CN109270407A - Fault cause identification method for UHV DC transmission line based on multi-source information fusion - Google Patents

Fault cause identification method for UHV DC transmission line based on multi-source information fusion Download PDF

Info

Publication number
CN109270407A
CN109270407A CN201811366886.6A CN201811366886A CN109270407A CN 109270407 A CN109270407 A CN 109270407A CN 201811366886 A CN201811366886 A CN 201811366886A CN 109270407 A CN109270407 A CN 109270407A
Authority
CN
China
Prior art keywords
fault
output
layer
uhv
neural network
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
Application number
CN201811366886.6A
Other languages
Chinese (zh)
Other versions
CN109270407B (en
Inventor
李玉敦
李宽
施雨
苏欣
张繁斌
王志远
刘萌
赵斌超
张婉婕
杨超
王昕�
麻常辉
张国辉
王永波
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.)
Electric Power Research Institute of State Grid Shandong Electric Power Co Ltd
Original Assignee
Electric Power Research Institute of State Grid Shandong 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 Electric Power Research Institute of State Grid Shandong Electric Power Co Ltd filed Critical Electric Power Research Institute of State Grid Shandong Electric Power Co Ltd
Priority to CN201811366886.6A priority Critical patent/CN109270407B/en
Publication of CN109270407A publication Critical patent/CN109270407A/en
Application granted granted Critical
Publication of CN109270407B publication Critical patent/CN109270407B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G01R31/088—Aspects of digital computing

Landscapes

  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Supply And Distribution Of Alternating Current (AREA)

Abstract

本发明涉及一种基于多源信息融合的特高压直流输电线路故障原因辨识方法，其包括：获取特高压直流输电线路的电气量故障数据和非电气量信息；提取特高压直流输电线路电气特征量，构造电气特征输入向量；提取特高压直流输电线路非电气特征量，构造非电气特征输入向量；对雷击、山火、污秽、风偏、鸟害分别构建综合神经网络辨识模型，神经网络模型的隐含层的层数采用最小误差的方法选取；采用自学习的方法进行具体故障原因辨识。本发明综合电气量信息和非电气量信息，寻找特征规律，利用神经网络本身的融合性将多源信息的进行融合，利用大数据方法的思想，利用多源故障信息，结合神经网络算法，进行原因的辨识，充分保证了原因辨识的准确性。The invention relates to a fault cause identification method for UHV DC transmission lines based on multi-source information fusion, which comprises: obtaining electrical quantity fault data and non-electrical quantity information of UHV DC transmission lines; and extracting electrical characteristic quantities of UHVDC transmission lines , construct electrical characteristic input vector; extract non-electrical characteristic quantity of UHV DC transmission line, construct non-electrical characteristic input vector; construct comprehensive neural network identification model for lightning strike, mountain fire, pollution, wind deviation and bird damage, neural network model The number of layers in the hidden layer is selected by the method of minimum error; the self-learning method is used to identify the specific fault cause. The invention integrates the electric quantity information and the non-electric quantity information, searches for the characteristic law, fuses the multi-source information by using the fusion of the neural network itself, utilizes the idea of the big data method, utilizes the multi-source fault information, and combines the neural network algorithm to perform The identification of the reason fully guarantees the accuracy of the cause identification.

Description

Extra high voltage direct current transmission line failure cause discrimination method based on Multi-source Information Fusion
Technical field
The present invention relates to ultra-high-tension power transmission line O&M field, specifically a kind of extra-high voltage based on Multi-source Information Fusion Direct current transmission line fault reason discrimination method.
Background technique
Extra high voltage direct current transmission line length is located at the locations such as mountain area hills up to thousands of kilometers mostly, is easy by each The influences such as kind natural calamity, external force destruction are broken down, and timely the Research on Identification of failure cause can instruct line walking, add Fast maintenance and line powering restore.
Transmission line malfunction mainly includes the failure as caused by the reasons such as lightning stroke, mountain fire, filth, bird pest and windage yaw.To event The identification for hindering cause needs premised on the understanding to various failure principles and process, on this basis to the spy of specific fault Sign carries out mining analysis, so as to form the foundation of reason identification.
Wherein, fault signature includes the fault signature of electrical quantity and the fault signature of non-electric quantity, the electricity under different reasons Tolerance feature is there is similitude, and there is also distinctive factor, relying solely on a kind of factor can not accurately know non-electric quantity Not.There is presently no the methods for the identification of extra high voltage direct current transmission line failure cause of complete set, send out in transmission line of electricity After raw failure, the auxiliary in decision can not be provided to operating maintenance personnel.It is a kind of based on electrical quantity and non-it is therefore desirable to provide The extra high voltage direct current transmission line failure cause discrimination method of the Multi-source Information Fusions such as electrical quantity.
Summary of the invention
The technical problem to be solved in the present invention is to provide a kind of extra high voltage direct current transmission lines based on Multi-source Information Fusion Failure cause discrimination method, this method combination electric quantity information and non-electric quantity information find characteristic rule, are subsequent nerve The foundation of network class model provides data source, meanwhile, the present invention construct the neural network of failure cause identification structure and Above multi-source information merge by whole cause identification framework using the amalgamation of neural network itself, utilizes big number According to the thought of method, the identification of reason is carried out, has fully ensured that reason in conjunction with neural network algorithm using multi-source fault message The accuracy of identification.
In order to solve the above technical problems, the extra high voltage direct current transmission line failure of the invention based on Multi-source Information Fusion is former Because of discrimination method, include the following steps:
The electrical quantity fault data and non-electric quantity information of step 1) acquisition extra high voltage direct current transmission line；
Step 2) extracts extra high voltage direct current transmission line electric characteristic amount, constructs electric characteristic input vector；
Step 3) extracts extra high voltage direct current transmission line non-electrical characteristic quantity, constructs non-electrical feature input vector；
Step 4) constructs global neurological network identification model, nerve net to lightning stroke, mountain fire, filth, windage yaw, bird pest respectively The number of plies of the hidden layer of network model is chosen using the method for minimal error；
Step 5) carries out specific failure cause identification using the method for self study.
In step 1), the electrical quantity fault data of extra high voltage direct current transmission line is obtained from protection information system, non- Electric quantity information is obtained from meteorological system, lightning location system, GIS-Geographic Information System, forest fire monitoring system；When the event of generation Barrier seldom, when the lazy weight of sample, then obtains data by PSACD simulated fault.
In step 2), if it is t that the moment, which occurs, for failure1, extract the data after extra high voltage direct current transmission line breaks down As electric characteristic amount, electric characteristic amount includes fault current maximum amplitude, the feature of voltage and current amplitude frequency diagram, transition resistance Mean value, transition resistance standard deviation；
Fault current maximum amplitude is | I |；Wherein electric current I is the value of fault current, is positive or is negative；
The feature of voltage and current amplitude frequency diagram needs to carry out discrete fourier to data to decompose to obtain, according to formula
To Current Voltage Fourier transformation, the waveform of frequency domain is obtained, and then extracts and obtains the feature of voltage and current amplitude frequency diagram Amount；
According to formula
Obtain transition resistance mean value；
According to formula
Transition resistance standard deviation E；Just electric characteristic input vector has been obtained.
In step 3), by non-electrical characteristic quantity be divided into weather, the period, season, landform, lightening activity frequency, temperature, 9 humidity, wind-force and historical failure information characteristic quantities, are indicated by orderly binary number, establish the input of non-electrical feature Vector.
In step 4), using three layers of BP neural network structure, input layer number is 33, and hidden layer is missed using minimum The method of difference is chosen, and output layer number of nodes is 5；The activation primitive of hidden layer and output layer is all made of log-sigmoid type letter Number, can be by output control within the scope of 0-1；Training method uses trainlm (Levenberg-Marquardt algorithm), Training objective error is set as 1e-4, and frequency of training is set as 1000 times.
BP neural network is according to the multilayer feedforward neural network of error backpropagation algorithm training, if input layer includes N A neuron, output layer include M neuron, and input vector X, output vector Y, hidden layer includes L neuron section Point, wijIndicate input layer to the connection weight between hidden layer, wjkIndicate hidden layer to the connection weight between output layer, θj Indicate the threshold value of hidden layer neuron, θkIndicate the threshold value of output layer neuron,Indicate the activation primitive of hidden layer, φ The activation primitive of () expression output layer.
The logarithmic function of S type are as follows:
The output signal o of j-th of neuron node of hidden layerjAre as follows:
The output signal y of k-th of neuron node of output layerkAre as follows:
Equipped with P training sample, for each training sample, the quadratic form criterion function of error are as follows:
Global error function of the network to P training sample are as follows:
WhereinWithRespectively indicate input training sample be p when, the expectation of k-th of neuron node of output layer Output and reality output.
It is to have difference between practical identification result and ideal output, with overall Identification Errors EbTo characterize this species diversity:
Ebi=Rsi-Rqi (11)
Wherein i=1~5 respectively indicate lightning stroke, mountain fire, filth, windage yaw, bird pest；Rsi、RqiRespectively indicate such reason Under reality output and desired output, EbiFor the reality output under such reason and the difference between desired output；
The range of a hidden layer is determined according to the empirical equation of hidden layer:
In formula, m and n points of ratios are the number of nodes for outputting and inputting layer, integer of a between 0-10；
The determination step of best hidden layer are as follows:
Step 4.1) calculates the value range of l, and the range of l is (6,16)；
Step 4.2) the implicit number of plies possible for each in l calculates Ebi、Eb；
Step 4.3) compares overall Identification Errors E corresponding to each possible implicit number of plies in lb, take the smallest EbThe corresponding implicit number of plies is best hidden layer.It should be noted that for different training samples, the number of best hidden layer Amount is not necessarily identical.
In step 5), when failure occurs, failure cause is differentiated using the network model of creation, troubleshooting knot Shu Hou, using the failure of generation as standard failure type typing historical failure data library, and by actual failure cause and model Differentiate that result compares, whether comparing result is as modifying the foundation in neural metwork training library；As overall Identification Errors Eb> When 0.5, the sample set in actual failure cause model training library is modified, the actual characteristic data that the secondary failure is added carry out weight New training, obtains revised neural network prediction model, so as to autonomous learning, adapts to the variation of environment, more accurately Identification of defective reason；If differentiating result and actual result deviation within the allowable range, do not need to modify in trained library Sample, but record the fault signature and failure cause.
The step of neural metwork training are as follows:
The formation of step 5.1) training sample, according to the electric characteristic amount and non-electrical feature in step 2) and step 3) The extraction of amount, constructs comprehensive input feature value, and input vector is input 1~input 33, the failure cause according to known to sample Type structure output vector, and then input vector and output vector are mapped composing training sample；
Step 5.2) initially takes l=6, and input layer, hidden layer, output layer are successively passed through in the forward-propagating of signal；
Step 5.3) calculate output error, if target error meets setting value 1e-3, complete training output as a result, Otherwise the 4th step is carried out；
The backpropagation of step 5.4) signal successively corrects Feedback error according to additional guide vanes trainlm Each layer of weight and threshold value；
Step 5.5) repeats step 5.2), step 5.3), step 5.4) according to new weight and threshold value, until error is full Sufficient setting value；Judge whether l takes, step 5.6) is carried out if taking, otherwise l+1, return step 5.2)；
Step 5.6) compares corresponding to each possible implicit number of plies in l according to the determination method of best hidden layer Overall Identification Errors Eb, take the smallest EbThe corresponding implicit number of plies is best hidden layer；
Step 5.7) carries out real case test, and after best hidden layer determines, the structure of neural network is just determined, will be real The fault condition on border is input in neural network, the result recognized；
Step 5.8) uses self-learning algorithm, and every time after identification, actual failure cause and prediction model are differentiated result It compares, whether comparing result is as modifying the foundation in neural metwork training library；If differentiating that result is deviated, The sample set for modifying actual failure cause model training library, the actual characteristic data that the secondary failure is added carry out re -training, Obtain revised neural network prediction model.
The beneficial effects of the present invention are: the generation of failure is related with the running environment of transmission line of electricity, and variety classes failure Feature show difference on Wave data, therefore to line fault occurs for this method the moment on the basis of principle analysis The non-electric quantities factors such as weather condition, the landform in period, season and scene, historical data and failure pole tension, The internal factor of the electrical measure feature such as electric current, transition resistance, frequency domain distribution is excavated, its characteristic rule is found, thus after being The foundation of continuous neural network classification model provides data source.Meanwhile the present invention constructs the neural network of failure cause identification Structure and whole cause identification framework, this method using neural network itself amalgamation by above multi-source information into Row fusion, using the thought of big data method, carries out distinguishing for reason in conjunction with neural network algorithm using multi-source fault message Know, has fully ensured that the accuracy of reason identification.
Detailed description of the invention
Invention is further described in detail with reference to the accompanying drawings and detailed description:
Fig. 1 is the false voltage amplitude frequency diagram after discrete Fourier transform, time domain is transformed into frequency, before containing The electric fault characteristic quantity stated: voltage wave cut off, voltage wave direct current amplitude；
Fig. 2 is the fault current amplitude frequency diagram after discrete Fourier transform, time domain is transformed into frequency, before containing The electric fault characteristic quantity stated: current wave cut off, current wave direct current amplitude；
Fig. 3 is global neurological network identification illustraton of model, totally 33 input vectors, and wherein 1-7 is electrical quantity characteristic value, 8- 33 be non-electric quantity characteristic value, and output result is one of lightning stroke, mountain fire, filth, windage yaw, bird pest；
Fig. 4 is neural network model figure, structure 33-L-5；
Fig. 5 is Neural Network Self-learning flow chart.
Specific embodiment
Referring to attached drawing, the extra high voltage direct current transmission line failure cause identification side of the invention based on Multi-source Information Fusion Method includes the following steps:
The electrical quantity fault data and non-electric quantity information of step 1) acquisition extra high voltage direct current transmission line；
Step 2) extracts extra high voltage direct current transmission line electric characteristic amount, constructs electric characteristic input vector；
Step 3) extracts extra high voltage direct current transmission line non-electrical characteristic quantity, constructs non-electrical feature input vector；
Step 4) constructs global neurological network identification model, nerve net to lightning stroke, mountain fire, filth, windage yaw, bird pest respectively The number of plies of the hidden layer of network model is chosen using the method for minimal error；
Step 5) carries out specific failure cause identification using the method for self study.
In step 1), the fault data of extra high voltage direct current transmission line electrical quantity is obtained from protection information system, non-electrical The information of tolerance is obtained from meteorological system, lightning location system, GIS-Geographic Information System, forest fire monitoring system etc..When generation Failure is seldom, when the lazy weight of sample, then obtains data by PSACD simulated fault.
In step 2), occurs for failure for t the moment1, this method extract extra high voltage direct current transmission line break down after number According to the mould of the maximum amplitude including fault current, the feature of voltage and current amplitude frequency diagram, transition resistance mean value, transition resistance standard Difference is used as electrical quantity fault eigenvalue.The feature of voltage and current amplitude frequency diagram needs to carry out discrete fourier to data to decompose to obtain.
Fault current maximum amplitude is | I |；Wherein electric current I is the value of fault current, and can be positive can be negative.
DC current sampled value i (n), n=0,1 after taking failure, 2 ... .N-1.Since current sampling data is discrete Value, can be broken down into n times harmonic wave with discrete Fourier transform (DFT).
Discrete Fourier transform principle:
K=0 in formula, 1,2 ... N-1 are overtone order, and N is the total number of sample points of a cycle.
Its fundametal compoment is calculated as follows:
Its DC component is calculated as follows:
Similarly, can also discrete Fourier transform be carried out to voltage.
For this method, voltage, electric current are a DC quantity before failure occurs, and failure has a lot humorous after occurring Wave component is distributed within a very wide frequency range, by discrete Fourier transform (DFT) by time domain after failure Waveform is transformed into frequency domain and is analyzed, to obtain the characteristic quantity under certain frequency domains.
This method illustrates the method by extracting corresponding failure characteristic value to voltage, electric current amplitude frequency diagram for being struck by lightning.
Fig. 1 and Fig. 2 is respectively the false voltage and fault current progress discrete Fourier transform after being struck by lightning to route Voltage, electric current amplitude frequency diagram afterwards.
As seen in Figure 1, the content of DC component is maximum, and harmonic wave focuses mostly within 1kHZ, humorous higher than 1kHZ Wave is seldom.This method is using 1KHZ as a characteristic value, and referred to as " voltage wave cut off ", while the amplitude of flip-flop is also made For a characteristic value.
As seen in Figure 2, the content of DC component is maximum, and harmonic wave focuses mostly within 10HZ, humorous higher than 10HZ Wave is seldom.This method is using 10HZ as a characteristic value, and referred to as " current wave cut off ", while the amplitude of flip-flop is also made For a characteristic value.
Using the average R of transition resistance instantaneous value after failure, reflect the resistance value size characteristic of transition resistance, wherein R Are as follows:
Transition resistance standard deviation E are as follows:
Wherein i=1,2,3....N, N are total number of sample points.
Therefore electric characteristic vector X1-X7 has just been obtained.
In step 3), the present invention by non-electric quantity be divided into weather, the period, season, landform, lightening activity frequency, temperature, 9 humidity, wind-force and historical failure information characteristic quantities, are indicated by orderly binary number, establish the feature of non-electric quantity Input vector, input vector are input 8- input 33.
Construction method is as follows:
1, the characteristic quantity input model of weather condition
Weather condition is set as four kinds of situations, is respectively as follows: fine day, cloudy day, misty rain, thunderstorm.Lightning stroke mostly occurs in thunderstorm Weather, this method indicates this four characteristic quantities with 4 bits, for example, using vector
[1 00 0] indicate fine day.Characteristic quantity input model are as follows:
Input 8 Input 9 Input 10 Input 11
Fine day Cloudy day Misty rain Thunderstorm Corresponding types
1 0 0 0 Fine day
0 1 0 0 Cloudy day
0 0 1 0 Misty rain
0 0 0 1 Thunderstorm
2, period characteristic quantity input model
This method indicates this four characteristic quantities with 4 bits, and early morning (when -8 when 4), daytime are (- 17 when 9 When), at dusk (when -21 when 18), midnight (when -3 when 22).For example, early morning is indicated with vector [1 00 0], characteristic quantity input Model are as follows:
Input 12 Input 13 Input 14 Input 15
Early morning Daytime At dusk Midnight Corresponding types
1 0 0 0 Early morning
0 1 0 0 Daytime
0 0 1 0 At dusk
0 0 0 1 Midnight
3, seasonal characteristic quantity input model
This method indicates spring, summer, autumn, this four characteristic quantities of winter with 4 bits, for example, using vector
[1 00 0] indicate spring.Characteristic quantity input model are as follows:
Input 16 Input 17 Input 18 Input 19
Spring Summer Autumn Winter Corresponding types
1 0 0 0 Spring
0 1 0 0 Summer
0 0 1 0 Autumn
0 0 0 1 Winter
4, features of terrain
Features of terrain is divided into Plain, hills, mountain area three types by this method, is indicated with 3 bits These three characteristic quantities, for example, indicating Plain with vector [1 0 0].Characteristic quantity input model are as follows:
Input 20 Input 21 Input 22
Plain Hills Mountain area Corresponding types
1 0 0 Plain
0 1 0 Hills
0 0 1 Mountain area
5, lightening activity frequency (thunderstorm day)
It is considered as area with less lightning activities that thunderstorm day Td, which is equal to 15 areas below, and the area more than 40 is area with more lightning activities, more than 90 Area is special strong minefield.This method indicates these three characteristic quantities with 3 bits, for example, with vector [1 0 0] area with less lightning activities is indicated.Characteristic quantity input model are as follows:
Input 23 Input 24 Input 25
Area with less lightning activities Area with more lightning activities Strong minefield Corresponding types
1 0 0 Area with less lightning activities
0 1 0 Area with more lightning activities
0 0 1 Strong minefield
6, temperature, humidity, wind-force
Input 26 Input 27 Input 28
Temperature Humidity Wind-force
X1 X2 X3
The unit of temperature is degree Celsius that humidity is percentage, and wind-force is 0-17 grades, and required parameter is obtained from meteorological system ?.
7, historical failure information
Historical failure information is divided into 5 classes according to fault type by this method: lightning stroke, mountain fire, filth, windage yaw, bird pest. Need to record be break down place ownership certain power supply company restriction occurred in history such failure time Number, is denoted as N1~N5, represents number.For example, somewhere occurred lightning fault 2 times in history, then N1=2 is remembered.
Input 29 Input 30 Input 31 Input 32 Input 33
Lightning stroke Mountain fire It is filthy Windage yaw Bird pest
N1 N2 N3 N4 N5
Step (4), the building of the global neurologicals network identification model such as lightning stroke, mountain fire, filth, windage yaw, bird pest, synthesis are distinguished Know model such as Fig. 3.This method uses three layers of BP neural network structure, and input layer number is 33, and hidden layer uses minimal error Method choose.Output layer number of nodes is 5.Neural network structure such as Fig. 4.The activation primitive of hidden layer and output layer is all made of Log-sigmoid type function (logarithmic function of S type), can be by output control within the scope of 0-1.Training method uses Trainlm (Levenberg-Marquardt algorithm), training objective error are set as 1e-3, and maximum frequency of training is set as 1000 times.
BP neural network is according to the multilayer feedforward neural network of error backpropagation algorithm training, such as Fig. 4, input layer Comprising N number of neuron, output layer includes M neuron, and input vector X, output vector Y, hidden layer includes L nerve First node, wijIndicate input layer to the connection weight between hidden layer, wjkIndicate hidden layer to the connection weight between output layer Value, θjIndicate the threshold value of hidden layer neuron, θkIndicate the threshold value of output layer neuron,Indicate the activation letter of hidden layer Number, φ () indicate the activation primitive of output layer.
The logarithmic function of S type are as follows:
The output signal o of j-th of neuron node of hidden layerjAre as follows:
The output signal y of k-th of neuron node of output layerkAre as follows:
Equipped with P training sample, for each training sample, the quadratic form criterion function of error are as follows:
Global error function of the network to P training sample are as follows:
WhereinWithRespectively indicate input training sample be p when, the expectation of k-th of neuron node of output layer is defeated Out and reality output.
Wherein, the number of plies of the hidden layer of neural network model is chosen using the method for minimal error.
The desired output Y of neural network are as follows:
Lightning stroke Mountain fire It is filthy Windage yaw Bird pest
0 or 1 0 or 1 0 or 1 0 or 1 0 or 1
Such as: when lightning stroke occurs, desired output is Y=[1 000 0], and when mountain fire occurs, desired output is Y=[0 1 0 0 0]。
But actually obtain value of the result between 0-1, such as certain identification result are as follows: [0.9 0.1 00 0].Institute To be differentiated between practical identification result and ideal output.With overall Identification Errors EbTo characterize this species diversity:
Ebi=Rsi-Rqi (13)
Wherein i=1~5 respectively indicate lightning stroke, mountain fire, filth, windage yaw, bird pest.Rsi、RqiRespectively indicate such reason Under reality output and desired output, EbiFor the reality output under such reason and the difference between desired output.
The range of a hidden layer is determined according to the empirical equation of hidden layer first:
In formula, m and n points of ratios are the number of nodes for outputting and inputting layer, integer of a between 0-10.Best hidden layer is really It is as follows to determine method:
The first step calculates the value range of l, and for this method, l range is (6,16).
Second step, the implicit number of plies possible for each in l calculate Ebi、Eb。
Third step compares overall Identification Errors E corresponding to each possible implicit number of plies in lb, take the smallest Eb The corresponding implicit number of plies is best hidden layer.It should be noted that for different training samples, the quantity of best hidden layer It is not necessarily identical.
In step 5), the identification of failure cause is carried out using the method for self study, it is specific: when an error occurs, to utilize The network model of creation differentiates failure cause, after troubleshooting, needs the failure that will occur as standard failure Type typing historical failure data library, and actual failure cause and Model checking result are compared, comparing result conduct Whether the foundation in neural metwork training library is modified.If differentiating result (using the overall Identification Errors E in step 5bTo characterize) It is deviated, this method, which takes, works as EbWhen > 0.5, need to modify the sample set in actual failure cause model training library, being added should The actual characteristic data of secondary failure carry out re -training, revised neural network prediction model are obtained, so as to independently learn It practises, adapts to the variation of environment, more accurate identification of defective reason.If differentiating that result and actual result deviation are allowing model In enclosing, does not then need to modify the sample in trained library, but record the fault signature and failure cause.Specific self study process Such as Fig. 5.
Steps are as follows for neural metwork training:
Step 1: the formation of training sample, according to the electric characteristic amount and non-electrical characteristic quantity in step 2) and step 3) Extraction, construct comprehensive input feature value, input vector is input 1~input 33.The failure cause class according to known to sample Type constructs output vector, and then input vector and output vector are mapped composing training sample.
Step 2: initially taking l=6, input layer, hidden layer, output layer are successively passed through in the forward-propagating of signal.
Step 3: calculating output error, if target error meets setting value 1e-3, training output is completed as a result, no Then carry out the 4th step.
Step 4: the backpropagation of signal successively corrects Feedback error every according to additional guide vanes trainlm One layer of weight and threshold value.
Step 5: repeating second step, third step, the 4th step according to new weight and threshold value, set until error meets Value.Whether l takes, and the 6th step is carried out if taking, otherwise l+1, returns to second step.
Step 6: each compared in l is possible implicit according to the determination method of the best hidden layer in step (5) Overall Identification Errors E corresponding to the number of pliesb, take the smallest EbThe corresponding implicit number of plies is best hidden layer.
Step 7: real case is tested.After best hidden layer determines, the structure of neural network is just determined, will actual event Barrier situation is input in neural network, the result that can be recognized.
Step 8: every time after identification, actual failure cause and prediction model are differentiated result using self-learning algorithm It compares, whether comparing result is as modifying the foundation in neural metwork training library.If differentiating that result is deviated, The sample set for modifying actual failure cause model training library, the actual characteristic data that the secondary failure is added carry out re -training, Obtain revised neural network prediction model.
This method is illustrated with the case being struck by lightning: according to method and step above-mentioned, by certain to lightning stroke sample training After obtained a neural network, with this trained neural network to test samples (test samples be not present in train sample In this) reason identification is carried out, test samples 1-5 is caused by cause of lightning stroke, and test samples 6-10 is distracter (non-lightning stroke), It is obtaining the result is as follows:
Sample Output 1 Output 2 Output 3 Output 4 Output 5 Diagnosis of Primary because
1 0.9859 0.0010 0.0067 0.0009 0.0104 Lightning stroke
2 0.9848 0.0018 0.0091 0.0002 0.0313 Lightning stroke
3 0.9885 0.0011 0.0018 0.0003 0.0166 Lightning stroke
4 0.9922 0.0008 0.0025 0.0016 0.0040 Lightning stroke
5 0.9949 0.0008 0.0024 0.0058 0.0250 Lightning stroke
6 0.0228 0.0007 0.0109 0.0001 0.3766 Non- lightning stroke
7 0.0328 0.0006 0.0001 0.0000 0.1732 Non- lightning stroke
8 0.2333 0.0008 0.0002 0.0000 0.4300 Non- lightning stroke
9 0.0328 0.0006 0.0001 0.0000 0.1732 Non- lightning stroke
10 0.0136 0.0035 0.0068 0.0000 0.4506 Non- lightning stroke
Reason Lightning stroke Nothing Nothing Nothing Nothing
It can be seen that by the identification result of upper table Lai for test samples, neural network has carried out effective identification, right It can effectively be identified in test samples 1-5, reach desired effect.
In conclusion the present invention is not limited to above-mentioned specific embodiments.Those skilled in the art are not departing from the present invention Under the premise of technical solution, several changes or modification can be done, above-mentioned change or modification each fall within protection scope of the present invention.

Claims (10)

1.一种基于多源信息融合的特高压直流输电线路故障原因辨识方法，其特征是包括如下步骤：A method for identifying fault causes of UHV DC transmission lines based on multi-source information fusion, which comprises the following steps: 步骤1)获取特高压直流输电线路的电气量故障数据和非电气量信息；Step 1) obtaining electrical fault data and non-electrical quantity information of the UHV DC transmission line; 步骤2)提取特高压直流输电线路电气特征量，构造电气特征输入向量；Step 2) extracting the electrical characteristic quantity of the UHV DC transmission line and constructing an electrical characteristic input vector; 步骤3)提取特高压直流输电线路非电气特征量，构造非电气特征输入向量；Step 3) extracting non-electrical characteristic quantities of the UHV DC transmission line and constructing a non-electrical characteristic input vector; 步骤4)对雷击、山火、污秽、风偏、鸟害分别构建综合神经网络辨识模型，神经网络模型的隐含层的层数采用最小误差的方法选取；Step 4) Construct a comprehensive neural network identification model for lightning strikes, mountain fires, pollution, wind deviations, and bird damages. The number of layers of the hidden layer of the neural network model is selected by the method of minimum error; 步骤5)采用自学习的方法进行具体故障原因辨识。Step 5) Use a self-learning method to identify specific fault causes. 2.如权利要求1所述的基于多源信息融合的特高压直流输电线路故障原因辨识方法，其特征是在步骤1)中，特高压直流输电线路的电气量故障数据从保护信息系统中获取，非电气量信息从气象系统、雷电定位系统、地理信息系统、山火监测系统中获取；当发生的故障很少，样本的数量不足时，则通过PSACD仿真故障来获取数据。2 . The method for identifying a fault cause of a UHV DC transmission line based on multi-source information fusion according to claim 1 , wherein in step 1), the electrical fault data of the UHV DC transmission line is obtained from the protection information system. Non-electrical quantity information is obtained from meteorological system, lightning positioning system, geographic information system, and mountain fire monitoring system; when there are few failures and the number of samples is insufficient, the data is acquired by PSACD simulation failure. 3.如权利要求1所述的基于多源信息融合的特高压直流输电线路故障原因辨识方法，其特征是在步骤2)中，设故障发生时刻为t1，提取特高压直流输电线路发生故障后的数据作为电气特征量，提取的时间范围视保护动作时间而定，一般为几毫秒到几十毫秒。电气特征量包括故障电流最大幅值、电压电流幅频图的特征、过渡电阻均值、过渡电阻标准差等；3 . The method for identifying fault causes of UHV DC transmission lines based on multi-source information fusion according to claim 1 , wherein in step 2), the fault occurrence time is t 1 , and the UHV DC transmission line is faulty. The latter data is used as the electrical feature quantity, and the extracted time range depends on the protection action time, and is generally several milliseconds to several tens of milliseconds. The electrical characteristic quantity includes the maximum amplitude of the fault current, the characteristics of the voltage-current amplitude-frequency diagram, the mean value of the transition resistance, the standard deviation of the transition resistance, and the like; 故障电流最大幅值为|I|；其中电流I为故障电流的值，为正或为负；The maximum amplitude of the fault current is |I|; where the current I is the value of the fault current, which is positive or negative; 电压电流幅频图的特征需要对数据进行离散傅里叶分解获得，根据公式The characteristics of the voltage and current amplitude-frequency diagrams need to be obtained by discrete Fourier decomposition of the data, according to the formula 对电流电压傅里叶变换，得到频域的波形，进而提取得到电压电流幅频图的特征量；For the current-voltage Fourier transform, the waveform in the frequency domain is obtained, and then the feature quantity of the voltage-current amplitude-frequency map is extracted; 根据公式According to the formula 得到过渡电阻均值；Obtaining the mean value of the transition resistance; 根据公式According to the formula 过渡电阻标准差E；Transition resistance standard deviation E; 建立电气特征输入向量。Establish an electrical feature input vector. 4.如权利要求1所述的基于多源信息融合的特高压直流输电线路故障原因辨识方法，其特征是在步骤3)中，将非电气特征量分为天气、时段、季节、地形、雷电活动频度、温度、湿度、风力和历史故障信息9个特征量，通过有序的二进制数来表示，建立非电气特征输入向量。4 . The method for identifying fault causes of UHV DC transmission lines based on multi-source information fusion according to claim 1 , wherein in step 3), the non-electrical characteristic quantities are divided into weather, time period, season, terrain, and lightning. The nine characteristic quantities of activity frequency, temperature, humidity, wind and historical fault information are represented by ordered binary numbers to establish a non-electrical characteristic input vector. 5.如权利要求1所述的基于多源信息融合的特高压直流输电线路故障原因辨识方法，其特征是在步骤4)中，采用三层BP神经网络结构，输入层节点数为33，隐含层采用最小误差的方法选取，输出层节点数为5；隐含层和输出层的激活函数均采用log-sigmoid型函数，可以将输出控制在0-1范围之内；训练方法采用trainlm算法，训练目标误差设置为1e-4，训练次数设置为1000次。5 . The method for identifying fault causes of UHV DC transmission lines based on multi-source information fusion according to claim 1 , wherein in step 4), a three-layer BP neural network structure is adopted, and the number of input layer nodes is 33, The layer is selected by the method of minimum error, and the number of nodes in the output layer is 5; the activation functions of the hidden layer and the output layer are all log-sigmoid type functions, and the output can be controlled within the range of 0-1; the training method adopts the trainlm algorithm. The training target error is set to 1e-4 and the number of training times is set to 1000. 6.如权利要求5所述的基于多源信息融合的特高压直流输电线路故障原因辨识方法，其特征是BP神经网络是根据误差反向传播算法训练的多层前馈神经网络，设输入层包含N个神经元，输出层包含M个神经元，输入向量为X，输出向量为Y，隐含层包含L个神经元节点，wij表示输入层到隐含层之间的连接权值，wjk表示隐含层到输出层之间的连接权值，θj表示隐含层神经元的阈值，θk表示输出层神经元的阈值，表示隐含层的激活函数，φ(·)表示输出层的激活函数。The method for identifying fault causes of UHV DC transmission lines based on multi-source information fusion according to claim 5, wherein the BP neural network is a multi-layer feedforward neural network trained according to an error back propagation algorithm, and the input layer is provided. Contains N neurons, the output layer contains M neurons, the input vector is X, the output vector is Y, the hidden layer contains L neuron nodes, and w ij represents the connection weight between the input layer and the hidden layer. w jk represents the connection weight between the hidden layer and the output layer, θ j represents the threshold of the hidden layer neuron, and θ k represents the threshold of the output layer neuron, Represents the activation function of the hidden layer, and φ(·) represents the activation function of the output layer. S型的对数函数为：The logarithmic function of the S type is: 隐含层第j个神经元节点的输出信号oj为：The output signal o j of the jth neuron node of the hidden layer is: 输出层第k个神经元节点的输出信号yk为：The output signal y k of the kth neuron node of the output layer is: 设有P个训练样本，对于每个训练样本,误差的二次型准则函数为：There are P training samples. For each training sample, the quadratic criterion function of the error is: 网络对P个训练样本的总体误差函数为：The overall error function of the network for P training samples is: 其中和分别表示输入训练样本为p时，输出层第k个神经元节点的期望输出和实际输出。among them with Represents the expected output and actual output of the kth neuron node of the output layer when the input training sample is p. 7.如权利要求6所述的基于多源信息融合的特高压直流输电线路故障原因辨识方法，其特征是实际辨识结果和理想输出之间是有差别，用总体辨识误差Eb来表征这种差异：7 . The method for identifying fault causes of UHV DC transmission lines based on multi-source information fusion according to claim 6 , wherein the actual identification result and the ideal output are different, and the overall identification error E b is used to represent the same. difference: Ebi＝Rsi-Rqi (11)E bi =R si -R qi (11) 其中i＝1～5，分别表示雷击、山火、污秽、风偏、鸟害；Rsi、Rqi分别表示此种原因下的实际输出和期望输出，Ebi为此种原因下的实际输出和期望输出之间的差值。Where i=1～5, respectively represent lightning strike, mountain fire, pollution, wind deviation, bird damage; R si and R qi respectively represent the actual output and expected output under this cause, and E bi is the actual output under this reason. The difference between the expected output and the output. 8.如权利要求7所述的基于多源信息融合的特高压直流输电线路故障原因辨识方法，其特征是根据隐含层的经验公式确定一个隐含层的范围：8 . The method for identifying fault causes of UHV DC transmission lines based on multi-source information fusion according to claim 7 , wherein the range of an implicit layer is determined according to an empirical formula of the hidden layer: 式中，m和n分别为输入和输出层的节点数，a为0-10之间的整数；Where m and n are the number of nodes of the input and output layers, respectively, and a is an integer between 0 and 10; 最佳隐含层的确定步骤为：The steps to determine the best hidden layer are: 步骤4.1)计算l的取值范围，l的范围为(6，16)；Step 4.1) Calculate the range of values of l, the range of l is (6, 16); 步骤4.2)对于l内的每一个可能的隐含层数，计算Ebi、Eb；Step 4.2) Calculate E bi and E b for each possible hidden layer in l; 步骤4.3)比较l内的每一个可能的隐含层数所对应的总体辨识误差Eb，取最小的Eb所对应的隐含层数为最佳隐含层。Step 4.3) Compare the total identification error E b corresponding to each possible hidden layer number in l, and take the minimum number of hidden layers corresponding to E b as the optimal hidden layer. 9.如权利要求7所述的基于多源信息融合的特高压直流输电线路故障原因辨识方法，其特征是在步骤5)中，故障发生时，利用创建的网络模型对故障原因进行判别，故障处理结束后，将发生的故障作为标准故障类型录入历史故障数据库，并将实际的故障原因与模型判别结果进行对比，对比结果作为是否修改神经网络训练库的依据；当总体辨识误差Eb>0.5时，修改实际的故障原因模型训练库的样本集，加入该次故障的实际特征数据进行重新训练，得到修正后的神经网络预测模型，从而能够自主学习，适应环境的变化，更加准确的辨识故障原因；如果判别结果与实际结果偏差在允许范围内，则不需要修改训练库中的样本，而是记录该故障特征及故障原因。9 . The method for identifying a fault cause of a UHV DC transmission line based on multi-source information fusion according to claim 7 , wherein in step 5), when the fault occurs, the fault is determined by using the created network model, and the fault is generated. After the processing is finished, the fault that occurs is entered into the historical fault database as the standard fault type, and the actual fault cause is compared with the model discriminant result. The comparison result is used as the basis for modifying the neural network training library; when the overall identification error E b > At 0.5 o'clock, modify the actual fault cause model training library sample set, add the actual feature data of the fault to retrain, and obtain the modified neural network prediction model, so that it can learn autonomously, adapt to environmental changes, and more accurately identify The cause of the fault; if the deviation between the discriminant result and the actual result is within the allowable range, it is not necessary to modify the sample in the training library, but to record the fault feature and the cause of the fault. 10.如权利要求9所述的基于多源信息融合的特高压直流输电线路故障原因辨识方法，其特征是神经网络训练的步骤为：10. The method for identifying a fault cause of a UHV DC transmission line based on multi-source information fusion according to claim 9, wherein the step of training the neural network is: 步骤5.1)训练样本的形成，根据步骤2)和步骤3)中的电气特征量和非电气特征量的提取，构造综合输入特征向量，输入向量为输入1～输入33，根据样本已知的故障原因类型构造输出向量，进而将输入向量和输出向量对应起来构成训练样本；Step 5.1) The formation of the training sample is constructed according to the extraction of the electrical feature quantity and the non-electrical feature quantity in step 2) and step 3), and the integrated input feature vector is constructed, and the input vector is input 1 to input 33, and the fault is known according to the sample. The cause type constructs an output vector, which in turn associates the input vector and the output vector to form a training sample; 步骤5.2)初始取l＝6，信号的正向传播，依次经过输入层、隐含层、输出层；Step 5.2) Initially taking l=6, the forward propagation of the signal passes through the input layer, the hidden layer, and the output layer in turn; 步骤5.3)计算输出误差，如果目标误差满足设定值1e-3，则完成训练输出结果，否则进行第四步；Step 5.3) calculating the output error, if the target error satisfies the set value 1e-3, the training output result is completed, otherwise the fourth step is performed; 步骤5.4)信号的反向传播，将误差反向传递，根据附加动量法trainlm依次修正每一层的权值和阈值；Step 5.4) The back propagation of the signal, the error is reversely transmitted, and the weight and threshold of each layer are sequentially corrected according to the additional momentum method trainlm; 步骤5.5)根据新的权值和阈值重复步骤5.2)、步骤5.3)、步骤5.4)，直到误差满足设定值；判断l是否取完，如果取完则进行步骤5.6)，否则l+1，返回步骤5.2)；Step 5.5) Repeat steps 5.2), 5.3), and 5.4) according to the new weight and threshold until the error satisfies the set value; judge whether l is taken, and if it is finished, proceed to step 5.6), otherwise l+1, Return to step 5.2); 步骤5.6)根据最佳隐含层的确定方法，比较l内的每一个可能的隐含层数所对应的总体辨识误差Eb，取最小的Eb所对应的隐含层数为最佳隐含层；Step 5.6) According to the method for determining the optimal hidden layer, compare the total identification error E b corresponding to each possible hidden layer number in l, and take the minimum number of hidden layers corresponding to E b as the best hidden Containing layer 步骤5.7)进行实际案例测试，最佳隐含层确定后，神经网络的结构便确定，将实际的故障情况输入到神经网络中，得到辨识的结果；Step 5.7) Perform the actual case test. After the optimal hidden layer is determined, the structure of the neural network is determined, and the actual fault condition is input into the neural network to obtain the identified result; 步骤5.8)采用自学习算法，每次辨识后，将实际的故障原因与预测模型判别结果进行对比，对比结果作为是否修改神经网络训练库的依据；如果判别结果有所偏差，那么修改实际的故障原因模型训练库的样本集，加入该次故障的实际特征数据进行重新训练，得到修正后的神经网络预测模型。Step 5.8) Using a self-learning algorithm, after each identification, the actual fault cause is compared with the prediction model discriminant result, and the comparison result is used as a basis for modifying the neural network training library; if the discriminating result is deviated, then the actual fault is modified. The sample set of the reason model training library is added to the actual feature data of the fault to be retrained, and the corrected neural network prediction model is obtained.
CN201811366886.6A 2018-11-16 2018-11-16 Fault identification method for UHVDC transmission lines based on multi-source information fusion Active CN109270407B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811366886.6A CN109270407B (en) 2018-11-16 2018-11-16 Fault identification method for UHVDC transmission lines based on multi-source information fusion

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811366886.6A CN109270407B (en) 2018-11-16 2018-11-16 Fault identification method for UHVDC transmission lines based on multi-source information fusion

Publications (2)

Publication Number Publication Date
CN109270407A true CN109270407A (en) 2019-01-25
CN109270407B CN109270407B (en) 2022-02-25

Family

ID=65189592

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811366886.6A Active CN109270407B (en) 2018-11-16 2018-11-16 Fault identification method for UHVDC transmission lines based on multi-source information fusion

Country Status (1)

Country Link
CN (1) CN109270407B (en)

Cited By (24)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110018389A (en) * 2019-02-21 2019-07-16 国网山东省电力公司临沂供电公司 A kind of transmission line of electricity on-line fault monitoring method and system
CN110045691A (en) * 2019-03-13 2019-07-23 东北大学 A kind of multitasking fault monitoring method of multi-source heterogeneous big data
CN110146789A (en) * 2019-06-27 2019-08-20 国网山东省电力公司菏泽供电公司 A method and device for intelligent transportation inspection notification
CN110533331A (en) * 2019-08-30 2019-12-03 广东电网有限责任公司江门供电局 A kind of fault early warning method and system based on transmission line of electricity data mining
CN110609206A (en) * 2019-08-21 2019-12-24 北京四方继保自动化股份有限公司 Fault identification method for high-voltage direct-current transmission system
CN111579978A (en) * 2020-05-18 2020-08-25 刘柱华 System and method for realizing relay fault identification based on artificial intelligence technology
CN111695543A (en) * 2020-06-23 2020-09-22 贵州电网有限责任公司 Method for identifying hidden danger discharge type of power transmission line based on traveling wave time-frequency characteristics
CN111880046A (en) * 2020-07-22 2020-11-03 国网河北省电力有限公司 A device and method for quickly identifying the cause of a line fault
CN111898729A (en) * 2020-06-10 2020-11-06 国网江苏省电力有限公司电力科学研究院 A method and system for identifying the cause of a transmission line fault
CN112147462A (en) * 2020-09-16 2020-12-29 国网江西省电力有限公司电力科学研究院 Power transmission line fault identification method based on deep learning
CN112782614A (en) * 2019-11-11 2021-05-11 株洲中车时代电气股份有限公司 Fault early warning method and device of converter based on multi-information fusion
CN113159345A (en) * 2021-05-08 2021-07-23 中国电力科学研究院有限公司 Power grid fault identification method and system based on fusion neural network model
CN113295935A (en) * 2021-06-01 2021-08-24 兰州资源环境职业技术学院 Lightning stroke risk assessment method based on high-precision lightning positioning technology
CN113466633A (en) * 2021-07-30 2021-10-01 武汉三相电力科技有限公司 Distributed fault monitoring integrated method and device for power transmission line
CN113533906A (en) * 2021-07-28 2021-10-22 广西电网有限责任公司电力科学研究院 Intelligent overhead transmission line fault type diagnosis method and system
CN113937740A (en) * 2021-12-20 2022-01-14 国网四川省电力公司电力科学研究院 A method and system for fault identification and disposal of distribution lines considering mountain fire risk
CN113960417A (en) * 2021-11-19 2022-01-21 国网湖南省电力有限公司 Power transmission line fault rapid diagnosis method, device, equipment and medium based on multi-source information fusion
CN113960409A (en) * 2021-09-14 2022-01-21 广州番禺电缆集团有限公司 Cable fault cause determination method, device, equipment and storage medium
CN114825418A (en) * 2022-06-07 2022-07-29 北京信云筑科技有限责任公司 Multifunctional flexible complementary digital power supply management method
CN115932484A (en) * 2023-02-15 2023-04-07 重庆大学 Method and device for identifying and ranging faults of power transmission line and electronic equipment
CN117216485A (en) * 2023-11-09 2023-12-12 国网山东省电力公司电力科学研究院 Objective weighting-based power transmission wave-recording bird damage fault judging method and system
CN117743944A (en) * 2024-01-04 2024-03-22 重庆大学 Power transmission line fault identification method and device based on multi-source data and electronic equipment
CN118962323A (en) * 2024-07-08 2024-11-15 广东电网能源发展有限公司 Cable fault identification and positioning method, device, equipment and storage medium
CN119247253A (en) * 2024-09-23 2025-01-03 北京合众伟奇科技股份有限公司 A method and device for monitoring the hidden danger of burning of metering equipment

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5041976A (en) * 1989-05-18 1991-08-20 Ford Motor Company Diagnostic system using pattern recognition for electronic automotive control systems
JPH11142466A (en) * 1997-11-05 1999-05-28 Chugoku Electric Power Co Inc:The Method and device for estimating cause of accident of distribution lines
US20080077836A1 (en) * 2006-09-27 2008-03-27 Khoche A Jay Diagnostic Information Capture from Memory Devices with Built-in Self Test
CN101975910A (en) * 2010-09-07 2011-02-16 昆明理工大学 Intelligent fault classification and location method for ultra-high voltage direct current transmission line
CN105676067A (en) * 2016-01-26 2016-06-15 云南电网有限责任公司电力科学研究院 Fault diagnosis method and system of power transmission lines
CN106405339A (en) * 2016-11-11 2017-02-15 中国南方电网有限责任公司 Power transmission line fault reason identification method based on high and low frequency wavelet feature association
CN107340456A (en) * 2017-05-25 2017-11-10 国家电网公司 Power distribution network operating mode intelligent identification Method based on multiple features analysis
CN108108877A (en) * 2017-11-29 2018-06-01 海南电网有限责任公司电力科学研究院 A kind of transmission line of electricity damage to crops caused by thunder methods of risk assessment based on BP neural network
CN108663600A (en) * 2018-05-09 2018-10-16 广东工业大学 A kind of method for diagnosing faults, device and storage medium based on power transmission network

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5041976A (en) * 1989-05-18 1991-08-20 Ford Motor Company Diagnostic system using pattern recognition for electronic automotive control systems
JPH11142466A (en) * 1997-11-05 1999-05-28 Chugoku Electric Power Co Inc:The Method and device for estimating cause of accident of distribution lines
US20080077836A1 (en) * 2006-09-27 2008-03-27 Khoche A Jay Diagnostic Information Capture from Memory Devices with Built-in Self Test
CN101975910A (en) * 2010-09-07 2011-02-16 昆明理工大学 Intelligent fault classification and location method for ultra-high voltage direct current transmission line
CN105676067A (en) * 2016-01-26 2016-06-15 云南电网有限责任公司电力科学研究院 Fault diagnosis method and system of power transmission lines
CN106405339A (en) * 2016-11-11 2017-02-15 中国南方电网有限责任公司 Power transmission line fault reason identification method based on high and low frequency wavelet feature association
CN107340456A (en) * 2017-05-25 2017-11-10 国家电网公司 Power distribution network operating mode intelligent identification Method based on multiple features analysis
CN108108877A (en) * 2017-11-29 2018-06-01 海南电网有限责任公司电力科学研究院 A kind of transmission line of electricity damage to crops caused by thunder methods of risk assessment based on BP neural network
CN108663600A (en) * 2018-05-09 2018-10-16 广东工业大学 A kind of method for diagnosing faults, device and storage medium based on power transmission network

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
L.XU,ETC: "Comparisons of logistic regression and artificial neural network on power distribution systems fault cause identification", 《PROCEEDINGS OF THE 2005 IEEE MIDNIGHT-SUMMER WORKSHOP ON SOFT COMPUTING IN INDUSTRIAL APPLICATION》 *
张虹: "输电线路故障原因综合辨识研究", 《中国优秀硕士学位论文全文数据库工程科技II辑》 *
申元 等: "基于BP神经网络的输电线路故障原因辨识研究", 《智能电网》 *

Cited By (34)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110018389B (en) * 2019-02-21 2021-11-12 国网山东省电力公司临沂供电公司 Online fault monitoring method and system for power transmission line
CN110018389A (en) * 2019-02-21 2019-07-16 国网山东省电力公司临沂供电公司 A kind of transmission line of electricity on-line fault monitoring method and system
CN110045691A (en) * 2019-03-13 2019-07-23 东北大学 A kind of multitasking fault monitoring method of multi-source heterogeneous big data
CN110045691B (en) * 2019-03-13 2021-03-16 东北大学 Multi-task processing fault monitoring method for multi-source heterogeneous big data
CN110146789A (en) * 2019-06-27 2019-08-20 国网山东省电力公司菏泽供电公司 A method and device for intelligent transportation inspection notification
CN110146789B (en) * 2019-06-27 2021-11-16 国网山东省电力公司菏泽供电公司 Intelligent operation and inspection reporting method and device
CN110609206A (en) * 2019-08-21 2019-12-24 北京四方继保自动化股份有限公司 Fault identification method for high-voltage direct-current transmission system
CN110533331A (en) * 2019-08-30 2019-12-03 广东电网有限责任公司江门供电局 A kind of fault early warning method and system based on transmission line of electricity data mining
CN110533331B (en) * 2019-08-30 2024-03-15 广东电网有限责任公司江门供电局 A fault early warning method and system based on transmission line data mining
CN112782614A (en) * 2019-11-11 2021-05-11 株洲中车时代电气股份有限公司 Fault early warning method and device of converter based on multi-information fusion
CN111579978A (en) * 2020-05-18 2020-08-25 刘柱华 System and method for realizing relay fault identification based on artificial intelligence technology
CN111579978B (en) * 2020-05-18 2024-01-02 珠海施诺电力科技有限公司 System and method for realizing relay fault identification based on artificial intelligence technology
CN111898729A (en) * 2020-06-10 2020-11-06 国网江苏省电力有限公司电力科学研究院 A method and system for identifying the cause of a transmission line fault
CN111695543A (en) * 2020-06-23 2020-09-22 贵州电网有限责任公司 Method for identifying hidden danger discharge type of power transmission line based on traveling wave time-frequency characteristics
CN111880046A (en) * 2020-07-22 2020-11-03 国网河北省电力有限公司 A device and method for quickly identifying the cause of a line fault
CN112147462A (en) * 2020-09-16 2020-12-29 国网江西省电力有限公司电力科学研究院 Power transmission line fault identification method based on deep learning
CN113159345A (en) * 2021-05-08 2021-07-23 中国电力科学研究院有限公司 Power grid fault identification method and system based on fusion neural network model
CN113295935A (en) * 2021-06-01 2021-08-24 兰州资源环境职业技术学院 Lightning stroke risk assessment method based on high-precision lightning positioning technology
CN113533906B (en) * 2021-07-28 2022-09-23 广西电网有限责任公司电力科学研究院 A kind of intelligent overhead transmission line fault type diagnosis method and system
CN113533906A (en) * 2021-07-28 2021-10-22 广西电网有限责任公司电力科学研究院 Intelligent overhead transmission line fault type diagnosis method and system
CN113466633A (en) * 2021-07-30 2021-10-01 武汉三相电力科技有限公司 Distributed fault monitoring integrated method and device for power transmission line
CN113960409B (en) * 2021-09-14 2023-10-24 广州番禺电缆集团有限公司 Cable fault cause determining method, device, equipment and storage medium
CN113960409A (en) * 2021-09-14 2022-01-21 广州番禺电缆集团有限公司 Cable fault cause determination method, device, equipment and storage medium
CN113960417A (en) * 2021-11-19 2022-01-21 国网湖南省电力有限公司 Power transmission line fault rapid diagnosis method, device, equipment and medium based on multi-source information fusion
CN113937740B (en) * 2021-12-20 2022-03-11 国网四川省电力公司电力科学研究院 Distribution line fault identification and disposal method and system considering forest fire risk
CN113937740A (en) * 2021-12-20 2022-01-14 国网四川省电力公司电力科学研究院 A method and system for fault identification and disposal of distribution lines considering mountain fire risk
CN114825418A (en) * 2022-06-07 2022-07-29 北京信云筑科技有限责任公司 Multifunctional flexible complementary digital power supply management method
CN115932484B (en) * 2023-02-15 2023-07-18 重庆大学 Transmission line fault identification and fault location method, device and electronic equipment
CN115932484A (en) * 2023-02-15 2023-04-07 重庆大学 Method and device for identifying and ranging faults of power transmission line and electronic equipment
CN117216485A (en) * 2023-11-09 2023-12-12 国网山东省电力公司电力科学研究院 Objective weighting-based power transmission wave-recording bird damage fault judging method and system
CN117216485B (en) * 2023-11-09 2024-01-30 国网山东省电力公司电力科学研究院 Objective weighting-based power transmission wave-recording bird damage fault judging method and system
CN117743944A (en) * 2024-01-04 2024-03-22 重庆大学 Power transmission line fault identification method and device based on multi-source data and electronic equipment
CN118962323A (en) * 2024-07-08 2024-11-15 广东电网能源发展有限公司 Cable fault identification and positioning method, device, equipment and storage medium
CN119247253A (en) * 2024-09-23 2025-01-03 北京合众伟奇科技股份有限公司 A method and device for monitoring the hidden danger of burning of metering equipment

Also Published As

Publication number Publication date
CN109270407B (en) 2022-02-25

Similar Documents

Publication Publication Date Title
CN109270407A (en) Fault cause identification method for UHV DC transmission line based on multi-source information fusion
CN106961249B (en) A photovoltaic array fault diagnosis and early warning method
CN110929918B (en) 10kV feeder fault prediction method based on CNN and LightGBM
Xie et al. Predicting lightning outages of transmission lines using generalized regression neural network
Xu et al. A classification approach for power distribution systems fault cause identification
CN106777984A (en) A kind of method that photovoltaic array Working state analysis and fault diagnosis are realized based on density clustering algorithm
CN109782124A (en) An integrated fault location method and system for main and auxiliary equipment based on gradient descent algorithm
CN107766879A (en) The MLP electric network fault cause diagnosis methods of feature based information extraction
CN109655712A (en) A kind of distribution network line fault analysis of causes method and system
CN112464995A (en) Power grid distribution transformer fault diagnosis method and system based on decision tree algorithm
Wang et al. Design and implementation of early warning system based on educational big data
CN116090821A (en) Power distribution network line security risk assessment method considering extreme weather
CN114492662B (en) Secondary system fault location method for smart substation based on Transformer-GRU
CN109917228A (en) A ground fault location method for distribution network based on traveling wave method based on RBF neural network
CN106327071A (en) Power line communication risk analysis method and power line communication risk analysis system
CN118521042B (en) A protection method and device for underground cables
CN119247048B (en) A method and system for locating power transmission line faults based on transfer learning
CN109657872A (en) A kind of prediction technique of network distribution transformer operating status
Sun et al. Early warning system for spatiotemporal prediction of fault events in a power transmission system
CN111126672A (en) A typhoon disaster prediction method for high-voltage overhead transmission lines based on classification decision tree
Yang et al. SILOS: An intelligent fault detection scheme for solar insecticidal lamp IoT with improved energy efficiency
CN108694479A (en) Consider the distribution network reliability prediction technique that weather influences time between overhaul
Min et al. Fault prediction for distribution network based on CNN and LightGBM algorithm
CN118607713A (en) Method, device and storage medium for predicting power grid equipment failure rate under typhoon weather
CN108121796A (en) Electric energy metering device failure analysis methods and device based on confidence level

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