CN117436692A - Electricity consumption-based default electricity consumption checking system - Google Patents

Electricity consumption-based default electricity consumption checking system Download PDF

Info

Publication number
CN117436692A
CN117436692A CN202311383876.4A CN202311383876A CN117436692A CN 117436692 A CN117436692 A CN 117436692A CN 202311383876 A CN202311383876 A CN 202311383876A CN 117436692 A CN117436692 A CN 117436692A
Authority
CN
China
Prior art keywords
data
electricity
electricity consumption
model
default
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
Application number
CN202311383876.4A
Other languages
Chinese (zh)
Inventor
常乐
周开保
郑抗震
刘辉舟
王海鸿
夏泽举
胡吕龙
庄磊
梁晓伟
蔺菲
金耀
王波
赵骞
陈徽
倪妍妍
蔡漪
陶琳
朱祥东
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Marketing Service Center of State Grid Anhui Electric Power Co Ltd
Original Assignee
Marketing Service Center of State Grid Anhui Electric Power Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Marketing Service Center of State Grid Anhui Electric Power Co Ltd filed Critical Marketing Service Center of State Grid Anhui Electric Power Co Ltd
Priority to CN202311383876.4A priority Critical patent/CN117436692A/en
Publication of CN117436692A publication Critical patent/CN117436692A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0635—Risk analysis of enterprise or organisation activities
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Electricity, gas or water supply
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Abstract

The invention discloses a default electricity consumption checking system based on electricity consumption, and relates to the technical field of electricity consumption checking. The system comprises an electricity acquisition module, a data screening module, a user evaluation module and a data evaluation module; the electricity acquisition module is used for acquiring first electricity data and transmitting the first electricity data and the user I D to the data screening module; the data screening module sets the first electric data with high default risk level as high-risk electric data, and sets the first electric data with low default risk level as low-risk electric data; the data evaluation module processes the high-risk electricity consumption data through the first evaluation model and judges the illegal electricity consumption condition, and processes the low-risk electricity consumption data through the second evaluation model and judges the illegal electricity consumption condition; and the user evaluation module sets the default risk level of each user according to the default electricity consumption condition. According to the method and the device, the electricity consumption data of different risks are distinguished and processed through different models, so that the detection precision and the calculation cost are considered.

Description

Electricity consumption-based default electricity consumption checking system
Technical Field
The invention relates to the technical field of electricity consumption inspection, in particular to a default electricity consumption inspection system based on electricity consumption.
Background
The electricity consumption inspection is an evaluation and monitoring method widely applied in the electric power field, and aims to detect and evaluate electricity consumption conditions of users and judge whether illegal electricity consumption behaviors exist. The electricity consumption inspection mainly relates to the problems of the electric energy quality, the electricity consumption safety, the energy efficiency and the like of users. The method is mainly used for guaranteeing safe, stable and reliable operation of the power system, reducing economic loss of power supply enterprises, and promoting informatization management and energy conservation of power utilization users. In the traditional manual inspection mode, workers are used for performing field investigation and collecting data, and then the electricity consumption condition is evaluated through manual judgment and data analysis. Obviously, the manual inspection mode needs to be carried out for a long time and a long period, so that the working efficiency is low, special technicians are required to process and judge the equipment, and the cost is high.
Therefore, an electric checking system for default based on artificial intelligence technology is generated. The existing illegal electricity inspection system is used for uniformly analyzing and processing the collected electricity data, and is difficult to consider the analysis accuracy and calculation efficiency.
Disclosure of Invention
In order to solve the problems existing in the prior art, the invention adopts the following technical scheme:
the system comprises an electricity acquisition module, a data screening module, a user evaluation module and a data evaluation module;
the electricity acquisition module is used for acquiring first electricity data and transmitting the first electricity data and the user ID to the data screening module;
the data screening module acquires the default risk level of the user according to the user ID, sets the first electricity data with high default risk level as high-risk electricity consumption data, and sets the first electricity data with low default risk level as low-risk electricity consumption data;
the data evaluation module comprises a first evaluation model and a second evaluation model, wherein the data evaluation module processes high-risk electricity consumption data and judges illegal electricity consumption conditions through the first evaluation model, and processes low-risk electricity consumption data and judges illegal electricity consumption conditions through the second evaluation model; the first evaluation model is a CNN-LSTM model; the second evaluation model is a gradient lifting model;
and the user evaluation module sets the default risk level of each user according to the default electricity consumption condition.
In a preferred embodiment of the present invention, the data screening module includes a feature extraction unit, and the feature extraction unit obtains a correlation coefficient between each feature in the power consumption data and the default power consumption condition through correlation detection.
As a preferable scheme of the invention, the high-risk electricity consumption data is obtained by screening the characteristics that the correlation coefficient in the first electricity data with high default risk level is larger than a first correlation threshold value, and the low-risk electricity consumption data is obtained by screening the characteristics that the correlation coefficient in the first electricity data with low default risk level is larger than a second correlation threshold value; the first correlation threshold is less than the second correlation threshold.
As a preferred embodiment of the present invention, the correlation detection is expressed as:
wherein t represents a correlation coefficient, and the larger the value of t is, the higher the correlation between the feature X and the default electricity consumption condition is;and->Average value of samples of characteristic X in electricity data when electricity is consumed in violation and in non-violation respectively, n 1 And n 2 Respectively violations ofSample size of characteristic X in electricity data at about and no violations of electricity consumption, ++>And->The sample variance of feature X in the electricity usage data when electricity is consumed is breached and not breached, respectively.
As a preferable scheme of the invention, the CNN-LSTM model comprises a signal input layer, a convolution layer, a pooling layer, a long-short-time memory layer and a full connection layer; the input layer receives the high risk of breach data as an input signal.
As a preferred embodiment of the present invention, the output of the convolution layer is expressed as:
wherein x is i (t) is the time sequence of the input signal, K is the number of convolution kernels, x represents the convolution operator, w k,i Is the weight of the kth convolution kernel ith input signal, b k Is the bias term for the kth convolution kernel, D is the dimension of the input signal.
As a preferable scheme of the invention, the training process of the gradient lifting model comprises the following steps:
s1, initializing a basic decision tree model, and fitting the basic decision tree model to serve as a gradient lifting model by using training data;
s2, calculating residual errors between the predicted result and the real result of the gradient lifting model;
s3, training a new decision tree model according to the residual error and training data;
s4, adding the new decision tree model and the gradient lifting model, and taking the obtained result as a new gradient lifting model;
s5, repeating the steps S2-S4 until the objective function of the gradient lifting model is not reduced or reaches the preset number of repeated rounds.
As a preferred embodiment of the present invention, the objective function of the gradient lifting model is expressed as:
wherein y is i Representing the true label of the i-th sample,representing the prediction result of the ith sample, n representing the number of samples, θ representing the parameter vector of the gradient lifting model, +.>Represents a loss function, k represents the number of decision trees, f k Represents the kth decision tree, Ω (f k ) To represent decision tree f k Regularization term of complexity.
As a preferred embodiment of the present invention, the step S1 further includes: a gain threshold is set.
The step S4 specifically includes: adding the new decision tree model and the gradient lifting model, and comparing objective functions before and after the addition to obtain splitting gain; if the splitting gain is larger than the gain threshold, taking the obtained result as a new gradient lifting model; if the split gain is less than the gain threshold, the next step is entered.
As a preferred scheme of the invention, the user evaluation module is also used for calculating user scores;
the user score is expressed as:
score＝max(X 1 μ 1 (t),X 2 μ 2 (t),...,X M μ M (t)),
wherein M is the total number of scoring items, M is a positive integer greater than 2, X 1 To X M Score, μ for score items 1-M 1 (t) to mu M (t) is the weight coefficient of the 1 st to M th scoring item;
the weight coefficient of the scoring term is expressed as:
μ i (t)＝μ i,1 +g i (t i )μ i,2 ,
i＝1,2,...,M,
wherein mu i,1 First weight, μ, of the ith scoring item i,2 Second weight, μ, for the i-th scoring item i,1 Sum mu i,2 Are all greater than 0 and less than 0.5; t is t i For the current time and fraction X i Time difference of latest update time, delta i Is a blending parameter which is independently set according to the scoring item.
Compared with the prior art, the invention has the following beneficial effects:
the data evaluation module comprises a CNN-LSTM model and a gradient lifting model, and the CNN-LSTM model is used for processing high-risk electricity consumption data, so that the illegal electricity consumption condition can be judged more accurately; and for low-risk electricity consumption data, a gradient lifting model is adopted for evaluation, so that compared with a CNN-LSTM model, the result can be processed more quickly in a shorter time, the evaluation effect is ensured, and meanwhile, the calculation efficiency is reduced. According to the method and the device, the high-risk electricity consumption data and the low-risk electricity consumption data are distinguished and processed through different models, so that the detection precision of the high-risk electricity consumption data and the calculation cost of the low-risk electricity consumption data are considered.
According to the method, the correlation detection is used as a feature extraction means, and the feature extraction unit is used for obtaining the correlation coefficient between each feature in the power consumption data and the illegal power consumption condition, so that the screened data is guaranteed to have higher correlation, and the judgment of the illegal power consumption behavior can be more accurately carried out. In the data screening module, different correlation thresholds are set according to different high-violation risk levels and different low-violation risk levels, so that different power consumption data are screened out. Because the first correlation threshold is smaller than the second correlation threshold, the high-risk electricity consumption data can retain more data characteristics than the low-risk electricity consumption data, so that the accuracy and the efficiency of electricity consumption inspection are considered.
According to the method, the low-risk electricity consumption data can be effectively evaluated and predicted through training and iteration gradient lifting models, the influence degree of the new decision tree model on the whole model can be judged through setting the gain threshold value and detecting the splitting gain in the training process, the influence of the new decision tree model on the whole gradient lifting model is prevented from being too large, the stability, the accuracy and the generalization performance of the whole system are effectively improved, and the model can be more suitable for diversified requirements and data scenes.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a schematic structural diagram of a system for checking electricity consumption violations based on electricity consumption according to an embodiment of the present invention;
fig. 2 is a schematic diagram of a training process of a gradient lifting model according to an embodiment of the invention.
Detailed Description
The technical solutions of the present invention will be clearly and completely described in connection with the embodiments, and it is obvious that the described embodiments are only some embodiments of the present invention, 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. It is noted that relational terms such as "first" and "second", and the like, are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions.
Referring to fig. 1, the invention provides a power consumption based default electricity consumption checking system, which comprises an electricity consumption acquisition module, a data screening module, a user evaluation module and a data evaluation module. The power utilization acquisition module, the data screening module and the data evaluation module are sequentially in communication connection, and the power utilization acquisition module is also in communication connection with the data evaluation module.
The electricity acquisition module is used for acquiring first electricity data and transmitting the first electricity data and the user ID to the data screening module; the electricity collection module is responsible for collecting electricity data of a user, and the electricity data obtained in the part is recorded as first electricity data for distinguishing the electricity data from the electricity data after subsequent processing. The first electricity data comprises electricity consumption, electricity consumption time and the like. The first electricity data can be automatically acquired through the intelligent ammeter, and a user ID is attached to the data section when the first electricity data is transmitted to the data screening module.
The data screening module classifies the first electricity data to obtain second electricity data and third electricity data; specifically, the data screening module acquires the default risk level of the user according to the user ID, sets the first electricity data with the high default risk level as high-risk electricity consumption data, and sets the first electricity data with the low default risk level as low-risk electricity consumption data; the user information includes at least a user ID, a user type, a power usage type, a grid user, and contract information. The user types include families, schools, hospitals, enterprises and the like, and the electricity types include commercial electricity and residential electricity.
The data evaluation module comprises a first evaluation model and a second evaluation model, wherein the data evaluation module processes high-risk electricity consumption data and judges illegal electricity consumption conditions through the first evaluation model, and processes low-risk electricity consumption data and judges illegal electricity consumption conditions through the second evaluation model.
And the user evaluation module sets the default risk level of each user according to the default electricity consumption condition. Specifically, through the condition of the default electricity consumption in a certain time, a user evaluation module calculates user scores, and when the user scores are larger than a high risk threshold, the corresponding users are divided into high default risk grades; otherwise, the corresponding user is classified as a low breach risk level.
The data evaluation module comprises a CNN-LSTM model and a gradient lifting model, and the CNN-LSTM model is used for processing high-risk electricity consumption data, so that more electricity consumption characteristics can be extracted, and the illegal electricity consumption condition can be judged more accurately. And for low-risk electricity consumption data, a gradient lifting model is adopted for evaluation, so that compared with a CNN-LSTM model, the result can be processed more quickly in a shorter time, the evaluation accuracy is ensured, and the calculation cost is reduced. According to the method and the device, the high-risk electricity consumption data and the low-risk electricity consumption data are distinguished and processed through different models, so that the detection precision of the high-risk electricity consumption data and the calculation cost of the low-risk electricity consumption data are considered.
As a preferred embodiment, the user score is expressed as:
score＝max(X 1 μ 1 (t),X 2 μ 2 (t),...,X M μ M (t))
wherein M is the total number of scoring items, M is a positive integer greater than 1, X 1 To X M Score, μ for score items 1-M 1 (t) to mu M And (t) is the weight coefficient of the 1 st to M th scoring items, which changes with time and is used for controlling the importance degree of different data. The scoring item at least comprises the default electricity consumption condition provided by the data evaluation module. In the present application, other scoring items come from channels other than the data evaluation module, such as channels for telecommunications reporting, offline inspection, and the like.
Further, the weight coefficient of the scoring item is expressed as:
μ i (t)＝μ i,1 +g i (t i )μ i,2
i＝1,2,...,M
wherein mu i,1 First weight, μ, of the ith scoring item i,2 Second weight, μ, for the i-th scoring item i,1 Sum mu i,2 Are all greater than 0 and less than 0.5; t is t i For the current time and fraction X i Time difference of latest update time, delta i Based on scoring itemsAnd (5) independently setting blending parameters.
As a preferred embodiment, the data screening module includes a feature extraction unit, where the feature extraction unit obtains, through correlation detection, a correlation coefficient between each feature in the power consumption data and the default power consumption condition, the high risk power consumption data is obtained by screening features with a correlation coefficient greater than a first correlation threshold in the first power consumption data with a high default risk level, and the low risk power consumption data is obtained by screening features with a correlation coefficient greater than a second correlation threshold in the first power consumption data with a low default risk level. The first correlation threshold is smaller than the second correlation threshold, namely the data size of the high-risk electricity consumption data is larger than the data size of the low-risk electricity consumption data.
The correlation detection is expressed as:
wherein t represents a correlation coefficient, and the larger the value of t is, the higher the correlation between the feature X and the default electricity consumption condition is;and->Average value of samples of characteristic X in electricity data when electricity is consumed in violation and in non-violation respectively, n 1 And n 2 Sample size of characteristic X in electricity data when electricity is consumed in violation and in non-violation respectively, +.>And->Power consumption when power consumption is not in default and not in default respectivelySample variance of feature X in the data.
In the embodiment, the correlation detection is used as a feature extraction means, and the feature extraction unit is used for obtaining the correlation coefficient between each feature in the power consumption data and the illegal power consumption condition, so that the screened data is ensured to have higher correlation, and the judgment of the illegal power consumption behavior can be more accurately carried out. In the data screening module, different correlation thresholds are set according to different high-violation risk levels and different low-violation risk levels, so that different power consumption data are screened out. Because the first correlation threshold is smaller than the second correlation threshold, the high-risk electricity consumption data can retain more data characteristics than the low-risk electricity consumption data, so that the accuracy and the efficiency of electricity consumption inspection are considered.
As a preferred embodiment, the first evaluation model is a CNN-LSTM model. The CNN-LSTM model comprises a signal input layer, a convolution layer, a pooling layer, a long-short-time memory layer and a full connection layer. Wherein the input layer receives high risk of breach data as an input signal.
The convolution layer is used to extract spatial features of the input signal. The output of the convolutional layer is expressed as:
wherein x is i (t) is the time sequence of the input signal, K is the number of convolution kernels, x represents the convolution operator, w k,i Is the weight of the kth convolution kernel ith input signal, b k Is the bias term for the kth convolution kernel, D is the dimension of the input signal.
The pooling layer reduces the dimension of the feature map and extracts the main features by pooling after the convolution layer. The pooling mode adopted in this embodiment is maximum pooling.
The long-time and short-time memory layer is used for carrying out time sequence modeling on the output characteristic sequence of the pooling layer. Specifically, the long-short-time memory model includes a memory unit, an input gate, a forgetting gate and an output gate, and the long-short-time memory model can be expressed as:
f t ＝σ(W f h t-1 +U f h p (t)+b f )
i t ＝σ(W i h t-1 +U i h p (t)+b i )
o t ＝σ(W o h t-1 +U o h p (t)+b o )
h t ＝o t ⊙tanh(c t )
wherein h is t Represents the hidden state of the current moment, h t-1 Indicating the hidden state, i, of the previous moment t H is the output of the input gate p (t) is the output of the pooling layer, f t C, outputting to the forgetting gate t Memory cell, c, representing the current time t A memory cell representing a previous time instant is shown,calculation result representing candidate cell state, o t Representing the output of the output gate, W i 、W f 、W c 、W o Weights are respectively connected for hidden states of the input gate, the forgetting gate, the memory unit and the output gate, U i 、U f 、U c 、U o Weights are respectively connected to input features of the input gate, the forget gate, the memory unit and the output gate, b i 、b f 、b c 、b o Bias vectors for the input gate, the forget gate, the memory cell and the output gate, respectively, ++denotes Hadamard product, (-) is a sigmoid function.
The full connection layer is used for mapping the output of the long-short time memory layer to the required output dimension after the long-short time memory layer, and applying an activation function to obtain a final measurement result.
The method adopts the CNN-LSTM model as the first evaluation model, and can capture the space-time characteristics of the electricity utilization data. The CNN layer can automatically extract local features in signal data, and based on the local features, the LSTM layer can model time sequence dependency relations among signal sequences, so that accuracy of a model is improved. In the task of processing high-risk electricity consumption data to judge the illegal electricity consumption condition, the CNN-LSTM model can accurately identify abnormal data in the electricity consumption data, and a high-precision classification effect is realized. And, since the CNN-LSTM model can respond to the input of each neuron, it is analyzed. For example, in electricity data processing, the model may present the user with visual results of the power load condition, the daily electricity amount, the electricity rate information, and the like. Because the LSTM layer is provided with an internal information storage unit in the data calculation process, the model can find and store long-term rules and rules of the data sequence, so that the model has better interpretability.
As a preferred embodiment, the second evaluation model is a gradient lifting model. Specifically, the gradient lifting model adopts the idea of gradient lifting decision trees, namely, classification and regression prediction are performed by using a plurality of decision trees.
Referring to fig. 2, the training process of the gradient lifting model includes the steps of:
s1, initializing a basic decision tree model, and fitting the basic decision tree model to serve as a gradient lifting model by using training data;
s2, calculating residual errors between the predicted result and the real result of the gradient lifting model;
s3, training a new decision tree model according to the residual error and training data; in the process, regularization terms including L1, L2, sub-sampling, column sampling and/or weighted sampling are used, so that the generalization performance of the model can be improved and the overfitting can be reduced;
s4, adding the new decision tree model and the gradient lifting model, and taking the obtained result as a new gradient lifting model;
s5, repeating the steps S2-S4 until the objective function of the gradient lifting model is not reduced or reaches the preset number of repeated rounds. It should be noted that the new decision tree model mainly focuses on samples with frequent prediction errors in training data, that is, residuals still exist after the current gradient lifting model is fitted. Therefore, by continually adding new decision tree models to the original base decision tree model, a balance of high accuracy and generalization performance can be achieved.
According to the method and the device, the low-risk electricity consumption data can be effectively evaluated and predicted through training and iterative gradient lifting models. The gradient lifting model is characterized in that the defect of a single decision tree can be overcome, and the prediction precision and effect of the model are improved. The regularization term and the sampling technology are utilized, so that the model can be prevented from being overfitted, and different data sets and requirements can be well adapted in practical application.
Wherein, the objective function of the gradient lifting model is expressed as:
wherein y is i Representing the true label of the i-th sample,representing the prediction result of the ith sample, n representing the number of samples, θ representing the parameter vector of the gradient lifting model, +.>Represents a loss function, k represents the number of decision trees, f k Represents the kth decision tree, Ω (f k ) To represent decision tree f k Regularization term of complexity.
As a preferred embodiment, the step S1 further includes:
a gain threshold is set.
The step S4 specifically includes:
adding the new decision tree model and the gradient lifting model, and comparing objective functions before and after the addition to obtain splitting gain; if the splitting gain is larger than the gain threshold, taking the obtained result as a new gradient lifting model; if the split gain is less than the gain threshold, the next step is entered.
According to the embodiment, the influence degree of the new decision tree model on the whole model can be judged by setting the gain threshold and detecting the splitting gain in the training process, so that the influence of the new decision tree model on the whole gradient lifting model is prevented from being too large, the stability, the accuracy and the generalization performance of the whole system are effectively improved, and the model can be more suitable for diversified requirements and data scenes.
In the several embodiments provided in this application, it should be understood that the disclosed units may be implemented in other ways. For example, the above-described embodiments of the units are merely illustrative, e.g., the division of the units is merely a logical functional division, and there may be additional divisions when actually implemented, e.g., multiple units or components may be combined or integrated into another unit, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection of modules, electrical, mechanical, or other forms.
The units described as separate parts may or may not be physically separate, and parts displayed as modules may or may not be physical modules, i.e. may be located in one place, or may be distributed over a plurality of units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist alone physically, or two or more modules may be integrated into one module. The integrated modules may be implemented in hardware or in software functional modules.
The integrated modules, if implemented in the form of software functional modules and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application may be embodied in essence or a part contributing to the prior art or all or part of the technical solution in the form of a software product stored in a storage medium, including several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a read-only memory (ROM), a random access memory (RAM, random access memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.

Claims (10)

1. The utility model provides a system is checked to electricity consumption-based default electricity consumption which characterized in that: the system comprises an electricity acquisition module, a data screening module, a user evaluation module and a data evaluation module;
the electricity acquisition module is used for acquiring first electricity data and transmitting the first electricity data and the user ID to the data screening module;
the data screening module acquires the default risk level of the user according to the user ID, sets the first electricity data with high default risk level as high-risk electricity consumption data, and sets the first electricity data with low default risk level as low-risk electricity consumption data;
the data evaluation module comprises a first evaluation model and a second evaluation model, wherein the data evaluation module processes high-risk electricity consumption data and judges illegal electricity consumption conditions through the first evaluation model, and processes low-risk electricity consumption data and judges illegal electricity consumption conditions through the second evaluation model; the first evaluation model is a CNN-LSTM model; the second evaluation model is a gradient lifting model;
and the user evaluation module sets the default risk level of each user according to the default electricity consumption condition.
2. The electricity consumption-based default electricity auditing system of claim 1, wherein: the data screening module comprises a feature extraction unit, and the feature extraction unit obtains a correlation coefficient of each feature in the power consumption data and the illegal power consumption condition through correlation detection.
3. The electricity consumption-based default electricity auditing system of claim 2, wherein: the high-risk electricity consumption data is obtained by screening the characteristics that the correlation coefficient in the first electricity data with high default risk level is larger than a first correlation threshold value, and the low-risk electricity consumption data is obtained by screening the characteristics that the correlation coefficient in the first electricity data with low default risk level is larger than a second correlation threshold value; the first correlation threshold is less than the second correlation threshold.
4. The electricity consumption-based default electricity auditing system of claim 2, wherein: the correlation detection is expressed as:
wherein t represents a correlation coefficient, and the larger the value of t is, the higher the correlation between the feature X and the default electricity consumption condition is;andaverage value of samples of characteristic X in electricity data when electricity is consumed in violation and in non-violation respectively, n 1 And n 2 Sample size of characteristic X in electricity data when electricity is consumed in violation and in non-violation respectively, +.>And->The sample variance of feature X in the electricity usage data when electricity is consumed is breached and not breached, respectively.
5. The electricity consumption-based default electricity auditing system of claim 1, wherein: the CNN-LSTM model comprises a signal input layer, a convolution layer, a pooling layer, a long-short-time memory layer and a full-connection layer; the input layer receives the high risk of breach data as an input signal.
6. The electricity consumption-based default electricity auditing system of claim 5, wherein:
the output of the convolution layer is expressed as:
wherein x is i (t) is the time sequence of the input signal, K is the number of convolution kernels, x represents the convolution operator, w k,i Is the weight of the kth convolution kernel ith input signal, b k Is the bias term for the kth convolution kernel, D is the dimension of the input signal.
7. The electricity consumption-based default electricity auditing system of claim 1, wherein: the training process of the gradient lifting model comprises the following steps:
s1, initializing a basic decision tree model, and fitting the basic decision tree model to serve as a gradient lifting model by using training data;
s2, calculating residual errors between the predicted result and the real result of the gradient lifting model;
s3, training a new decision tree model according to the residual error and training data;
s4, adding the new decision tree model and the gradient lifting model, and taking the obtained result as a new gradient lifting model;
s5, repeating the steps S2-S4 until the objective function of the gradient lifting model is not reduced or reaches the preset number of repeated rounds.
8. The electricity consumption-based default electricity auditing system of claim 7, wherein: the objective function of the gradient lifting model is expressed as:
wherein y is i Representing the true label of the i-th sample,representing the prediction result of the ith sample, n representing the number of samples, θ representing the parameter vector of the gradient lifting model, +.>Represents a loss function, k represents the number of decision trees, f k Represents the kth decision tree, Ω (f k ) To represent decision tree f k Regularization term of complexity.
9. The electricity consumption-based default electricity auditing system of claim 7, wherein: the step S1 further includes: setting a gain threshold;
the step S4 specifically includes: adding the new decision tree model and the gradient lifting model, and comparing objective functions before and after the addition to obtain splitting gain; if the splitting gain is larger than the gain threshold, taking the obtained result as a new gradient lifting model; if the split gain is less than the gain threshold, the next step is entered.
10. The electricity consumption-based default electricity auditing system of claim 1, wherein: the user evaluation module is also used for calculating user scores;
the user score is expressed as:
score＝max(X 1 μ 1 (t)，X 2 μ 2 (t)，...，X M μ M (t))，
wherein M is the total number of scoring items, M is a positive integer greater than 2, X 1 To X M Score, μ for score items 1-M 1 (t) to mu M (t) is the weight coefficient of the 1 st to M th scoring item;
the weight coefficient of the scoring term is expressed as:
μ i (t)＝μ i，1 +g i (t i )μ i，2 ，
i＝1，2，....，M，
wherein mu i，1 First weight, μ, of the ith scoring item i，2 Second weight, μ, for the i-th scoring item i，1 Sum mu i，2 Are all greater than 0 and less than 0.5; t is t i For the current time and fraction X i Time difference of latest update time, delta i Is a blending parameter which is independently set according to the scoring item.
CN202311383876.4A 2023-10-24 2023-10-24 Electricity consumption-based default electricity consumption checking system Pending CN117436692A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311383876.4A CN117436692A (en) 2023-10-24 2023-10-24 Electricity consumption-based default electricity consumption checking system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311383876.4A CN117436692A (en) 2023-10-24 2023-10-24 Electricity consumption-based default electricity consumption checking system

Publications (1)

Publication Number Publication Date
CN117436692A true CN117436692A (en) 2024-01-23

Family

ID=89549259

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311383876.4A Pending CN117436692A (en) 2023-10-24 2023-10-24 Electricity consumption-based default electricity consumption checking system

Country Status (1)

Country Link
CN (1) CN117436692A (en)

Similar Documents

Publication Publication Date Title
CN110263846B (en) Fault diagnosis method based on fault data deep mining and learning
CN102955902B (en) Method and system for evaluating reliability of radar simulation equipment
CN113887616B (en) Real-time abnormality detection method for EPG connection number
CN111343147B (en) Network attack detection device and method based on deep learning
CN112085621B (en) Distributed photovoltaic power station fault early warning algorithm based on K-Means-HMM model
CN112700324A (en) User loan default prediction method based on combination of Catboost and restricted Boltzmann machine
CN110119758A (en) A kind of electricity consumption data abnormality detection and model training method, device
CN115660262B (en) Engineering intelligent quality inspection method, system and medium based on database application
CN108879732A (en) Transient stability evaluation in power system method and device
CN112001644A (en) Power distribution network operation reliability detection method, device, terminal and storage medium
CN116610998A (en) Switch cabinet fault diagnosis method and system based on multi-mode data fusion
CN115423594A (en) Enterprise financial risk assessment method, device, equipment and storage medium
CN115617784A (en) Data processing system and processing method for informationized power distribution
CN111461923A (en) Electricity stealing monitoring system and method based on deep convolutional neural network
CN111612149A (en) Main network line state detection method, system and medium based on decision tree
CN113268870B (en) Monte Carlo-based image recognition reliability evaluation method under outdoor environment condition
CN112365093A (en) GRU deep learning-based multi-feature factor red tide prediction model
CN116384223A (en) Nuclear equipment reliability assessment method and system based on intelligent degradation state identification
CN117436692A (en) Electricity consumption-based default electricity consumption checking system
CN113151842B (en) Method and device for determining conversion efficiency of wind-solar complementary water electrolysis hydrogen production
CN115146715A (en) Power utilization potential safety hazard diagnosis method, device, equipment and storage medium
Li et al. Social network change detection using a genetic algorithm based back propagation neural network model
CN114818849A (en) Convolution neural network based on big data information and anti-electricity-stealing method based on genetic algorithm
CN113657726A (en) Personnel risk analysis method based on random forest
CN113962508A (en) Identification method and identification device for electricity object and electronic equipment

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