CN106779069A - A kind of abnormal electricity consumption detection method based on neutral net - Google Patents
A kind of abnormal electricity consumption detection method based on neutral net Download PDFInfo
- Publication number
- CN106779069A CN106779069A CN201611122796.3A CN201611122796A CN106779069A CN 106779069 A CN106779069 A CN 106779069A CN 201611122796 A CN201611122796 A CN 201611122796A CN 106779069 A CN106779069 A CN 106779069A
- Authority
- CN
- China
- Prior art keywords
- data
- hidden layer
- model
- output
- formula
- 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
Classifications
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
Abstract
A kind of abnormal electricity consumption detection method based on neutral net, whether the method is based on setting up abnormal electricity consumption detection model, is diagnosed and is analyzed by the operating condition to equipment, judge measuring equipment in normal operating condition, decision-making function is realized, described method specifically includes following steps:1)Data acquisition, described data are mainly derived from the standby class data such as electric energy measurement data in electric energy meter and acquisition terminal, operating condition data and logout;2)Data cleansing, the data for using need to be allowed for access model after data cleansing and screening;3)Data are classified, and are completed after data cleansing, and data are demarcated, and are classified according to the numeral of classification in a last plus list registration of data, and the data integration for completing data scaling is training data;4)Modeling procedure, the developing algorithm model by the way of supervised learning;5)Model realization;6)Interpretation of result, the final abnormal electricity consumption of model finds that accuracy rate maintains higher level.
Description
Technical field
The present invention relates to a kind of abnormal electricity consumption detection method based on neutral net, belong to the exception electric-examination of power network
Survey technology field.
Background technology
With the continuous improvement of the level of informatization, the big data epoch have arrived, from the relatively low data of a large amount of value densities
In excavate valuable information, have become all trades and professions close attention hot issue.For power industry, with
The continuous improvement of electric power information degree and increasing rapidly with electricity consumption data volume, all kinds of apparatus and systems have substantial amounts of
Data will be processed, and data scale is huge, and the event information species for containing is various, but up to the present, still suffer from " data sea
The major issue of amount, absence of information ".
Simultaneously as the reason such as various communication failures, equipment fault, power network fluctuation and user exception electricity consumption behavior, goes out
The abnormal phenomenon of mass data is showed.These abnormal datas have impact on accuracy, completeness, self-consistency and the dynamic of energy data
Property, but also contained the critical event information of power network, therefore the algorithm excavated suitable for extensive electricity consumption data is studied, and set up
Effective anomaly model carries out the analysis of abnormal power information, identification and processes, for power industry analysis mining thing
The development of part information and intelligent grid is significant.
For equipment fault and the detection of user's exception electricity consumption, early stage uses in-situ check and test method, i.e. technology people more
Member is investigated to electricity consumption scene.This processing mode extremely labor intensive material resources, efficiency is low, effect is poor, meanwhile, it is this
There is great human factor in mode, be unfavorable for the management of power industry.
In recent years, people seeking always it is a kind of efficiently accurately algorithm is found and is sorted out to anomalous event, and
Shi Faxian, prevention and treatment power consumer multiplexing electric abnormality, improve power consumer service quality.
The content of the invention
It is an object of the invention to overcome the shortcomings of that prior art is present, and provide a kind of based on the neutral net for having supervision
Abnormal electricity consumption detection model by large data sets training after, quickly to find multiplexing electric abnormality, accuracy rate is high, can effectively help
Help relevant department to reduce investigation scope, and for Utilities Electric Co. uses manpower and material resources sparingly resource, reduces cost based on neutral net
Abnormal electricity consumption detection method.
The purpose of the present invention is completed by following technical solution, a kind of exception electro-detection based on neutral net
Method, the method is based on setting up abnormal electricity consumption detection model, is diagnosed and is analyzed by the operating condition to equipment, judges meter
Whether amount equipment is in normal operating condition, realizes decision-making function;Described method specifically includes following steps:
1) data acquisition, described data are mainly derived from electric energy measurement data, operation in electric energy meter and acquisition terminal
The standby class data such as floor data and logout;
2) data cleansing, the data for using need to be allowed for access model after data cleansing and screening, reject content and be not inconsistent
Situations such as data of data protocol call format, including data mess code are closed with number completion is answered according to for sky;Reject content substantially wrong
Data, including Data writing time is later than current time feelings after a few days earlier than equipment set-up time and Data writing time
Condition;
3) data classification, is completed after data cleansing, and data are demarcated, in a last plus list registration of data
Numeral according to classification is classified, and the data integration for completing data scaling is training data;
4) modeling procedure, the developing algorithm model by the way of supervised learning, will input data be referred to as training data,
Every group of training data has clearly mark or a result, when algorithm model is set up, sets up a learning process, learns
The immanent structure of data obtains result reasonably to organize data;
The BP neural network of many hidden layers is set up with Matlab.BP neural network algorithm includes two aspects:The forward direction of signal
Propagate the backpropagation with error.Carried out by from the direction for being input to output when namely calculating reality output, and weights and threshold
The adjustment of value is then carried out by the direction from exporting to being input into;
5) model realization:
The first step carries out data normalization treatment, maps the data into [0,1] or [- 1,1] interval or smaller interval
Interior, prevent neutral net causes convergence rate excessively slow because the unit and scope of input data are inconsistent;
Second step is selected each function, and activation primitive is selected first, and the activation primitive of BP neural network must be place
Place can be micro-, most frequently be the logarithmic function of S types or the tan and linear function of S types, general neutral net
Algorithm model, in hidden layer transfer function S type activation primitives, the transfer function of output layer then selects linear activation primitive;Its
Secondary is the selection to training function, using the adjusting learning rate function traingdx of momentum one;
3rd step is the selection to the network number of plies and hidden layer nodes, on the premise of it ensure that precision, is preferentially examined
Hidden layer nodes as few as possible are taken in worry;Finally determine with formulaHidden layer nodes are determined, for difference
Training data, hidden layer nodes are also uncertain;
4th step carries out the selection of each function, and the establishment function newff carried using Matlab creates a BP nerve net
Network;Neutral net is after foundation, in addition it is also necessary to set some parameters manually, and these parameters are all to continuously attempt to choose by experiment
, while also there is the reference of some historical experiences, epochs is the maximum train epochs of network, and goal is the minimum of training objective
Error, show represents the gap periods of training result display frequency, and lr is learning rate;
What is finally carried out is model initialization and emulation, the neutral net initialization function init carried using Matlab,
And emulated using sim functions;
6) interpretation of result, by constantly adjust and change each parameter and hidden layer nodes after, the final exception of model
Electricity consumption finds that accuracy rate maintains higher level.
As preferred:Described step 2) in, after rejecting the data of apparent error, in artificial rejecting or modification normal data
The not clear noise data of some reasons, because these noise datas easily allow neural network model to learn to excessive information, leads
Cause the accuracy rate reduction of neural network model;
Described step 4), error is backpropagation, i.e., each layer neuron is successively calculated first by output layer
Output error, then adjusts the weights and threshold value of each layer according to error gradient descent method, makes the final defeated of amended network
Going out can be close to desired value;
According to error gradient descent method, shown in output layer weighed value adjusting formula such as formula (1):
Shown in output layer adjusting thresholds formula such as formula (2):
Shown in hidden layer weighed value adjusting formula such as formula (3):
Shown in hidden layer adjusting thresholds formula such as formula (4):
In formula:η is learning rate;TP KIt is desired output;OP KIt is the output after sample input network;netkIt is hidden layer kth
The input of individual node;yiIt is the output of i-th node of hidden layer;
Described step 5), in second step, the S type activation primitives of selection are logsig functions, the value of function input
Between (∞, ∞), output is then on (0,1) interval;
Described step 5), in the 3rd step, it is generally the case that the network number of plies is more, and the error of neutral net is just smaller, when
So it is not excluded for increasing with the network number of plies, the situation that the error of neutral net increases on the contrary;If increasing hidden layer simply
Quantity, often increase network complexity, also have great possibility can make network occur " over-fitting " phenomenon;Most
Conventional neutral net is usually 3 layer networks, that is, have 1 hidden layer, meanwhile, give increase hidden layer by the reducing error of the task
Nodes undertake, and are so easier to realize;To hidden layer nodes, in BP neural network, the selection of hidden layer nodes
Seem and be even more important, hidden layer nodes can not only cause very big influence to the performance of neural network model, and selection is not
" over-fitting " phenomenon when suitable hidden layer nodes can also directly result in training, but just do not have still from the point of view of current development
The theoretical and method for having set of system explicitly defines hidden layer nodes, and how this selects, and, although there are some empirical functions,
Such as:
M=log2n
In formula:M is hidden layer interstitial content;N is input layer number;, l is output layer nodes:α be 1-10 between
Arbitrary constant;
Step 6) in, due to the limitation of own hardware condition, the amount of test data of selection is not huge enough, compared to electric power
Less for industry data volume hundreds of millions of daily, these data can not completely represent the electricity consumption abnormal behavior of user;Base
In the limitation of above force majeure factor etc., the accuracy rate of model is restricted;Therefore model should be put into real work
In, after being trained using substantial amounts of power industry user power utilization data, it will be largely overcoming due to data deficiencies
The not high enough situation of caused model accuracy rate.
Abnormal electricity consumption detection model of the present invention based on the neutral net for having supervision, can be fast after large data sets are trained
Speed finds multiplexing electric abnormality, and accuracy rate is high, can effectively help relevant department to reduce investigation scope, and for Utilities Electric Co. has saved manpower
Material resources, reduces cost;The present invention has practical value higher, is suitable for large range promotion use;Be conducive to side simultaneously
Help others establish the consciousness of sincere electricity consumption, promote power industry to develop healthy and orderlyly.
Brief description of the drawings
Fig. 1 is the structure chart of BP neural network of the present invention.
Fig. 2 is overall flow chart of steps of the present invention.
Fig. 3 is algorithm routine flow chart of the present invention.
Specific embodiment
The present invention will be described in detail below in conjunction with the accompanying drawings and the specific embodiments:It is of the present invention a kind of based on god
Through the abnormal electricity consumption detection method of network, the method is based on setting up abnormal electricity consumption detection model, by the operating condition to equipment
Diagnosed and analyzed, judged that whether measuring equipment, in normal operating condition, realizes decision-making function;Described method tool
Body comprises the following steps:
1) data acquisition, described data are mainly derived from electric energy measurement data, operation in electric energy meter and acquisition terminal
The standby class data such as floor data and logout;
2) data cleansing, the data for using need to be allowed for access model after data cleansing and screening, reject content and be not inconsistent
Situations such as data of data protocol call format, including data mess code are closed with number completion is answered according to for sky;Reject content substantially wrong
Data, including Data writing time is later than current time feelings after a few days earlier than equipment set-up time and Data writing time
Condition;
3) data classification, is completed after data cleansing, and data are demarcated, in a last plus list registration of data
Numeral according to classification is classified, and the data integration for completing data scaling is training data;
4) modeling procedure, the developing algorithm model by the way of supervised learning, will input data be referred to as training data,
Every group of training data has clearly mark or a result, when algorithm model is set up, sets up a learning process, learns
The immanent structure of data obtains result reasonably to organize data;
The BP neural network of many hidden layers is set up with Matlab.BP neural network algorithm includes two aspects:The forward direction of signal
Propagate the backpropagation with error;Carried out by from the direction for being input to output when namely calculating reality output, and weights and threshold
The adjustment of value is then carried out by the direction from exporting to being input into;
5) model realization:
The first step carries out data normalization treatment, maps the data into [0,1] or [- 1,1] interval or smaller interval
Interior, prevent neutral net causes convergence rate excessively slow because the unit and scope of input data are inconsistent;
Second step is selected each function, and activation primitive is selected first, and the activation primitive of BP neural network must be place
Place can be micro-, most frequently be the logarithmic function of S types or the tan and linear function of S types, general neutral net
Algorithm model, in hidden layer transfer function S type activation primitives, the transfer function of output layer then selects linear activation primitive;Its
Secondary is the selection to training function, using the adjusting learning rate function traingdx of momentum one;
3rd step is the selection to the network number of plies and hidden layer nodes, on the premise of it ensure that precision, is preferentially examined
Hidden layer nodes as few as possible are taken in worry;Finally determine with formulaHidden layer nodes are determined, for difference
Training data, hidden layer nodes are also uncertain;
4th step carries out the selection of each function, and the establishment function newff carried using Matlab creates a BP nerve net
Network;Neutral net is after foundation, in addition it is also necessary to set some parameters manually, and these parameters are all to continuously attempt to choose by experiment
, while also there is the reference of some historical experiences, epochs is the maximum train epochs of network, and goal is the minimum of training objective
Error, show represents the gap periods of training result display frequency, and lr is learning rate;
What is finally carried out is model initialization and emulation, the neutral net initialization function init carried using Matlab,
And emulated using sim functions;
6) interpretation of result, by constantly adjust and change each parameter and hidden layer nodes after, the final exception of model
Electricity consumption finds that accuracy rate maintains higher level.
Further embodiment of the present invention is:
Described step 2) in, it is artificial to reject or change some reasons in normal data after rejecting the data of apparent error
Not clear noise data, because these noise datas easily allow neural network model to learn to excessive information, causes nerve net
The accuracy rate reduction of network model;
Described step 4), error is backpropagation, i.e., each layer neuron is successively calculated first by output layer
Output error, then adjusts the weights and threshold value of each layer according to error gradient descent method, makes the final defeated of amended network
Going out can be close to desired value;
According to error gradient descent method, shown in output layer weighed value adjusting formula such as formula (1):
Shown in output layer adjusting thresholds formula such as formula (2):
Shown in hidden layer weighed value adjusting formula such as formula (3):
Shown in hidden layer adjusting thresholds formula such as formula (4):
In formula:η is learning rate;TP KIt is desired output;OP KIt is the output after sample input network;netkIt is hidden layer kth
The input of individual node;yiIt is the output of i-th node of hidden layer;
Described step 5), in second step, the S type activation primitives of selection are logsig functions, the value of function input
Between (∞, ∞), output is then on (0,1) interval;
Described step 5), in the 3rd step, it is generally the case that the network number of plies is more, and the error of neutral net is just smaller, when
So it is not excluded for increasing with the network number of plies, the situation that the error of neutral net increases on the contrary;If increasing hidden layer simply
Quantity, often increase network complexity, also have great possibility can make network occur " over-fitting " phenomenon;Most
Conventional neutral net is usually 3 layer networks, that is, have 1 hidden layer, meanwhile, give increase hidden layer by the reducing error of the task
Nodes undertake, and are so easier to realize;To hidden layer nodes, in BP neural network, the selection of hidden layer nodes
Seem and be even more important, hidden layer nodes can not only cause very big influence to the performance of neural network model, and selection is not
" over-fitting " phenomenon when suitable hidden layer nodes can also directly result in training, but just do not have still from the point of view of current development
The theoretical and method for having set of system explicitly defines hidden layer nodes, and how this selects, and, although there are some empirical functions,
Such as:
M=log2n
In formula:M is hidden layer interstitial content;N is input layer number;, l is output layer nodes:α be 1-10 between
Arbitrary constant;
Step 6) in, due to the limitation of own hardware condition, the amount of test data of selection is not huge enough, compared to electric power
Less for industry data volume hundreds of millions of daily, these data can not completely represent the electricity consumption abnormal behavior of user;Base
In the limitation of above force majeure factor etc., the accuracy rate of model is restricted;Therefore model should be put into real work
In, after being trained using substantial amounts of power industry user power utilization data, it will be largely overcoming due to data deficiencies
The not high enough situation of caused model accuracy rate.
Embodiment:With the development of artificial intelligence, application of the artificial neural network in terms of classification and prediction is just by people
Accept extensively and study, the principle of artificial neural network algorithm is to allow computer oneself to go summary is how to recognize these exceptions
Classification, and using the data set classified as training data, allow algorithm oneself to look for structured knowledge therein, not only divide
Class result is preferable, and the efficiency of algorithm is very high.
Neutral net brief introduction:Neural network algorithm is one of most important algorithm of machine learning, simulates the god of human brain
Through network, by learning to conclude, obtain can be used for the algorithm model classified and predict;The similar mankind of the operation principle of neutral net
This neuron working principle, simple neutral net is segmented into input layer, hidden layer and output layer.Input layer is responsible for connecing
The collection of letters number, is then passed to hidden layer, and hidden layer is responsible for being analyzed treatment to the data for receiving, through dividing for excessive hidden layer
After analysis treatment, result is integrated into output to output layer;There are corresponding parameter, the process of these parameter adjustments between each layer and each layer
Can just regard that human nerve's network constantly learns the process for updating as.
The neural network algorithm of many hidden layers has excellent feature learning ability, and the feature that study is obtained has more to data
Essential portrays, so as to be conducive to visualizing and classify.Difficulty of the deep neural network in training, can be by " just successively
Beginningization " effectively overcomes;Thus traditional neural network difficulty computationally is not only solved, while also illustrate that deep layer nerve
Network superiority in the study.
The present invention is modeled using the neutral net of many hidden layers;Neutral net can be divided into by mode of learning supervision
4 kinds of study, unsupervised learning, semi-supervised learning and intensified learning, present invention developing algorithm mould by the way of supervised learning
Type.So-called supervised learning mode, is exactly referred to as training data by input data, and every group of training data has a clearly mark
Or result, when algorithm model is set up, a learning process is set up, the immanent structure of learning data is reasonably to organize
Data obtain result.
The BP neural network of many hidden layers is set up with Matlab.BP neural network algorithm includes two aspects:The forward direction of signal
Propagate the backpropagation with error;Carried out by from the direction for being input to output when namely calculating reality output, and weights and threshold
The adjustment of value is then carried out by the direction from exporting to being input into.The structure of BP neural network is shown in Fig. 1.
In BP neural network, between layers using full mutual contact mode, in the absence of interconnection, hidden layer between same layer
The determination that can have one to multilayer, the hidden layer number of plies is obtained by specific experiment.XjThe input of j-th node of input layer is represented,
(j=1,2,3 ..., m);ωi,jRepresent i-th node of hidden layer to the weights between j-th node of input layer;θiRepresent and hide
I-th threshold value of node of layer;φ represents the excitation function of hidden layer;ωk,iRepresent k-th node of output layer to i-th of hidden layer
Weights between node, (i=1,2,3 ..., q);αkRepresent the threshold value of k-th node of output layer, (k=1,2,3 ..., L);ψ
Represent the excitation function of output layer;οkRepresent the output of k-th node of output layer.
The foundation of model:Model Establishing process is by data acquisition, data cleansing, data classification, modeling procedure, model reality
Existing, interpretation of result composition, master-plan flow chart is as shown in Figure 2.
Model is mainly analyzed to the data produced by user power utilization, including the use including the data such as power consumption and voltage
Power information.By the analysis to user power utilization data, abnormality detection model is obtained, so as to be carried out quickly to the exception that future occurs
Accurately find and position.
Whether the present invention is diagnosed and analyzed by the operating condition to equipment, judge measuring equipment in normal operation
State, realizes decision-making function.
The data of model are mainly derived from electric energy measurement data in electric energy meter and acquisition terminal, operating condition data and thing
The standby class data such as part record.By to electric energy meter, the Production conditions of acquisition terminal Various types of data, incidence relation analysis, with reference to
Related service application demand, setting up abnormal electricity consumption detection model carries out abnormity diagnosis analysis.
With reference to metering on-line monitoring and intelligent diagnostics analysis, diagnostic model is divided into single anomaly analysis and associates abnormal point
Analysis.Single anomaly analysis refer to the Frequency analysis frequency ratio in same event in a period of certain, judge single equipment
Real-time status or state variation tendency.Abnormal association analysis is to the incidence relation degree of being associated between single exception point
Analysis, the intelligent data analysis of multiplexing electric abnormality is carried out according to related degree model.Single anomaly analysis include electricity abnormity diagnosis, electricity
Current voltage abnormity diagnosis, exception electrodiagnosis, load abnormity diagnosis, clock abnormity diagnosis, wiring abnormity diagnosis, expense control exception are examined
Break;Abnormal association analysis includes that doubtful stealing, equipment fault, error connection, distribution transforming need dilatation, on-site maintenance, battery failure, return
Road exception, multiplexing electric abnormality etc..
Data cleansing:The data that algorithm model is used need to be allowed for access model after data cleansing and screening, reject in
Not situations such as appearance does not meet the data of data protocol call format, including data mess code with number completion is answered according to for sky;Reject content obvious
Wrong data, including Data writing time is later than current time after a few days earlier than equipment set-up time and Data writing time
Situation etc..After rejecting the data of apparent error, the not clear noise data of some reasons in artificial rejecting or modification normal data,
Because these noise datas easily allow neural network model to learn to excessive information, the accuracy rate of neural network model is caused to drop
It is low.
Data are classified:Complete after data cleansing, data are demarcated, in a last plus list registration evidence of data
The numeral of classification is classified, and the data integration for completing data scaling is training data.In general, training data is more,
The accuracy rate of model is higher.But this trend is not absolute, if training data is excessive, brings the same of loss in efficiency
When, it is easier to cause the excessive non-structured details of model learning, make the reduction of its accuracy rate.
Modeling procedure:Algorithm routine flow chart is shown in Fig. 3, wherein, error is backpropagation, i.e., first by output layer
The output error of each layer neuron is successively calculated, the weights and threshold value of each layer are then adjusted according to error gradient descent method, made
The final output of amended network can be close to desired value;
According to error gradient descent method, shown in output layer weighed value adjusting formula such as formula (1):
Shown in output layer adjusting thresholds formula such as formula (2):
Shown in hidden layer weighed value adjusting formula such as formula (3):
Shown in hidden layer adjusting thresholds formula such as formula (4):
In formula:η is learning rate;TP KIt is desired output;OP KIt is the output after sample input network;netkIt is hidden layer kth
The input of individual node;yiIt is the output of i-th node of hidden layer.
Model realization:The first step carries out data normalization treatment, maps the data into [0,1] or [- 1,1] interval or more
In small interval, prevent neutral net causes convergence rate excessively slow because the unit and scope of input data are inconsistent;
Second step is selected each function, and activation primitive is selected first, and the activation primitive of BP neural network must be place
Place can be micro-, most frequently be the logarithmic function of S types or the tan and linear function of S types, general neutral net
Algorithm model, in hidden layer transfer function S type activation primitives, the transfer function of output layer then selects linear activation primitive;Mould
The S type activation primitives that type is selected are logsig functions, and the value of function input is exported then in (0,1) between (∞, ∞)
On interval.Next to that the selection to training function, using the adjusting learning rate function traingdx of momentum one;
3rd step is the selection to the network number of plies and hidden layer nodes, it is generally the case that the network number of plies is more, nerve net
The error of network is just smaller, is not excluded for increasing with the network number of plies certainly, the situation that the error of neutral net increases on the contrary.If
Increase the quantity of hidden layer simply, often increase the complexity of network, also there is great possibility to occur network
The phenomenon of " over-fitting ".The most frequently used neutral net is usually 3 layer networks, that is, have 1 hidden layer, meanwhile, error will be reduced
Task gives increase hidden layer nodes to undertake, and is so easier to realize.To hidden layer nodes, in BP neural network,
The selection of hidden layer nodes seems and is even more important that hidden layer nodes not only can cause very big to the performance of neural network model
Influence, and " over-fitting " phenomenon when selecting the inappropriate hidden layer nodes also to directly result in training, but with regard to mesh
From the point of view of preceding development, how this selects to explicitly define hidden layer nodes still without the theoretical of set of system and method, and, though
So there are some empirical functions, such as:
M=log2n
In formula:M is hidden layer interstitial content;N is input layer number;, l is output layer nodes:α be 1-10 between
Arbitrary constant;
But, the hidden layer nodes that various computing formula are obtained might have larger difference, so, we follow one
Most basic principle:On the premise of it ensure that precision, pay the utmost attention to take hidden layer nodes as few as possible.It is final true
Determine with formulaDetermine hidden layer nodes, for different training datas, hidden layer nodes are also uncertain
's;
4th step carries out the selection of each function, and the establishment function newff carried using Matlab creates a BP nerve net
Network.Neutral net is after foundation, in addition it is also necessary to set some parameters manually, and these parameters are all to continuously attempt to choose by experiment
, while also there is the reference of some historical experiences, epochs is the maximum train epochs of network, and goal is the minimum of training objective
Error, show represents the gap periods of training result display frequency, and lr is learning rate;
What is finally carried out is model initialization and emulation, the neutral net initialization function init carried using Matlab,
And emulated using sim functions.
Interpretation of result:By constantly adjust and change each parameter and hidden layer nodes after, the final abnormal use of model
Electricity finds that accuracy rate maintains higher level, but simultaneously because the limitation of own hardware condition etc., the test data of selection
Amount is not huge enough, and less compared to for power industry data volume hundreds of millions of daily, these data can not be completely represented
The electricity consumption abnormal behavior of user.Based on the limitation of above force majeure factor etc., the accuracy rate of model is restricted.If by mould
Type is put into real work, after being trained using substantial amounts of power industry user power utilization data, it will largely
Overcome due to the not high enough situation of model accuracy rate caused by data deficiencies.
Conventional abnormal electricity consumption detection method is required for by Multiple-Scan data set, and carries out large amount of complex calculating,
Abnormal judgement can be drawn.And then can by the neutral net exception electricity consumption detection algorithm for having supervision after large-scale dataset training
The whether quick judgement of user power utilization Information abnormity is enough realized, is substantially increased and is found efficiency that power information is abnormal and accurate
Rate, so as to save the resources such as substantial amounts of manpower and materials.After the algorithm model detects the abnormal user of power information, will
Such user is returned, and abnormal user can be submitted to relevant departments' row on-site inspection by analysis department, reduce the field of investigation, be carried
The accuracy rate that abnormal electricity consumption behavior high finds.
Claims (2)
1. a kind of abnormal electricity consumption detection method based on neutral net, the method is based on setting up abnormal electricity consumption detection model, passes through
The operating condition of equipment is diagnosed and analyzed, is judged that whether measuring equipment, in normal operating condition, realizes aid decision
Function, it is characterised in that described method specifically includes following steps:
1) data acquisition, described data are mainly derived from electric energy measurement data, operating condition in electric energy meter and acquisition terminal
The standby class data such as data and logout;
2) data cleansing, the data for using need to be allowed for access model after data cleansing and screening, reject content and do not meet number
According to the data of protocol format requirement, including data mess code and answer number completion according to for it is empty situations such as;The substantially wrong data of content are rejected,
Including Data writing time current time situation after a few days is later than earlier than equipment set-up time and Data writing time;
3) data classification, is completed after data cleansing, and data are demarcated, in the last plus list registration evidence point of data
The numeral of class is classified, and the data integration for completing data scaling is training data;
4) modeling procedure, the developing algorithm model by the way of supervised learning, will input data be referred to as training data, every group
Training data has clearly mark or a result, when algorithm model is set up, sets up a learning process, learning data
Immanent structure reasonably to organize data to obtain result;
The BP neural network of many hidden layers is set up with Matlab.BP neural network algorithm includes two aspects:The propagated forward of signal
With the backpropagation of error.Carried out by from the direction for being input to output when namely calculating reality output, and weights and threshold value
Adjustment is then carried out by the direction from exporting to being input into;
5) model realization:
The first step carries out data normalization treatment, maps the data into [0,1] or [- 1,1] interval or smaller interval, prevents
Only neutral net causes convergence rate excessively slow because the unit and scope of input data are inconsistent;
Second step is selected each function, and activation primitive is selected first, and the activation primitive of BP neural network must everywhere may be used
It is micro-, most frequently be the logarithmic function of S types or the tan and linear function of S types, general neural network algorithm
Model, in hidden layer transfer function S type activation primitives, the transfer function of output layer then selects linear activation primitive;Next to that
Selection to training function, using the adjusting learning rate function traingdx of momentum one;
3rd step is the selection to the network number of plies and hidden layer nodes, on the premise of it ensure that precision, pays the utmost attention to adopt
Exhaust the hidden layer nodes that may lack;Finally determine with formulaHidden layer nodes are determined, for different instructions
Practice data, hidden layer nodes are also uncertain;
4th step carries out the selection of each function, and the establishment function newff carried using Matlab creates a BP neural network;God
Through network in foundation after, in addition it is also necessary to some parameters are set manually, these parameters be all by experiment continuously attempt to choose, while
Also there is the reference of some historical experiences, epochs is the maximum train epochs of network, and goal is the minimal error of training objective,
Show represents the gap periods of training result display frequency, and lr is learning rate;
What is finally carried out is model initialization and emulation, the neutral net initialization function init carried using Matlab, and is made
Emulated with sim functions;
6) interpretation of result, by constantly adjust and change each parameter and hidden layer nodes after, the final abnormal electricity consumption of model
It was found that accuracy rate maintains higher level.
2. the abnormal electricity consumption detection method based on neutral net according to claim 1, it is characterised in that:
Described step 2) in, after rejecting the data of apparent error, some reasons are failed to understand in artificial rejecting or modification normal data
Noise data because these noise datas easily allow neural network model to learn to excessive information, cause neutral net mould
The accuracy rate reduction of type;
Described step 4), error is backpropagation, i.e., the output of each layer neuron is successively calculated first by output layer
Error, then adjusts the weights and threshold value of each layer according to error gradient descent method, enables the final output of amended network
Close to desired value;
According to error gradient descent method, shown in output layer weighed value adjusting formula such as formula (1):
Shown in output layer adjusting thresholds formula such as formula (2):
Shown in hidden layer weighed value adjusting formula such as formula (3):
Shown in hidden layer adjusting thresholds formula such as formula (4):
In formula:η is learning rate;TP KIt is desired output;OP KIt is the output after sample input network;netkSaved for k-th for hidden layer
The input of point;yiIt is the output of i-th node of hidden layer;
Described step 5), in second step, the S type activation primitives of selection are logsig functions, and the value of function input is (one
∞, ∞) between, output is then on (0,1) interval;
Described step 5), in the 3rd step, it is generally the case that the network number of plies is more, and the error of neutral net is just smaller, certainly not
Exclude increasing with the network number of plies, the situation that the error of neutral net increases on the contrary;If increasing the number of hidden layer simply
Amount, often increases the complexity of network, also has great possibility that network can be made the phenomenon of " over-fitting " occur;It is the most frequently used
Neutral net be usually 3 layer networks, that is, have 1 hidden layer, meanwhile, by reduce error task give increase hide node layer
Count to undertake, be so easier to realize;To hidden layer nodes, in BP neural network, the selection of hidden layer nodes seems
It is even more important, hidden layer nodes can not only cause very big influence to the performance of neural network model, and select improper
Hidden layer nodes " over-fitting " phenomenon when can also directly result in training, but just from the point of view of current development, still without one
The theoretical and method of set system explicitly defines hidden layer nodes, and how this selects, and, although there are some empirical functions, than
Such as:
M=log2n
In formula:M is hidden layer interstitial content;N is input layer number;, l is output layer nodes:α be 1-10 between appoint
Meaning constant;
Described step 6) in, due to the limitation of own hardware condition, the amount of test data of selection is not huge enough, compared to electricity
Less for Lixing industry data volume hundreds of millions of daily, these data can not completely represent the electricity consumption abnormal behavior of user;
Based on the limitation of above force majeure factor etc., the accuracy rate of model is restricted;Therefore model should be put into actual work
In work, after being trained using substantial amounts of power industry user power utilization data, it will be largely overcoming due to data not
The not high enough situation of model accuracy rate caused by foot.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611122796.3A CN106779069A (en) | 2016-12-08 | 2016-12-08 | A kind of abnormal electricity consumption detection method based on neutral net |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611122796.3A CN106779069A (en) | 2016-12-08 | 2016-12-08 | A kind of abnormal electricity consumption detection method based on neutral net |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106779069A true CN106779069A (en) | 2017-05-31 |
Family
ID=58877302
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201611122796.3A Pending CN106779069A (en) | 2016-12-08 | 2016-12-08 | A kind of abnormal electricity consumption detection method based on neutral net |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106779069A (en) |
Cited By (42)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107632521A (en) * | 2017-08-30 | 2018-01-26 | 中国矿业大学 | A kind of potentiostat control strategy based on decision tree and neutral net |
CN107730040A (en) * | 2017-09-30 | 2018-02-23 | 国网山东省电力公司电力科学研究院 | Power information system log information comprehensive characteristics extracting method and device based on RBM |
CN107831465A (en) * | 2017-10-31 | 2018-03-23 | 国网黑龙江省电力有限公司电力科学研究院 | A kind of intelligent electric energy meter fault judgment method based on BP neural network |
CN107968299A (en) * | 2017-12-06 | 2018-04-27 | 芯海科技(深圳)股份有限公司 | A kind of breed fish intelligent socket and recognizer based on artificial neural network |
CN107977710A (en) * | 2017-12-21 | 2018-05-01 | 南方电网科学研究院有限责任公司 | Multiplexing electric abnormality data detection method and device |
CN108062616A (en) * | 2017-11-20 | 2018-05-22 | 国网福建省电力有限公司 | A kind of multiplexing electric abnormality detection model with constraints |
CN108099959A (en) * | 2018-01-26 | 2018-06-01 | 山西省自动化研究所 | Foreign body intrusion intellectual monitoring alarm system |
CN108287782A (en) * | 2017-06-05 | 2018-07-17 | 中兴通讯股份有限公司 | A kind of multidimensional data method for detecting abnormality and device |
CN108572648A (en) * | 2018-04-24 | 2018-09-25 | 中南大学 | A kind of automatic driving vehicle power supply multi-source fusion prediction technique and system |
CN109635950A (en) * | 2018-11-30 | 2019-04-16 | 国网上海市电力公司 | The electric energy meter method for monitoring operation states clustered based on genetic algorithm and corporations |
CN109787846A (en) * | 2019-03-27 | 2019-05-21 | 湖北大学 | A kind of 5G network service quality exception monitoring and prediction technique and system |
CN109919193A (en) * | 2019-01-31 | 2019-06-21 | 中国科学院上海光学精密机械研究所 | A kind of intelligent stage division, system and the terminal of big data |
CN110096723A (en) * | 2019-01-22 | 2019-08-06 | 国网山西省电力公司长治供电公司 | High-tension switch cabinet state of insulation analysis method based on O&M detection big data |
CN110263949A (en) * | 2019-06-21 | 2019-09-20 | 安徽智寰科技有限公司 | Merge the data processing method and system of machine mechanism and intelligent algorithm system |
CN110309884A (en) * | 2019-07-05 | 2019-10-08 | 国网四川省电力公司经济技术研究院 | Electricity consumption data anomalous identification system based on ubiquitous electric power Internet of Things net system |
CN110321947A (en) * | 2019-06-27 | 2019-10-11 | 和远智能科技股份有限公司 | A kind of multiplexing electric abnormality pre-judging method based on convolutional neural networks |
CN110363384A (en) * | 2019-06-03 | 2019-10-22 | 杭州电子科技大学 | Exception electric detection method based on depth weighted neural network |
CN110427484A (en) * | 2018-04-26 | 2019-11-08 | 上海意仕腾教育科技有限公司 | A kind of Chinese natural language processing method based on deep learning |
CN110472695A (en) * | 2019-08-22 | 2019-11-19 | 李世鹏 | A kind of the unusual service condition detection and classification method of industrial processes |
CN110490329A (en) * | 2019-07-05 | 2019-11-22 | 国网浙江省电力有限公司电力科学研究院 | A kind of extensive electricity exception data detection method and system based on machine learning |
CN110601764A (en) * | 2019-09-16 | 2019-12-20 | 西南交通大学 | Radio frequency modulation format identification method based on optical assistance |
CN110751264A (en) * | 2019-09-19 | 2020-02-04 | 清华大学 | Electricity consumption mode identification method based on orthogonal self-coding neural network |
CN110750527A (en) * | 2019-10-24 | 2020-02-04 | 南方电网科学研究院有限责任公司 | Data cleaning method for electric power big data |
CN110796243A (en) * | 2019-11-27 | 2020-02-14 | 重庆大学 | Continuous operation monitoring data simulation generation method and device |
CN111080484A (en) * | 2019-12-21 | 2020-04-28 | 国网山东省电力公司泰安供电公司 | Method and device for monitoring abnormal data of power distribution network |
CN111098464A (en) * | 2018-10-25 | 2020-05-05 | 发那科株式会社 | State determination device and method |
CN111126727A (en) * | 2018-10-30 | 2020-05-08 | 广州供电局有限公司 | Electric power charging distributed parallel anomaly detection method |
CN111539492A (en) * | 2020-07-08 | 2020-08-14 | 武汉格蓝若智能技术有限公司 | Abnormal electricity utilization judgment system and method based on reinforcement learning |
CN112098715A (en) * | 2020-09-18 | 2020-12-18 | 宁波智诚祥科技发展有限公司 | Electric energy monitoring and early warning system based on 5G and corrected GCN diagram neural network |
CN112235273A (en) * | 2020-09-30 | 2021-01-15 | 福建奇点时空数字科技有限公司 | Server abnormal behavior monitoring system based on BS structure |
CN112329350A (en) * | 2020-11-17 | 2021-02-05 | 南京航空航天大学 | Airplane lead-acid storage battery abnormity detection semi-supervision method based on isolation |
CN112561383A (en) * | 2020-12-24 | 2021-03-26 | 航天科工网络信息发展有限公司 | Real-time anomaly detection method based on generation countermeasure network |
CN112801388A (en) * | 2021-02-04 | 2021-05-14 | 天津德尔塔科技有限公司 | Power load prediction method and system based on nonlinear time series algorithm |
CN113486926A (en) * | 2021-06-08 | 2021-10-08 | 同济大学 | Automatic change pier equipment anomaly detection system |
CN113640683A (en) * | 2021-08-06 | 2021-11-12 | 江苏金帆电源科技有限公司 | Method for identifying abnormal battery |
CN114034969A (en) * | 2021-10-25 | 2022-02-11 | 浙江万胜智能科技股份有限公司 | Energy controller current loop event generation method based on correction algorithm |
CN114492578A (en) * | 2021-12-24 | 2022-05-13 | 西南交通大学 | Electrical equipment operation data classification method based on BP neural network |
CN114968990A (en) * | 2022-04-12 | 2022-08-30 | 青岛沃柏斯智能实验科技有限公司 | Design method of diagnosis model for influencing factors of experimental data |
CN115170003A (en) * | 2022-09-08 | 2022-10-11 | 国网山东省电力公司兰陵县供电公司 | Electricity stealing monitoring method and system, storage medium and terminal |
CN115426282A (en) * | 2022-07-29 | 2022-12-02 | 苏州浪潮智能科技有限公司 | Voltage abnormality detection method, system, electronic device, and storage medium |
CN116739389A (en) * | 2023-08-14 | 2023-09-12 | 广东创能科技股份有限公司 | Smart city management method and system based on cloud computing |
CN117369425A (en) * | 2023-12-08 | 2024-01-09 | 南昌华翔汽车内外饰件有限公司 | Automobile instrument assembly fault diagnosis method, system, storage medium and computer |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP1764728A1 (en) * | 2004-06-25 | 2007-03-21 | Shin Caterpillar Mitsubishi Ltd. | Data compressing device and method, data analyzing device and method, and data managing system |
CN104459373A (en) * | 2014-11-11 | 2015-03-25 | 广东电网有限责任公司东莞供电局 | Method for calculating node voltage temporary drop magnitudes based on BP neural network |
-
2016
- 2016-12-08 CN CN201611122796.3A patent/CN106779069A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP1764728A1 (en) * | 2004-06-25 | 2007-03-21 | Shin Caterpillar Mitsubishi Ltd. | Data compressing device and method, data analyzing device and method, and data managing system |
CN104459373A (en) * | 2014-11-11 | 2015-03-25 | 广东电网有限责任公司东莞供电局 | Method for calculating node voltage temporary drop magnitudes based on BP neural network |
Non-Patent Citations (1)
Title |
---|
杨光: ""基于神经网络的异常用电检测算法模型研究"", 《供用电》 * |
Cited By (56)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108287782A (en) * | 2017-06-05 | 2018-07-17 | 中兴通讯股份有限公司 | A kind of multidimensional data method for detecting abnormality and device |
CN107632521B (en) * | 2017-08-30 | 2020-09-29 | 中国矿业大学 | Constant potential rectifier control method based on decision tree and neural network |
CN107632521A (en) * | 2017-08-30 | 2018-01-26 | 中国矿业大学 | A kind of potentiostat control strategy based on decision tree and neutral net |
CN107730040A (en) * | 2017-09-30 | 2018-02-23 | 国网山东省电力公司电力科学研究院 | Power information system log information comprehensive characteristics extracting method and device based on RBM |
CN107831465A (en) * | 2017-10-31 | 2018-03-23 | 国网黑龙江省电力有限公司电力科学研究院 | A kind of intelligent electric energy meter fault judgment method based on BP neural network |
CN108062616A (en) * | 2017-11-20 | 2018-05-22 | 国网福建省电力有限公司 | A kind of multiplexing electric abnormality detection model with constraints |
CN108062616B (en) * | 2017-11-20 | 2024-01-23 | 国网福建省电力有限公司 | Electricity consumption anomaly detection system with constraint conditions |
CN107968299A (en) * | 2017-12-06 | 2018-04-27 | 芯海科技(深圳)股份有限公司 | A kind of breed fish intelligent socket and recognizer based on artificial neural network |
CN107977710A (en) * | 2017-12-21 | 2018-05-01 | 南方电网科学研究院有限责任公司 | Multiplexing electric abnormality data detection method and device |
CN107977710B (en) * | 2017-12-21 | 2020-12-25 | 南方电网科学研究院有限责任公司 | Electricity consumption abnormal data detection method and device |
CN108099959A (en) * | 2018-01-26 | 2018-06-01 | 山西省自动化研究所 | Foreign body intrusion intellectual monitoring alarm system |
CN108572648A (en) * | 2018-04-24 | 2018-09-25 | 中南大学 | A kind of automatic driving vehicle power supply multi-source fusion prediction technique and system |
CN108572648B (en) * | 2018-04-24 | 2020-08-25 | 中南大学 | Unmanned vehicle power supply multi-source fusion prediction method and system |
CN110427484A (en) * | 2018-04-26 | 2019-11-08 | 上海意仕腾教育科技有限公司 | A kind of Chinese natural language processing method based on deep learning |
CN111098464A (en) * | 2018-10-25 | 2020-05-05 | 发那科株式会社 | State determination device and method |
US11628609B2 (en) | 2018-10-25 | 2023-04-18 | Fanuc Corporation | State determination device and method |
CN111126727A (en) * | 2018-10-30 | 2020-05-08 | 广州供电局有限公司 | Electric power charging distributed parallel anomaly detection method |
CN111126727B (en) * | 2018-10-30 | 2022-08-02 | 广东电网有限责任公司广州供电局 | Electric power charging distributed parallel anomaly detection method |
CN109635950A (en) * | 2018-11-30 | 2019-04-16 | 国网上海市电力公司 | The electric energy meter method for monitoring operation states clustered based on genetic algorithm and corporations |
CN110096723A (en) * | 2019-01-22 | 2019-08-06 | 国网山西省电力公司长治供电公司 | High-tension switch cabinet state of insulation analysis method based on O&M detection big data |
CN109919193B (en) * | 2019-01-31 | 2023-05-12 | 中国科学院上海光学精密机械研究所 | Intelligent grading method, system and terminal for big data |
CN109919193A (en) * | 2019-01-31 | 2019-06-21 | 中国科学院上海光学精密机械研究所 | A kind of intelligent stage division, system and the terminal of big data |
CN109787846A (en) * | 2019-03-27 | 2019-05-21 | 湖北大学 | A kind of 5G network service quality exception monitoring and prediction technique and system |
CN110363384A (en) * | 2019-06-03 | 2019-10-22 | 杭州电子科技大学 | Exception electric detection method based on depth weighted neural network |
CN110263949B (en) * | 2019-06-21 | 2021-08-31 | 安徽智寰科技有限公司 | Data processing method and system fusing machine mechanism and artificial intelligence algorithm system |
CN110263949A (en) * | 2019-06-21 | 2019-09-20 | 安徽智寰科技有限公司 | Merge the data processing method and system of machine mechanism and intelligent algorithm system |
CN110321947A (en) * | 2019-06-27 | 2019-10-11 | 和远智能科技股份有限公司 | A kind of multiplexing electric abnormality pre-judging method based on convolutional neural networks |
CN110490329A (en) * | 2019-07-05 | 2019-11-22 | 国网浙江省电力有限公司电力科学研究院 | A kind of extensive electricity exception data detection method and system based on machine learning |
CN110309884A (en) * | 2019-07-05 | 2019-10-08 | 国网四川省电力公司经济技术研究院 | Electricity consumption data anomalous identification system based on ubiquitous electric power Internet of Things net system |
CN110472695A (en) * | 2019-08-22 | 2019-11-19 | 李世鹏 | A kind of the unusual service condition detection and classification method of industrial processes |
CN110472695B (en) * | 2019-08-22 | 2022-10-18 | 李世鹏 | Abnormal working condition detection and classification method in industrial production process |
CN110601764A (en) * | 2019-09-16 | 2019-12-20 | 西南交通大学 | Radio frequency modulation format identification method based on optical assistance |
CN110751264A (en) * | 2019-09-19 | 2020-02-04 | 清华大学 | Electricity consumption mode identification method based on orthogonal self-coding neural network |
CN110750527A (en) * | 2019-10-24 | 2020-02-04 | 南方电网科学研究院有限责任公司 | Data cleaning method for electric power big data |
CN110796243A (en) * | 2019-11-27 | 2020-02-14 | 重庆大学 | Continuous operation monitoring data simulation generation method and device |
CN111080484A (en) * | 2019-12-21 | 2020-04-28 | 国网山东省电力公司泰安供电公司 | Method and device for monitoring abnormal data of power distribution network |
CN111539492B (en) * | 2020-07-08 | 2020-11-20 | 武汉格蓝若智能技术有限公司 | Abnormal electricity utilization judgment system and method based on reinforcement learning |
CN111539492A (en) * | 2020-07-08 | 2020-08-14 | 武汉格蓝若智能技术有限公司 | Abnormal electricity utilization judgment system and method based on reinforcement learning |
CN112098715B (en) * | 2020-09-18 | 2023-06-06 | 宁波智诚祥科技发展有限公司 | Electric energy monitoring and early warning system based on 5G and correction GCN graph neural network |
CN112098715A (en) * | 2020-09-18 | 2020-12-18 | 宁波智诚祥科技发展有限公司 | Electric energy monitoring and early warning system based on 5G and corrected GCN diagram neural network |
CN112235273A (en) * | 2020-09-30 | 2021-01-15 | 福建奇点时空数字科技有限公司 | Server abnormal behavior monitoring system based on BS structure |
CN112329350A (en) * | 2020-11-17 | 2021-02-05 | 南京航空航天大学 | Airplane lead-acid storage battery abnormity detection semi-supervision method based on isolation |
CN112561383A (en) * | 2020-12-24 | 2021-03-26 | 航天科工网络信息发展有限公司 | Real-time anomaly detection method based on generation countermeasure network |
CN112801388A (en) * | 2021-02-04 | 2021-05-14 | 天津德尔塔科技有限公司 | Power load prediction method and system based on nonlinear time series algorithm |
CN113486926B (en) * | 2021-06-08 | 2022-09-13 | 同济大学 | Automatic change pier equipment anomaly detection system |
CN113486926A (en) * | 2021-06-08 | 2021-10-08 | 同济大学 | Automatic change pier equipment anomaly detection system |
CN113640683A (en) * | 2021-08-06 | 2021-11-12 | 江苏金帆电源科技有限公司 | Method for identifying abnormal battery |
CN114034969A (en) * | 2021-10-25 | 2022-02-11 | 浙江万胜智能科技股份有限公司 | Energy controller current loop event generation method based on correction algorithm |
CN114492578A (en) * | 2021-12-24 | 2022-05-13 | 西南交通大学 | Electrical equipment operation data classification method based on BP neural network |
CN114968990A (en) * | 2022-04-12 | 2022-08-30 | 青岛沃柏斯智能实验科技有限公司 | Design method of diagnosis model for influencing factors of experimental data |
CN115426282B (en) * | 2022-07-29 | 2023-08-18 | 苏州浪潮智能科技有限公司 | Voltage abnormality detection method, system, electronic device and storage medium |
CN115426282A (en) * | 2022-07-29 | 2022-12-02 | 苏州浪潮智能科技有限公司 | Voltage abnormality detection method, system, electronic device, and storage medium |
CN115170003A (en) * | 2022-09-08 | 2022-10-11 | 国网山东省电力公司兰陵县供电公司 | Electricity stealing monitoring method and system, storage medium and terminal |
CN116739389A (en) * | 2023-08-14 | 2023-09-12 | 广东创能科技股份有限公司 | Smart city management method and system based on cloud computing |
CN117369425A (en) * | 2023-12-08 | 2024-01-09 | 南昌华翔汽车内外饰件有限公司 | Automobile instrument assembly fault diagnosis method, system, storage medium and computer |
CN117369425B (en) * | 2023-12-08 | 2024-02-27 | 南昌华翔汽车内外饰件有限公司 | Automobile instrument assembly fault diagnosis method, system, storage medium and computer |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106779069A (en) | A kind of abnormal electricity consumption detection method based on neutral net | |
CN106650797B (en) | Power distribution network electricity stealing suspicion user intelligent identification method based on integrated ELM | |
CN104407589B (en) | Workshop manufacturing process-oriented active sensing and anomaly analysis method of real-time production performance | |
CN101464964B (en) | Pattern recognition method capable of holding vectorial machine for equipment fault diagnosis | |
CN110361176A (en) | A kind of intelligent failure diagnosis method for sharing neural network based on multitask feature | |
CN108171259A (en) | A kind of wisdom tip water quality monitoring system and method based on NB-IoT | |
CN108228716A (en) | SMOTE_Bagging integrated sewage handling failure diagnostic methods based on weighting extreme learning machine | |
CN108520357A (en) | A kind of method of discrimination, device and the server of line loss abnormal cause | |
CN110455537A (en) | A kind of Method for Bearing Fault Diagnosis and system | |
CN110334764A (en) | Rotating machinery intelligent failure diagnosis method based on integrated depth self-encoding encoder | |
CN105512448A (en) | Power distribution network health index assessment method | |
CN109000930A (en) | A kind of turbogenerator performance degradation assessment method based on stacking denoising self-encoding encoder | |
CN106482938A (en) | Brake fluid system multi-source fusion fault predicting method based on GA BP network | |
CN114676742A (en) | Power grid abnormal electricity utilization detection method based on attention mechanism and residual error network | |
Khuat et al. | Ensemble learning for software fault prediction problem with imbalanced data. | |
CN107295537A (en) | A kind of method and system for wireless sensor network reliability of testing and assessing | |
CN114565038A (en) | Intelligent electric meter fault diagnosis method based on improved capsule network | |
CN110318731A (en) | A kind of oil well fault diagnostic method based on GAN | |
CN110111885A (en) | Attribute forecast method, apparatus, computer equipment and computer readable storage medium | |
CN114429152A (en) | Rolling bearing fault diagnosis method based on dynamic index antagonism self-adaption | |
CN113782113B (en) | Method for identifying gas fault in transformer oil based on deep residual error network | |
CN113344470B (en) | Intelligent power supply system management method based on block chain | |
CN105426665B (en) | Method is determined based on the DYNAMIC RELIABILITY of status monitoring | |
CN106202668B (en) | Complex equipment quality risk appraisal procedure based on quality problems data and reverse conduction neural network | |
CN117036103A (en) | Electric power system operation situation prediction method based on LSTM (least squares) circulating neural network |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170531 |
|
RJ01 | Rejection of invention patent application after publication |