CN110245617A - Artificial intelligence analysis's method based on transient state recording waveform - Google Patents
Artificial intelligence analysis's method based on transient state recording waveform Download PDFInfo
- Publication number
- CN110245617A CN110245617A CN201910522198.2A CN201910522198A CN110245617A CN 110245617 A CN110245617 A CN 110245617A CN 201910522198 A CN201910522198 A CN 201910522198A CN 110245617 A CN110245617 A CN 110245617A
- Authority
- CN
- China
- Prior art keywords
- data
- subset
- sample
- waveform
- feature
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/08—Locating faults in cables, transmission lines, or networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The present invention provides a kind of artificial intelligence analysis's method based on transient state recording waveform, the following steps are included: step S1, there is the transient-wave disturbed in acquisition from power grid environment, after transient state recording collector is triggered, it will record the Wave data and the Wave data in rear several periods in several periods before triggering moment；The waveform data sample of acquisition includes six variables of A, B, C three-phase voltage and A, B, C three-phase current；Step S2 samples the Wave data of acquisition, several data points of each periodic sampling；Wave data after sampling is input in trained intelligent algorithm model by step S3, and intelligent algorithm model provides type of waveform.The present invention improves the waveform recognition efficiency of electric network fault, has saved manpower, reduces the cost of operation and maintenance.
Description
Technical field
The present invention relates to power grid transient state recording waveform field, especially a kind of artificial intelligence based on transient state recording waveform point
Analysis method.
Background technique
Some stable states such as related voltage deviation, frequency departure, stable state harmonic distortion, voltage fluctuation and flicker in power grid at present
Detection there has been common method and practical detection device.But such as voltage sag, voltage swells, periodicity are fallen into
The automatic detection analysis of the electric system waveform transition effect such as wave lacks effective solution method always.Present processing means
Waveform that recording is got off with corresponding normal waveform manually carry out it is point-to-point compared with to judge transient state recording type.
Summary of the invention
It is an object of the present invention to overcome the shortcomings of the prior art and provide a kind of people based on transient state recording waveform
Work intelligent analysis method improves the waveform recognition efficiency of electric network fault.The technical solution adopted by the present invention is that:
A kind of artificial intelligence analysis's method based on transient state recording waveform, comprising the following steps:
Step S1, acquiring the transient-wave for occurring disturbing from power grid environment can remember after transient state recording collector is triggered
The Wave data and the Wave data in rear several periods in several periods before record triggering moment；
The waveform data sample of acquisition includes six variables of A, B, C three-phase voltage and A, B, C three-phase current；
Step S2 samples the Wave data of acquisition, several data points of each periodic sampling；
Wave data after sampling is input in trained intelligent algorithm model by step S3, and artificial intelligence is calculated
Method model provides type of waveform.
Further, the training process of intelligent algorithm model is as follows:
Step 1, waveform data sample, including normal, short-circuit, three kinds of Wave datas of ground connection, every kind of waveform number are collected in advance
It include six variables of A, B, C three-phase voltage and A, B, C three-phase current according to sample；
Step 2, the residual error data for calculating each variable, on the basis of the data point of variable a cycle, the change in sample
Data point in amount subsequent cycle is subtracted each other with corresponding position data point in a cycle, obtains the residual error data of variable；
Step 3, calculate characteristic value: calculate waveform data sample A, B, C three-phase voltage and A, B, C three-phase current six change
The degree of bias, kurtosis, the very poor, mean value, standard deviation of amount, counting residual error data in each variable respectively, to subtract the numerical value after mean value big
In the data point number of 2 times of standard deviations, and greater than the data point number of 3 times of standard deviations；By the calculated degree of bias of each variable,
Kurtosis, very poor, residual error data subtracts the data point number that the numerical value after mean value is greater than 2 times of standard deviations, greater than the number of 3 times of standard deviations
Input feature vector of the strong point number as intelligent algorithm model；
Step 4, the random forest being made of more decision trees is constructed, using comentropy as the standard of feature selecting, to defeated
Enter feature to be differentiated；
The building process of decision tree:
Step 401, the initial information entropy of current sample data set is calculated first, and the initial information entropy before unallocated subset is pressed
It is calculated according to original sample data set D；
Step 402, the comentropy of 6 variables, 30 features of each sample is then calculated；Realization process is to use binary tree,
Current sample data set is divided into two subset DsleftAnd Dright, the comentropy formula of feature:
Wherein N indicates that current sample data concentrates number of samples to mean that sample in D for original sample data set D
This number means that the number of samples in the subset if current sample data set is the subset after a division；NleftExpression is worked as
The a subset D that preceding sample data set separatesleftMiddle number of samples, NrightIndicate another height that current sample data set separates
Collect DrightMiddle number of samples；E () indicates to calculate comentropy；
Divide two subset DsleftAnd DrightMethod be by an input feature vector according to from small to large sequence arrange, according to
The secondary characteristic value for taking this feature is grouped into D less than or equal to this feature valueleftIn subset, D is grouped into greater than this feature valuerightSon
Collection calculates each corresponding I of all possible subset division situations of this feature, takes wherein I the smallest as this feature
Comentropy；
Calculate the comentropy of all features；
Step 403, I is calculatedn-Einit, InThe comentropy for indicating n-th of feature, takes In-EinitIt is worth maximum feature conduct to cut
Current sample data set is divided into two subsets by branch；
Step 404, in sub-portion recursive call step 401~step 403 until subset information entropy is 0 or In-Einit
When less than threshold value, stop continuing dividing subset, model training terminates；
In the subset of the bottom, the largest number of types of sample type just represent the type of subset；
Intelligent algorithm model returns to the type of subset corresponding with input feature vector according to input feature vector.
The present invention has the advantages that the present invention improves the waveform recognition efficiency of electric network fault, manpower is saved, has been reduced
The cost of operation and maintenance decreases erroneous judgement probability.
Detailed description of the invention
Fig. 1 is training process schematic diagram of the invention.
Specific embodiment
Below with reference to specific drawings and examples, the invention will be further described.
Artificial intelligence analysis's method based on transient state recording waveform, comprising the following steps:
Step S1, acquiring the transient-wave for occurring disturbing from power grid environment can remember after transient state recording collector is triggered
The Wave data and the Wave data in rear 8 periods in 4 periods before record triggering moment；
The waveform data sample of acquisition includes six variables of A, B, C three-phase voltage and A, B, C three-phase current；
Step S2 samples the Wave data of acquisition, such as 80 data points of each periodic sampling；
So variable just samples 960 data points in 12 periods；
Wave data after sampling is input in trained intelligent algorithm model by step S3, and artificial intelligence is calculated
Method model provides type of waveform, for example, normally, short circuit or ground connection；
The training process of intelligent algorithm model is as follows:
Step 1, waveform data sample, including normal, short-circuit, three kinds of Wave datas of ground connection, every kind of waveform number are collected in advance
It include six variables of A, B, C three-phase voltage and A, B, C three-phase current according to sample；
Step 2, the residual error data for calculating each variable, on the basis of the data point of variable a cycle, the change in sample
Data point in amount subsequent cycle is subtracted each other with corresponding position data point in a cycle, obtains the residual error data of variable；
By taking A phase voltage as an example, on the basis of taking A phase voltage 80 data points of a cycle, behind the 2nd period to the 12nd
80 data points in a each period in period are subtracted each other with corresponding position data point in a cycle, obtain the residual of A phase voltage
Difference data；
Step 3, calculate characteristic value: calculate waveform data sample A, B, C three-phase voltage and A, B, C three-phase current six change
The degree of bias, kurtosis, the very poor, mean value, standard deviation of amount, counting residual error data in each variable respectively, to subtract the numerical value after mean value big
In the data point number of 2 times of standard deviations, and greater than the data point number of 3 times of standard deviations；By the calculated degree of bias of each variable,
Kurtosis, very poor, residual error data subtracts the data point number that the numerical value after mean value is greater than 2 times of standard deviations, greater than the number of 3 times of standard deviations
Input feature vector of the strong point number as intelligent algorithm model；
Step 4, the random forest being made of more decision trees is constructed, using comentropy as the standard of feature selecting, to defeated
Enter feature to be differentiated；
Comentropy is a kind of measurement most common index of sample set purity；
It is assumed that ratio shared by kth class sample is classified as p in current sample set Dk(k=1,2 ..., n), then the comentropy of D is fixed
Justice is
The value of E (D) is smaller, then the purity of D is higher；
The building process of decision tree:
Step 401, the initial information entropy of current sample data set is calculated first, and the initial information entropy before unallocated subset is pressed
It is calculated according to original sample data set D；It is respectively normal p1, short-circuit p2 that waveform, which is divided into three types, is grounded p3；
Initial information entropy Einit=-p1log2p1-p2log2p2-p3log2p3；
Step 402, the comentropy of 6 variables, 30 features of each sample is then calculated；Realization process is to use binary tree,
Current sample data set is divided into two subset DsleftAnd Dright, the comentropy formula of feature:
Wherein N indicates that current sample data concentrates number of samples to mean that sample in D for original sample data set D
This number means that the number of samples in the subset if current sample data set is the subset after a division；NleftExpression is worked as
The a subset D that preceding sample data set separatesleftMiddle number of samples, NrightIndicate another height that current sample data set separates
Collect DrightMiddle number of samples；E () indicates to calculate comentropy；
Divide two subset DsleftAnd DrightMethod be by an input feature vector according to from small to large sequence arrange, according to
The secondary characteristic value for taking this feature is grouped into D less than or equal to this feature valueleftIn subset, D is grouped into greater than this feature valuerightSon
Collection calculates each corresponding I of all possible subset division situations of this feature, takes wherein I the smallest as this feature
Comentropy；
Calculate the comentropy of all features；
Step 403, I is calculatedn-Einit, InThe comentropy for indicating n-th of feature, takes In-EinitIt is worth maximum feature conduct to cut
Current sample data set is divided into two subsets by branch；
Step 404, in sub-portion recursive call step 401~step 403 until subset information entropy is 0 or In-Einit
When less than threshold value, stop continuing dividing subset, model training terminates；
In the subset of the bottom, the largest number of types of sample type just represent the type of subset.
Intelligent algorithm model returns to the type of subset corresponding with input feature vector according to input feature vector, such as just
Often, short circuit or ground connection.
One example as shown in Figure 1,
1) the corresponding initial information entropy E of original sample data set D is first calculatedinitAnd the comentropy I of each featuren,
In-EinitMaximum difference is " A phase voltage residual error is very poor " this feature, selects this feature for cut-off；
2) being divided into subset D 1 greater than 10 by the value in " A phase voltage residual error is very poor " feature, drawing less than or equal to 10
It assigns in subset D 2；
3) the comentropy E of D1 is calculatedinitAnd each characteristic information entropy I in D1n, In-EinitMaximum difference is " B phase electricity
Press residual error very poor " this feature, select this feature for cut-off；
4) being divided into subset D 3 greater than 8.5 by the value in " B phase voltage residual error is very poor " feature, less than or equal to 8.5
It is divided into subset D 4；
5) division of 2 branch of subset D is similar with D1's；
6) D3, D4, D5 are finally obtained, D6 tetra- meet the subset of information entropy condition, in each subset, sample type number
Most types just represents the type of subset.
It should be noted last that the above specific embodiment is only used to illustrate the technical scheme of the present invention and not to limit it,
Although being described the invention in detail referring to example, those skilled in the art should understand that, it can be to the present invention
Technical solution be modified or replaced equivalently, without departing from the spirit and scope of the technical solution of the present invention, should all cover
In the scope of the claims of the present invention.
Claims (3)
1. a kind of artificial intelligence analysis's method based on transient state recording waveform, which comprises the following steps:
Step S1, acquiring the transient-wave for occurring disturbing from power grid environment will record touching after transient state recording collector is triggered
The Wave data and the Wave data in rear several periods in several periods before the hair moment；
The waveform data sample of acquisition includes six variables of A, B, C three-phase voltage and A, B, C three-phase current；
Step S2 samples the Wave data of acquisition, several data points of each periodic sampling；
Wave data after sampling is input in trained intelligent algorithm model, intelligent algorithm mould by step S3
Type provides type of waveform.
2. artificial intelligence analysis's method as described in claim 1 based on transient state recording waveform, which is characterized in that
The training process of intelligent algorithm model is as follows:
Step 1, waveform data sample, including normal, short-circuit, three kinds of Wave datas of ground connection, every kind of Wave data sample are collected in advance
This includes six variables of A, B, C three-phase voltage and A, B, C three-phase current；
Step 2, the residual error data for calculating each variable, on the basis of the data point of variable a cycle, in sample after the variable
Data point in the continuous period is subtracted each other with corresponding position data point in a cycle, obtains the residual error data of variable；
Step 3, it calculates characteristic value: calculating six variables of A, B, C three-phase voltage and A, B, C three-phase current of waveform data sample
The degree of bias, kurtosis, very poor, mean value, standard deviation, count residual error data in each variable respectively and subtract the numerical value after mean value and be greater than 2
The data point number of times standard deviation, and the data point number greater than 3 times of standard deviations；By the calculated degree of bias of each variable, peak
Degree, very poor, residual error data subtracts the data point number that the numerical value after mean value is greater than 2 times of standard deviations, greater than the data of 3 times of standard deviations
Input feature vector of the point number as intelligent algorithm model；
Step 4, the random forest being made of more decision trees is constructed, it is special to input using comentropy as the standard of feature selecting
Sign is differentiated；
The building process of decision tree:
Step 401, the initial information entropy of current sample data set is calculated first, and the initial information entropy before unallocated subset is according to original
The sample data set D of beginning is calculated；
Step 402, the comentropy of 6 variables, 30 features of each sample is then calculated；Realization process will be worked as with binary tree
Preceding sample data set is divided into two subset DsleftAnd Dright, the comentropy formula of feature:
Wherein N indicates that current sample data concentrates number of samples, for original sample data set D, means that sample in D
Number means that the number of samples in the subset if current sample data set is the subset after a division；NleftIndicate current sample
The a subset D that notebook data collection separatesleftMiddle number of samples, NrightIndicate another subset that current sample data set separates
DrightMiddle number of samples；E () indicates to calculate comentropy；
Divide two subset DsIeftAnd DrightMethod be by an input feature vector according to from small to large sequence arrange, successively take
The characteristic value of this feature is grouped into D less than or equal to this feature valueleftIn subset, D is grouped into greater than this feature valuerightSubset, meter
The each corresponding I for calculating all possible subset division situations of this feature, takes the wherein the smallest information as this feature of I
Entropy；
Calculate the comentropy of all features；
Step 403, I is calculatedn-Einit,InThe comentropy for indicating n-th of feature, takes In-EinitIt is worth maximum feature as cut-off
Current sample data set is divided into two subsets；
Step 404, in sub-portion recursive call step 401~step 403 until subset information entropy is 0 or In-EinitIt is less than
When threshold value, stop continuing dividing subset, model training terminates；
In the subset of the bottom, the largest number of types of sample type just represent the type of subset；
Intelligent algorithm model returns to the type of subset corresponding with input feature vector according to input feature vector.
3. artificial intelligence analysis's method as described in claim 1 based on transient state recording waveform, which is characterized in that
After transient state recording collector is triggered, the Wave data and the wave in rear 8 periods in 4 periods before triggering moment will record
Graphic data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910522198.2A CN110245617B (en) | 2019-06-17 | 2019-06-17 | Artificial intelligence analysis method based on transient recording waveform |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910522198.2A CN110245617B (en) | 2019-06-17 | 2019-06-17 | Artificial intelligence analysis method based on transient recording waveform |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110245617A true CN110245617A (en) | 2019-09-17 |
| CN110245617B CN110245617B (en) | 2022-06-24 |
Family
ID=67887546
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910522198.2A Active CN110245617B (en) | 2019-06-17 | 2019-06-17 | Artificial intelligence analysis method based on transient recording waveform |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110245617B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113033837A (en) * | 2021-03-05 | 2021-06-25 | 国网电力科学研究院武汉南瑞有限责任公司 | Artificial intelligence fault identification system and method based on transient waveform of power transmission line |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105067963A (en) * | 2015-09-24 | 2015-11-18 | 广东电网有限责任公司佛山供电局 | Distribution network fault early warning method and system based on transient waveforms |
| CN107167702A (en) * | 2017-05-04 | 2017-09-15 | 国网福建省电力有限公司 | A kind of distribution feeder fault type recognition method and device |
| CN107329040A (en) * | 2017-06-16 | 2017-11-07 | 国电南瑞科技股份有限公司 | A kind of power distribution automation main station system single-phase earth fault localization method based on transient state recorder data |
| CN107609569A (en) * | 2017-07-31 | 2018-01-19 | 北京映翰通网络技术股份有限公司 | A kind of distribution net work earthing fault localization method based on multidimensional characteristic vectors |
| CN109002762A (en) * | 2018-06-14 | 2018-12-14 | 南方电网科学研究院有限责任公司 | Method and system for identifying fault current waveform of switching-off and switching-on coil |
-
2019
- 2019-06-17 CN CN201910522198.2A patent/CN110245617B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105067963A (en) * | 2015-09-24 | 2015-11-18 | 广东电网有限责任公司佛山供电局 | Distribution network fault early warning method and system based on transient waveforms |
| CN107167702A (en) * | 2017-05-04 | 2017-09-15 | 国网福建省电力有限公司 | A kind of distribution feeder fault type recognition method and device |
| CN107329040A (en) * | 2017-06-16 | 2017-11-07 | 国电南瑞科技股份有限公司 | A kind of power distribution automation main station system single-phase earth fault localization method based on transient state recorder data |
| CN107609569A (en) * | 2017-07-31 | 2018-01-19 | 北京映翰通网络技术股份有限公司 | A kind of distribution net work earthing fault localization method based on multidimensional characteristic vectors |
| CN109002762A (en) * | 2018-06-14 | 2018-12-14 | 南方电网科学研究院有限责任公司 | Method and system for identifying fault current waveform of switching-off and switching-on coil |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113033837A (en) * | 2021-03-05 | 2021-06-25 | 国网电力科学研究院武汉南瑞有限责任公司 | Artificial intelligence fault identification system and method based on transient waveform of power transmission line |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110245617B (en) | 2022-06-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110263172A (en) | A kind of evented autonomous classification method of power system monitor warning information | |
| CN109660206B (en) | Wasserstein GAN-based photovoltaic array fault diagnosis method | |
| CN110336289A (en) | A kind of monitoring recognition methods, device, equipment and the storage medium of power load | |
| CN113887912B (en) | Non-invasive load identification method for embedded equipment under deep learning | |
| CN111371188A (en) | Real-time early warning system for abnormal power consumption data | |
| CN114325256A (en) | Power equipment partial discharge identification method, system, equipment and storage medium | |
| CN110245617A (en) | Artificial intelligence analysis's method based on transient state recording waveform | |
| CN111044845A (en) | Power distribution network accident identification method and system based on Apriori algorithm | |
| CN117277566B (en) | Power grid data analysis power dispatching system and method based on big data | |
| CN105445577A (en) | Power quality disturbance source working condition identifying method | |
| CN115808600B (en) | DC arc detection method based on mathematical morphology and pattern recognition | |
| CN111475548A (en) | Power utilization abnormity analysis decision system based on big data mining technology | |
| CN116664870A (en) | Power quality disturbance identification method based on transfer learning | |
| US10120953B2 (en) | System data compression system and method thereof | |
| CN111103490B (en) | Load identification event detection method, device and equipment | |
| CN108037541A (en) | D.C. high voltage transmission judges optimized treatment method and system certainly to geomagnetic observation interference | |
| CN113985266A (en) | Centralized load identification method for multiple intelligent electric meters | |
| CN112614005A (en) | Enterprise rework state processing method and device | |
| CN111177502A (en) | Decision tree algorithm-based power communication network redundant data identification method and system | |
| CN111143296A (en) | Substation SCD file classification method and device, terminal and storage medium | |
| CN113899944B (en) | Detection method and device for power load switching point | |
| CN113569904B (en) | Bus wiring type identification method, system, storage medium and computing device | |
| CN113484662A (en) | Fault indicator starting wave recording method | |
| CN114579827B (en) | Method, device and equipment for processing data performance curve of industrial equipment | |
| CN113033607B (en) | Non-invasive load monitoring event extraction method and storage medium |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |