the model performance to some extent. By adding a random inactivation function, the feature crossover network carries out random inactivation on the crossed features in each layer, so that excessive occurrence of noise features can be relieved to a certain extent, and generalization capability of the model is improved; meanwhile, the existing characteristic crossover network alleviates the problem of gradient disappearance to a certain extent by using a residual connection mode. However, the whole network model is also a linear model, and functions are activated by adding Elu, so that the model has partial nonlinear capability on one hand; on the other hand, the Elu activation function has certain advantages in solving the gradient disappearance problem, namely, the problem that the multilayer characteristic crossover network needs to face gradient disappearance after the network layer number is gradually increased is indirectly weakened.
S6: and (2) inputting the daily electricity quantity behavior representation vector sequence of the user obtained in the step (S2) into a condition level normalization network and a maximum pooling network to obtain a user related behavior representation vector.
Preferably, step S6 comprises the steps of:
and S61, inputting the daily electric quantity behavior characterization vector sequence of the user obtained in the step S2 into a condition level normalization network to obtain a correlation behavior feature matrix.
The specific formula of the condition level normalization network is as follows:
wherein V is ij A vector representing the (i, j) th position in the correlation matrix, S ij Representing a user per day power behavior characterization vector sequenceAfter expansion in the second dimension (dimension [ batch_size,7, inner_dim]) Vectors of (i, j) th position, h j Represents the j-th vector in the vector sequence of the per-daily-electricity-consumption behavior representation of the user, and mu represents h j Vector-corresponding mean value, σ represents h j The standard deviation corresponding to the vector is used,representing a stitching operation. For h j Dividing by a certain weight to prevent h j The larger condition of each element in the numerical value is unfavorable for gradient update, and V obtained by the splicing operation ij Information characterization between the current i and j days is obtained simultaneously.
S62, inputting the correlation behavior feature matrix into a two-dimensional maximum pooling network with a convolution kernel of 7 to obtain a user correlation behavior characterization vector.
S7: splicing the characteristic cross vector obtained in the step S5 and the user-related behavior characterization vector obtained in the step S6, and inputting the characteristic cross vector and the user-related behavior characterization vector into a fully-connected neural network for two classification to obtain a classification result; and calculating model classification loss according to the classification result, adjusting model parameters according to the loss, converging a loss function, and finishing training to obtain a trained abnormal electricity utilization detection model.
Calculating the total loss of the model, namely the full connection classification loss:
where loss represents the full connection class loss, n represents the number of users, y i A real tag indicating whether the i-th user is an abnormal user,the probability prediction value of the expression model for the ith user is an abnormal user or not, eta represents regularization parameters, l represents the layer number of the multi-layer cross network, and the model is easier to be over-fitted along with the increase of the layer number of the cross network, so that the regularization term along with the crossIncreasing the number of network layers increases, θ j Representing parameters of the j-th layer in the multi-layer crossover network.
By adopting the loss function, the model classification loss is calculated, and the model parameters are adjusted according to the loss, so that the overfitting of the characteristic crossover network caused by a plurality of network layers can be effectively prevented, and the detection accuracy of the model is further improved.
As shown in fig. 2, the model structure of the abnormal electricity detection model in the embodiment of the present application mainly includes a multi-layer feature cross network, a condition-level normalization network, a max-pooling network, and a fully-connected neural network. In the abnormal electricity consumption detection method provided by the embodiment of the application, in the training process of the abnormal electricity consumption detection model, the flow of data processing is specifically as follows:
and (2) manually labeling the power grid user data set through the step (S1) to obtain the power grid user data set with the label.
Through the step S2, the daily electricity consumption behavior representation vector sequence of the user and the user electricity consumption behavior representation vector are extracted through the LSTM model according to the standardized processing and the like of the user data set of the power grid.
Through the step S3, clustering is carried out on the user electricity consumption behavior characterization vector by adopting a clustering algorithm, and user category characteristics are obtained.
Through the step S4, the discrete data such as the user category characteristics, the week, the resident population of the user, the occupied area of the house and the like are converted into the corresponding continuous characteristic vectors by adopting the embedded model.
And (5) splicing the user electricity consumption behavior characterization vector and the continuous feature vector and inputting the spliced user electricity consumption behavior characterization vector and the continuous feature vector into a multi-layer feature crossover network to obtain the feature crossover vector of the user electricity consumption behavior.
Through the step S6, the user daily electricity quantity behavior representation vector sequence is input into the condition level normalization network and the maximum pooling network to obtain the user related behavior representation vector.
And S7, splicing the characteristic cross vectors and the user related behavior characterization vectors, inputting the characteristic cross vectors and the user related behavior characterization vectors into a fully-connected neural network for two-classification, obtaining a classification result, calculating model classification loss according to the classification result, adjusting model parameters according to the loss, converging a loss function, and finishing training to obtain the trained abnormal electricity utilization detection model.
The beneficial effects of the invention are as follows: according to the invention, the influence of electricity consumption caused by resident population, house occupation area and the like can be weakened by a specific clustering algorithm, and the class dug after the influence is weakened is the class which truly classifies the electricity consumption of the user, so that the user class characteristics with more analysis and utilization values are obtained; the problem that certain cross features can be ignored when the traditional machine learning model extracts the features manually is solved by adopting the feature cross network, the continuous features and the category type features are fully interacted through the feature cross network, the false alarm rate of the model is reduced, and the detection accuracy is improved; the condition level normalization network is adopted, and the periodicity and diversity of different users in electricity behavior habit are fully excavated by interacting time sequence type features on each time step, so that the most effective feature vector is provided for the prediction of a final model, and the detection accuracy is improved. The method provided by the invention fully considers the characteristic influence of the electricity consumption behavior such as the diversity of users and the periodicity of electricity consumption habits of users, obviously improves the detection accuracy of abnormal electricity consumption, and has the advantages of high precision, high timeliness, high reliability, high automation degree and the like.
While the foregoing is directed to embodiments, aspects and advantages of the present invention, other and further details of the invention may be had by the foregoing description, it will be understood that the foregoing embodiments are merely exemplary of the invention, and that any changes, substitutions, alterations, etc. which may be made herein without departing from the spirit and principles of the invention.
Claims (8)
1. The intelligent detection method for abnormal electricity utilization is characterized by comprising the following steps of;
acquiring power grid user data to be detected;
inputting the power grid user data to be detected into a trained abnormal electricity utilization detection model, and outputting abnormal electricity utilization detection results of power grid users;
the training process of the abnormal electricity utilization detection model comprises the following steps:
s1: obtaining a model training sample, wherein the model training sample is a labeled power grid user data set, and the power grid user data set comprises power consumption recording time, daily power consumption value, week, user resident population and house occupation area;
s2: extracting a daily electricity consumption behavior representation vector sequence of a user and a user electricity consumption behavior representation vector from the power grid user data set;
s3: clustering the user electricity consumption behavior characterization vector by adopting a clustering algorithm to obtain user category characteristics;
s4: converting discrete data into corresponding continuous feature vectors by adopting an embedded model, wherein the discrete data comprises the user category features, weeks, user resident population and house floor area;
s5: splicing the user electricity consumption behavior characterization vector and the continuous feature vector, and inputting the spliced user electricity consumption behavior characterization vector and the continuous feature vector into a multi-layer feature crossover network to obtain a feature crossover vector of the user electricity consumption behavior;
s6: inputting the daily electricity quantity behavior representation vector sequence of the user into a condition level normalization network and a maximum pooling network to obtain a user related behavior representation vector;
s7: and splicing the characteristic cross vector and the user related behavior characterization vector, inputting the characteristic cross vector and the user related behavior characterization vector into a fully-connected neural network for two-classification, obtaining a classification result, calculating model classification loss according to the classification result, adjusting model parameters according to the loss, performing iterative training, ending training when a loss function converges, and obtaining a trained abnormal electricity utilization detection model.
2. The method for intelligently detecting abnormal electricity consumption according to claim 1, wherein in step S1, the grid user data set is manually marked, mark 1 indicates that the user is a user with abnormal electricity consumption, mark 0 indicates that the user is a user with normal electricity consumption, and a grid user data set with a label is obtained.
3. The method for intelligently detecting abnormal electricity consumption according to claim 1, wherein the step S2 comprises the steps of:
s21: acquiring daily electricity consumption values in the power grid user data set, constructing a daily electricity consumption representation vector with the length of 24 by taking an hour as a unit, and normalizing the daily electricity consumption representation vector to obtain a normalized daily electricity consumption representation vector;
s22: splicing the standardized daily electricity quantity representation vectors of 7 adjacent days in the time dimension, inputting an LSTM model, outputting a hidden layer obtained through the LSTM model as a daily electricity quantity behavior representation vector sequence of a user, and finally outputting a final output obtained through the LSTM model as a user electricity quantity behavior representation vector.
4. The method for intelligently detecting abnormal electricity consumption according to claim 1 or 3, wherein the step S3 is specifically: inputting the electricity consumption behavior characterization vector of the user into a focal-means clustering algorithm model, determining the number of clustering centers through an elbow method, taking the number of the clustering centers as the classification number of user categories, calculating the distance between each user and the clustering centers through a distance formula of the focal-means clustering algorithm, and determining the category corresponding to the clustering centers as the category characteristics of the user according to the distance;
the distance calculation formula of the focal-means clustering algorithm is as follows:
wherein x is i Representing the ith dimension, c, in the x user power usage behavior characterization vector i Represents the ith dimension in the cluster center token vector c,representing the average number of household population of all users belonging to the c cluster center, initializing the average number of household population of all users, μ representing the number of household population of x users, A a Representing the house floor space of x users +.>And (3) representing the average value of the occupied areas of the houses of all the users belonging to the c cluster center, wherein the average value of the occupied areas of the houses of all the users is used in the initialization process.
5. The method for intelligently detecting abnormal electricity consumption according to claim 1, wherein in the step S4, the embedded model is formed by two fully-connected layers, and the initializing weight parameters of the fully-connected neural network satisfy xavier uniform distribution.
6. The method for intelligently detecting abnormal electricity consumption according to claim 1, wherein the step S5 includes:
s51, splicing the user electricity consumption behavior characterization vector and the corresponding continuous feature vector to obtain a fusion feature vector;
s52, inputting the fusion feature vector into a multi-layer feature crossover network to obtain a feature crossover vector of the electricity consumption behavior of the user.
7. The method for intelligently detecting abnormal electricity consumption according to claim 1 or 6, wherein in the step S5, each layer of the multi-layer feature cross network adopts a single-layer feature cross network, the single-layer feature cross network comprises a Elu activation function, and the specific formula of the single-layer feature cross network is as follows:
wherein v is l+1 Representing multi-layer feature interactionsLayer (l+1) feature cross vector, v in a fork network l Representing layer I feature cross vector, v 0 Representing the fused feature vector, w l The weight parameter at the first layer is represented, D (x) represents the random deactivation function, and a (x) represents the Elu activation function.
8. The method for intelligently detecting abnormal electricity consumption according to claim 1, wherein the step S6 includes the steps of:
s61, inputting a daily electricity behavior characterization vector sequence of a user into a condition level normalization network to obtain a correlation behavior feature matrix;
s62, inputting the correlation behavior feature matrix into a two-dimensional maximum pooling network with a convolution kernel of 7 to obtain a user correlation behavior characterization vector.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311575884.9A CN117786564A (en) | 2023-11-23 | 2023-11-23 | Abnormal electricity consumption intelligent detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311575884.9A CN117786564A (en) | 2023-11-23 | 2023-11-23 | Abnormal electricity consumption intelligent detection method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117786564A true CN117786564A (en) | 2024-03-29 |
Family
ID=90391780
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311575884.9A Pending CN117786564A (en) | 2023-11-23 | 2023-11-23 | Abnormal electricity consumption intelligent detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117786564A (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111209168A (en) * | 2020-01-14 | 2020-05-29 | 中国人民解放军陆军炮兵防空兵学院郑州校区 | Log sequence anomaly detection framework based on nLSTM-self attention |
| CN113010504A (en) * | 2021-03-04 | 2021-06-22 | 江苏方天电力技术有限公司 | Electric power data anomaly detection method and system based on LSTM and improved K-means algorithm |
| US20230063489A1 (en) * | 2021-08-25 | 2023-03-02 | Bank Of America Corporation | Malware Detection with Multi-Level, Ensemble Artificial Intelligence Using Bidirectional Long Short-Term Memory Recurrent Neural Networks and Natural Language Processing |
| US20230252478A1 (en) * | 2022-02-08 | 2023-08-10 | Paypal, Inc. | Clustering data vectors based on deep neural network embeddings |
| CN116861306A (en) * | 2023-06-20 | 2023-10-10 | 华中科技大学 | Abnormal electricity utilization detection method based on electricity utilization trend quantification network |
-
2023
- 2023-11-23 CN CN202311575884.9A patent/CN117786564A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111209168A (en) * | 2020-01-14 | 2020-05-29 | 中国人民解放军陆军炮兵防空兵学院郑州校区 | Log sequence anomaly detection framework based on nLSTM-self attention |
| CN113010504A (en) * | 2021-03-04 | 2021-06-22 | 江苏方天电力技术有限公司 | Electric power data anomaly detection method and system based on LSTM and improved K-means algorithm |
| US20230063489A1 (en) * | 2021-08-25 | 2023-03-02 | Bank Of America Corporation | Malware Detection with Multi-Level, Ensemble Artificial Intelligence Using Bidirectional Long Short-Term Memory Recurrent Neural Networks and Natural Language Processing |
| US20230252478A1 (en) * | 2022-02-08 | 2023-08-10 | Paypal, Inc. | Clustering data vectors based on deep neural network embeddings |
| CN116861306A (en) * | 2023-06-20 | 2023-10-10 | 华中科技大学 | Abnormal electricity utilization detection method based on electricity utilization trend quantification network |
Non-Patent Citations (2)
| Title |
|---|
| 贾志强 等: ""基于集成学习算法的消费行为预测"", 《计算机技术与发展》, 10 May 2022 (2022-05-10) * |
| 陈晓斌: ""基于数据挖掘的反窃电预警方法研究"", 《中国优秀硕士学位论文全文数据库》, 15 March 2023 (2023-03-15) * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Li et al. | Electricity theft detection in power grids with deep learning and random forests | |
| CN110223196B (en) | Anti-electricity-stealing analysis method based on typical industry feature library and anti-electricity-stealing sample library | |
| CN113962364B (en) | Multi-factor power load prediction method based on deep learning | |
| CN108133225A (en) | A kind of icing flashover fault early warning method based on support vector machines | |
| Cao et al. | CNN-LSTM coupled model for prediction of waterworks operation data | |
| Wu et al. | AdaBoost-SVM for electrical theft detection and GRNN for stealing time periods identification | |
| CN117495109B (en) | Power stealing user identification system based on neural network | |
| CN115186012A (en) | Power consumption data detection method, device, equipment and storage medium | |
| Kumar et al. | Cloud-based electricity consumption analysis using neural network | |
| CN112529638B (en) | Service demand dynamic prediction method and system based on user classification and deep learning | |
| Asghari et al. | Spatial rainfall prediction using optimal features selection approaches | |
| CN116226748A (en) | Multi-label co-occurrence network discrimination-based electricity stealing type detection method and system | |
| Rodrigues et al. | A deep learning approach for the detection and classification of power quality disturbances with windowed signals | |
| CN113935413A (en) | Distribution network wave recording file waveform identification method based on convolutional neural network | |
| CN117786564A (en) | Abnormal electricity consumption intelligent detection method | |
| CN117892122A (en) | Training method of breaker fault detection model, fault detection method and device | |
| Wang et al. | A Novel Multi‐Input AlexNet Prediction Model for Oil and Gas Production | |
| CN117933531A (en) | Distributed photovoltaic power generation power prediction system and method | |
| Wang et al. | An improved deep learning algorithm in enabling load data classification for power system | |
| CN116304941A (en) | Ocean data quality control method and device based on multi-model combination | |
| CN115147242A (en) | Power grid data management system based on data mining | |
| CN115081551A (en) | RVM line loss model building method and system based on K-Means clustering and optimization | |
| CN114818849A (en) | Convolution neural network based on big data information and anti-electricity-stealing method based on genetic algorithm | |
| Chen et al. | LPR‐MLP: A Novel Health Prediction Model for Transmission Lines in Grid Sensor Networks | |
| CN118504847B (en) | Intelligent beam field management method and system based on digital twin technology |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |