CN116070151B - Ultra-high voltage direct current transmission line fault detection method based on generalized regression neural network - Google Patents

Ultra-high voltage direct current transmission line fault detection method based on generalized regression neural network Download PDF

Info

Publication number
CN116070151B
CN116070151B CN202310259217.3A CN202310259217A CN116070151B CN 116070151 B CN116070151 B CN 116070151B CN 202310259217 A CN202310259217 A CN 202310259217A CN 116070151 B CN116070151 B CN 116070151B
Authority
CN
China
Prior art keywords
faults
neural network
regression neural
value
fault
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.)
Active
Application number
CN202310259217.3A
Other languages
Chinese (zh)
Other versions
CN116070151A (en
Inventor
张学友
李冀
董翔宇
李坚林
谢佳
罗沙
殷振
阮巍
俞斌
马欢
邵华
贺成成
郑海鑫
张东欣
阴春锦
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.)
Super High Voltage Branch Of State Grid Anhui Electric Power Co ltd
Original Assignee
Super High Voltage Branch Of State Grid Anhui Electric Power Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Super High Voltage Branch Of State Grid Anhui Electric Power Co ltd filed Critical Super High Voltage Branch Of State Grid Anhui Electric Power Co ltd
Priority to CN202310259217.3A priority Critical patent/CN116070151B/en
Publication of CN116070151A publication Critical patent/CN116070151A/en
Application granted granted Critical
Publication of CN116070151B publication Critical patent/CN116070151B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

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
    • 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
    • G01R31/50—Testing of electric apparatus, lines, cables or components for short-circuits, continuity, leakage current or incorrect line connections
    • G01R31/52—Testing for short-circuits, leakage current or ground faults
    • 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
    • G01R31/50—Testing of electric apparatus, lines, cables or components for short-circuits, continuity, leakage current or incorrect line connections
    • G01R31/58—Testing of lines, cables or conductors
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02H—EMERGENCY PROTECTIVE CIRCUIT ARRANGEMENTS
    • H02H7/00—Emergency protective circuit arrangements specially adapted for specific types of electric machines or apparatus or for sectionalised protection of cable or line systems, and effecting automatic switching in the event of an undesired change from normal working conditions
    • H02H7/26—Sectionalised protection of cable or line systems, e.g. for disconnecting a section on which a short-circuit, earth fault, or arc discharge has occured
    • H02H7/268—Sectionalised protection of cable or line systems, e.g. for disconnecting a section on which a short-circuit, earth fault, or arc discharge has occured for dc systems
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • H02J3/36—Arrangements for transfer of electric power between ac networks via a high-tension dc link
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
    • Y04S10/52—Outage or fault management, e.g. fault detection or location

Landscapes

  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Business, Economics & Management (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • Economics (AREA)
  • Public Health (AREA)
  • Water Supply & Treatment (AREA)
  • Human Resources & Organizations (AREA)
  • Marketing (AREA)
  • Primary Health Care (AREA)
  • Strategic Management (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Supply And Distribution Of Alternating Current (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a fault detection method of an extra-high voltage direct current transmission line based on a generalized regression neural network, which is used for extracting fault characteristic quantity under a frequency domain based on generalized S transformation and constructing input data of the generalized regression neural network; carrying out normalization processing on sample data, and dividing the sample data into two samples of a test set and a training set; optimizing generalized regression neural network parameters by using a chaotic quantum particle swarm algorithm, forming an ideal network model by taking the lowest fitness function as a principle, and better learning fault characteristics of the ultra-high voltage direct current transmission line; the deep characteristic quantity is input into a Softmax classifier for classification, fault identification is divided into out-of-zone faults, bus faults and line faults, the faults are classified into positive faults, negative faults and bipolar faults, and an identification result is output. The invention realizes the accurate identification of different faults such as extra-region faults, bus faults, line faults, positive faults, negative faults, bipolar faults and the like of the extra-high voltage direct current line, and has high fault detection speed.

Description

Ultra-high voltage direct current transmission line fault detection method based on generalized regression neural network
Technical Field
The invention relates to the technical field of control of high-voltage direct-current transmission systems, in particular to a fault detection method of an extra-high-voltage direct-current transmission line based on a generalized regression neural network.
Background
With the proposal of the 'double carbon' target, the ultra-high voltage direct current transmission technology has wide deepened application in the aspect of high-capacity transmission because of the characteristics of modularization, low harmonic wave, low loss and the like. With the increase of high-proportion distributed power supplies, extra-high voltage direct current transmission will become one development direction of a future smart grid. However, direct current transmission is a low inertia system, if a short circuit fault occurs in a line, fault current can rise rapidly, and equipment such as an inverter and the like is burned out. Therefore, the extra-high voltage direct current circuit breaker is needed to reliably and rapidly cut off faults, safe and stable operation of a non-fault section is realized, a protection system is needed to complete fault detection and positioning within 3ms, and high requirements are provided for protection of an extra-high voltage direct current transmission line.
The current fault protection method for the extra-high voltage direct current transmission line mainly comprises a traveling wave method and a wavelet method. Although the traveling wave method can be applied to conventional direct current transmission, the defect of difficult detection of a wave head and difficult detection of a short-distance fault can not be applied for a long time; the wavelet method can greatly improve the detection speed of the traveling wave head, but cannot perform fault pole selection.
Disclosure of Invention
In order to solve the problems of low protection reliability and low detection speed of the extra-high voltage direct current transmission line, the invention provides the method for detecting the faults of the extra-high voltage direct current transmission line based on the generalized regression neural network, which utilizes the advantages of strong classification capability and high learning speed of the generalized regression neural network, and the network can also have good classification effect under the condition of few data samples, builds an extra-high voltage direct current transmission line fault identification model, optimizes the model by using a chaotic quantum particle swarm algorithm, effectively improves the protection reliability and the speed of the extra-high voltage direct current transmission line, and has good practical significance.
In order to achieve the above object, the present invention is realized by the following technical scheme:
the invention relates to a fault detection method for an extra-high voltage direct current transmission line based on a generalized regression neural network, which comprises the following steps:
step 1, constructing a protection starting criterion by using a fault starting element, and carrying out subsequent fault identification when the criterion is met;
step 2, extracting fault characteristic quantity under frequency domain characteristics by using generalized S transformation, and extracting transient energy of direct current voltage, bus voltage, positive reactance voltage and negative reactance voltage to form four-dimensional characteristic data as input of a generalized regression neural network;
step 3, carrying out normalization processing on the input four-dimensional characteristic data, and dividing the four-dimensional characteristic data into a test set and a training set;
step 4, optimizing parameters of the generalized regression neural network by using a chaotic vector particle swarm algorithm, finding out a smoothing factor under a minimum fitness function, and forming a classification model;
step 5, inputting a training set into the classification model obtained in the step 4, deeply learning fault characteristics of the ultra-high voltage direct current transmission line, inputting the learned fault characteristics into a Softmax classifier for classification, classifying faults into out-of-zone faults, bus faults and line faults, setting the fault classification into positive faults, negative faults and bipolar faults, setting parameters of the Softmax classifier, outputting a recognition result, and continuously iterating and optimizing parameters of a generalized regression neural network to finally form a network model;
and 6, verifying the validity of the network model obtained in the step 5 by using the test set.
The invention further improves that: step 1 is to set the bus voltage change rate as a fault starting element, and the constructed protection starting criterion is as follows:
Figure SMS_1
wherein: />
Figure SMS_2
For DC line voltage, ">
Figure SMS_3
A threshold is initiated for the dc protection algorithm. The invention further improves that: the generalized S transformation formula in the step 2 is as follows:
Figure SMS_4
wherein:Xdiscrete points that are discrete fourier transforms of the original signal;nin order to be able to take time,n=0,1,2,3,...N-1；
Figure SMS_5
the dimension is the number of sampling points;hrepresenting the quantity for the imaginary part; m is the dimension of the sampling point; />
Figure SMS_6
、/>
Figure SMS_7
N is the total number of sampling points, and T is the sampling period.
Generalized S-transform transient energy sum of signal in specific frequency band
Figure SMS_9
The method comprises the following steps: />
Figure SMS_12
Wherein:
Figure SMS_14
for complex time-frequency matrix, row vector +.>
Figure SMS_10
Column vector +.>
Figure SMS_11
Is the amplitude-frequency characteristic of the corresponding moment; p is the total number of rows of the complex time-frequency matrix; q is the total column number of the complex time-frequency matrix; />
Figure SMS_13
Is->
Figure SMS_15
Matrix element absolute value. The invention further improves that: the data normalization processing in the step 3 is as follows:
Figure SMS_8
wherein:
Figure SMS_17
for normalized sample numberdDimensional data; />
Figure SMS_19
Is the original firstdDimensional data; min, max are minimum and maximum functions. The invention further improves that: the constructed generalized regression neural network consists of an input layer, a mode layer, a summation layer and an outputLayer composition, outputzAt the position ofAThe above regression was: />
Figure SMS_22
Wherein: z is the actual output value +.>
Figure SMS_18
The predicted output value is the predicted output value of the generalized regression neural network; a is a generalized regression neural network input value; />
Figure SMS_21
Is a andzis a joint probability density function of (1); if->
Figure SMS_23
Satisfies the normal distribution, then:
Figure SMS_25
wherein: />
Figure SMS_16
In order to be the size of the sample,gfor the dimension of variable a, use +.>
Figure SMS_20
Substitute->
Figure SMS_24
The method comprises the following steps of: />
Figure SMS_26
Wherein:
Figure SMS_27
inputting a value for a generalized regression neural network; />
Figure SMS_28
Is->
Figure SMS_29
Learning samples corresponding to the neurons; />
Figure SMS_30
Smoothing for generalized regression neural networksFactors.
The invention further improves that: the quantum particle swarm algorithm in the step 4 is described as follows:
the speed of each dimension in the iteration of the traditional particle swarm algorithm is constrained, so that the particle search range cannot contain all feasible domains, and the global convergence to an optimal value cannot be guaranteed. The quantum particle swarm algorithm is a probability algorithm based on population, the quantum mechanics law is utilized to endow particles with quantum characteristics, and the particles can perform specific probability density motion at any position in a feasible region, so that the global optimal value can be obtained in the whole feasible region.X 1 In order to be the location of the particles,tfor the number of iterations, a wave function is used
Figure SMS_31
The state quantity such as the momentum and the energy of the particles is expressed, the probability of the particles appearing at a certain position is obtained by using a probability density function, and the probability of the particles is determined by the potential field. Solving by utilizing the Schrodinger equation to obtain a normalized probability distribution function as follows: />
Figure SMS_32
In the method, in the process of the invention,Lto determine the search interval coefficients of the particles. The particles are searched according to the following iterative equation.
Figure SMS_33
Figure SMS_34
In the method, in the process of the invention,u ij a random number is uniformly distributed between 0 and 1;kis a random number, and ranges from 0 to 1;βfor the expansion factor, the convergence rate of the particles is adjusted, and the calculation formula is as follows:
Figure SMS_35
in the method, in the process of the invention,T max for the maximum number of iterations to be performed,tfor the current iteration number
Figure SMS_36
Is the firstjThe average optimal position of each particle in the dimension is calculated as follows: />
Figure SMS_37
In the method, in the process of the invention,Mis a group rule module;
Figure SMS_38
is->
Figure SMS_39
Particle NoDDimension intOptimal position at the time of iteration.
Figure SMS_40
Is a local attractorPbestAndGbestrandom points in the space, and the calculation formula is as follows:
Figure SMS_41
in the method, in the process of the invention,
Figure SMS_42
for t iterations particle->
Figure SMS_43
First, thejWeight coefficient of local optimal solution is maintained, +.>
Figure SMS_44
For the t-th iteration particle->
Figure SMS_45
First, thejMaintaining the value of the local optimal solution, +.>
Figure SMS_46
Is the value of the j-th dimension global optimal solution at the t-th iteration
The chaotic search algorithm can traverse all states in a specific interval according to the law of the chaotic search algorithm, avoids the phenomenon of local optimal values in the optimizing process, and has strong traversal.
And optimizing the parameters by using a method combining quantum behavior characteristics and chaos search. The method comprises the steps of firstly carrying out global search by using a quantum particle swarm algorithm, searching an optimal value, then adding micro disturbance by taking the value as a center, and carrying out secondary optimization. The algorithm combined by the two methods can realize global optimum, and the optimizing result is unique.
The secondary optimization adopts a classical chaotic system mapping model, and is as follows:
Figure SMS_47
in the method, in the process of the invention,cis chaos factor->
Figure SMS_48
Is the firsteIndividual variable->
Figure SMS_49
Value of sub-chaos search,/>
Figure SMS_50
For chaotic mapping parameters, the values are [0,4 ]]Between (I)>
Figure SMS_51
For the number of chaotic searches, +.>
Figure SMS_52
Is the sequence number of the variable which is,Gthe maximum chaos search times.
The invention further improves that: the step of optimizing parameters of the generalized regression neural network by the chaotic quantum particle swarm algorithm in the step 4 is as follows:
s1: initializing a population; setting population dimension, population scale, iteration frequency upper limit and optimization parameter maximum and minimum values;
s2: the fitness function is constructed, and the expression is as follows:
Figure SMS_53
in the method, in the process of the invention,
Figure SMS_54
is->
Figure SMS_55
Fitness value of individual particles; />
Figure SMS_56
Is->
Figure SMS_57
Predicted coordinates of the individual particles; (x i ,y i ) Is the first
Figure SMS_58
The actual coordinates of the individual nodes are used,Nthe total number of sampling points;
after setting the fitness function, calculating the fitness function of each particle of all populations after initialization, setting the fitness function as an optimal value of the corresponding particle, comparing all the optimal values to obtain a global optimal value, and recording the particle corresponding to the global optimal value as the current global optimal position;
s3: updating the particle position corresponding to the global optimal value by using the Schrodinger wave equation, and restraining the particle position;
s4: calculating the fitness value of the particles again to obtain a current global optimal value and optimal particles;
s5: chaos searching is carried out by using a chaos mapping model, so that chaos traversing space is enlarged; generating a chaotic sequence on the basis of all optimal values of a quantum particle swarm algorithm, and if a better position of a current global optimal position is found in a new interval, replacing the current optimal position to ensure that the population is separated from the danger of local optimization;
s6: judging whether the maximum iteration times or the set searching precision are reached, if the conditions are met, finishing the optimizing, otherwise returning to the step S2 until the termination conditions are met;
s7: the finally obtained optimal
Figure SMS_59
And (3) reconstructing a generalized regression neural network model by values, and inputting input data into the network model with trained parameters to obtain an optimal output value.
The inventionThe further improvement is that: the step 5 is to use 1 and 0 to represent the out-of-zone fault
Figure SMS_60
Bus failure->
Figure SMS_61
Line fault->
Figure SMS_62
Positive electrode failure->
Figure SMS_63
Negative electrode failure->
Figure SMS_64
Bipolar failure->
Figure SMS_65
Whether or not to occur, output fault class phasors of
Figure SMS_66
. The beneficial effects of the invention are as follows: the method overcomes the defects of low fault recognition rate and low detection speed of the traditional extra-high voltage direct current protection, utilizes the advantages of strong classification capability and high learning speed of the generalized regression neural network, can have good classification effect under the condition of few data samples, builds an extra-high voltage direct current transmission line fault recognition model, optimizes the model by using a chaotic quantum particle swarm algorithm, effectively improves the reliability and the quick action of the extra-high voltage direct current transmission line protection, and has good practical significance.
Drawings
FIG. 1 is a flow chart of steps of a method for detecting faults of an extra-high voltage direct current transmission line based on a generalized regression neural network;
FIG. 2 is a graph of network identification accuracy and iteration number in an embodiment of the present invention;
FIG. 3 is a graph of network loss function versus iteration number in an embodiment of the present invention;
fig. 4 is a schematic diagram of accuracy of identifying various faults of an extra-high voltage direct current transmission line in an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention. In addition, the technical features of the embodiments of the present invention described below may be combined with each other as long as they do not collide with each other.
As shown in fig. 1, the method for detecting the fault of the ultra-high voltage direct current transmission line based on the generalized regression neural network mainly comprises network parameter optimization and fault type identification, and specifically comprises the following steps:
step 1, constructing a protection starting criterion by using a fault starting element, and carrying out subsequent fault identification when the criterion is met;
step 2, extracting fault characteristic quantity under frequency domain characteristics by using generalized S transformation, and extracting transient energy of direct current voltage, bus voltage, positive reactance voltage and negative reactance voltage to form four-dimensional characteristic data as input of a generalized regression neural network;
step 3, carrying out normalization processing on the input four-dimensional characteristic data, and dividing the four-dimensional characteristic data into a test set and a training set;
step 4, optimizing parameters of the generalized regression neural network by using a chaotic vector particle swarm algorithm, finding out a smoothing factor under a minimum fitness function, and forming a classification model;
step 5, inputting a training set into the classification model obtained in the step 4, deeply learning the fault characteristics of the ultra-high voltage direct current transmission line, inputting the learned fault characteristics into a Softmax classifier for classification, classifying the fault into an out-of-zone fault, a bus fault and a line fault, setting the fault classification into a positive fault, a negative fault and a bipolar fault, setting parameters of the Softmax classifier, outputting a recognition result, and continuously iterating and optimizing parameters of a generalized regression neural network to finally form a network model with better performance;
and 6, verifying the validity of the network model obtained in the step 5 by using the test set.
In step 1, setting a bus voltage change rate as a fault starting element, and constructing a protection starting criterion as follows:
Figure SMS_67
wherein:
Figure SMS_68
for DC line voltage, ">
Figure SMS_69
A threshold is initiated for the dc protection algorithm. In the step 2, the generalized S transformation adds an adjustable factor, so that compared with the traditional S transformation, the time-frequency resolution is higher, and the transformation formula is as follows:
Figure SMS_70
wherein:Xdiscrete points that are discrete fourier transforms of the original signal;nin order to be able to take time,n=0,1,2,3,...N-1；
Figure SMS_71
the dimension is the number of sampling points;hrepresenting the quantity for the imaginary part; m is the dimension of the sampling point; />
Figure SMS_72
、/>
Figure SMS_73
N is the total number of sampling points, and T is the sampling period.
Generalized S-transform transient energy sum of signal in specific frequency band
Figure SMS_74
The method comprises the following steps: />
Figure SMS_75
Wherein:
Figure SMS_76
for complex time-frequency matrix, row vector +.>
Figure SMS_77
For the time domain feature at a certain frequency, column vector +.>
Figure SMS_78
Is the amplitude-frequency characteristic of a certain moment; p is the total number of rows of the complex time-frequency matrix; q is the total column number of the complex time-frequency matrix; />
Figure SMS_79
Is->
Figure SMS_80
Matrix element absolute value.
In the step 3, the sample data is normalized according to the following principle:
Figure SMS_81
wherein: />
Figure SMS_82
For normalized sample numberdDimensional data; />
Figure SMS_83
Is the original firstdDimensional data; min, max are minimum and maximum functions.
The generalized regression neural network constructed in the step 4 comprises an input layer, a mode layer, a summation layer and an output layer, and outputszAt the position ofAThe above regression was:
Figure SMS_85
wherein: z is the actual output value +.>
Figure SMS_89
The predicted output value is the predicted output value of the generalized regression neural network; a is a generalized regression neural network input value;f(a,z) Is a andzif the joint probability density function of (a)f(a,z) Satisfies the normal distribution, then: />
Figure SMS_93
Wherein: />
Figure SMS_86
In order to be the size of the sample,gfor the dimension of variable a, use +.>
Figure SMS_88
Substitute->
Figure SMS_91
The method comprises the following steps of: />
Figure SMS_94
Wherein: />
Figure SMS_84
Inputting a value for a generalized regression neural network; />
Figure SMS_87
Is->
Figure SMS_90
Learning samples corresponding to the neurons; />
Figure SMS_92
Is a smoothing factor of a generalized regression neural network. The generalized regression neural network smoothing factor has great influence on the prediction performance of the network, and the smoothing factor needs to be optimized by using a chaotic quantum particle swarm algorithm, so that the generalization capability of the network is improved, and the prediction precision of the network is further improved.
The speed of each dimension in the iteration of the traditional particle swarm algorithm is constrained, so that the particle search range cannot contain all feasible domains, and the global convergence to an optimal value cannot be guaranteed. The quantum particle swarm algorithm is a probability algorithm based on population, the quantum mechanics law is utilized to endow particles with quantum characteristics, and the particles can perform specific probability density motion at any position in a feasible region, so that the global optimal value can be obtained in the whole feasible region.
X 1 In order to be the location of the particles,tfor the number of iterations, a wave function is used
Figure SMS_95
The state quantity such as the momentum and the energy of the particles is expressed, the probability of the particles appearing at a certain position is obtained by using a probability density function, and the probability of the particles is determined by the potential field. Solving by utilizing the Schrodinger equation to obtain a normalized probability distribution function as follows: />
Figure SMS_96
In the method, in the process of the invention,Lto determine the search interval coefficients of the particles. The particles are searched according to the following iterative equation. />
Figure SMS_97
In (1) the->
Figure SMS_98
A random number is uniformly distributed between 0 and 1;kis a random number, and ranges from 0 to 1;βfor the expansion factor, the convergence rate of the particles is adjusted, and the calculation formula is as follows: />
Figure SMS_99
In the method, in the process of the invention,T max for the maximum number of iterations to be performed,tfor the current iteration number
Figure SMS_100
Is the firstjThe average optimal position of each particle in the dimension is calculated as follows:
Figure SMS_101
in the method, in the process of the invention,Mis a group rule module; />
Figure SMS_102
Is->
Figure SMS_103
The D dimension of the individual particles is->
Figure SMS_104
Optimal position at the time of iteration.
Figure SMS_105
Is a local attractorPbestAndGbestrandom points in the space, and the calculation formula is as follows:
Figure SMS_106
in (1) the->
Figure SMS_107
For t iterations particle->
Figure SMS_108
First, thejWeight coefficient of local optimal solution is maintained, +.>
Figure SMS_109
For the t-th iteration particle->
Figure SMS_110
First, thejMaintaining the value of the local optimal solution, +.>
Figure SMS_111
Is the value of the j-th dimension global optimal solution at the t-th iteration
The chaotic search algorithm can traverse all states in a specific interval according to the law of the chaotic search algorithm, avoids the phenomenon of local optimal values in the optimizing process, and has strong traversal.
And optimizing the parameters by using a method combining quantum behavior characteristics and chaos search. The method comprises the steps of firstly carrying out global search by using a quantum particle swarm algorithm, searching an optimal value, then adding micro disturbance by taking the value as a center, and carrying out secondary optimization. The algorithm combined by the two methods can realize global optimum, and the optimizing result is unique.
The secondary optimization adopts a classical chaotic system mapping model, and is as follows:
Figure SMS_112
in the method, in the process of the invention,cis chaos factor->
Figure SMS_113
Is the firsteIndividual variable->
Figure SMS_114
Value of sub-chaos search,/>
Figure SMS_115
For chaotic mapping parameters, the values are [0,4 ]]Between (I)>
Figure SMS_116
Number of chaotic searches, ++>
Figure SMS_117
Is the sequence number of the variable which is,Gthe maximum chaos search times.
The step of optimizing parameters of the generalized regression neural network by the chaotic quantum particle swarm algorithm in the step 4 is as follows:
s1: setting the population dimension, population scale, iteration frequency upper limit and optimization parameter maximum and minimum value;
s2: the fitness function is constructed as follows:
Figure SMS_118
in (1) the->
Figure SMS_119
Is->
Figure SMS_120
Fitness value of individual particles; />
Figure SMS_121
Is->
Figure SMS_122
Predicted coordinates of the individual particles; (x i ,y i ) Is->
Figure SMS_123
The actual coordinates of the individual nodes are used,Nthe total number of sampling points;
after setting the fitness function, calculating the fitness function of each particle of all populations after initialization, setting the fitness function as an optimal value of the corresponding particle, comparing all the optimal values to obtain a global optimal value, and recording the particle corresponding to the global optimal value as the current global optimal position;
s3: updating the particle position corresponding to the global optimal value by using the Schrodinger wave equation, and restraining the particle position;
s4: calculating the fitness value of the particles again to obtain a current global optimal value and optimal particles;
s5: generating a chaotic sequence on the basis of all optimal values of a quantum particle swarm algorithm, and if a better position is found in a new interval, replacing the current optimal position to ensure that the population breaks away from the danger of local optimization;
s6: judging whether the maximum iteration times or the set searching precision are reached, if the conditions are met, finishing the optimizing, otherwise returning to the step S2 until the termination conditions are met;
s7: and reconstructing the generalized regression neural network model by the finally obtained optimal value, and inputting the input data into the network model with trained parameters to obtain the optimal output value.
After optimization, the generalized regression neural network is output to a Softmax classifier for classification, the output dimension of the classifier is set to be 6, the classifier respectively represents an out-of-zone fault, a bus fault, a line fault, an anode fault, a cathode fault and a bipolar fault, the occurrence or non-occurrence of the corresponding faults are respectively represented by '1' and '0', and the output fault category phasors are
Figure SMS_124
。
In order to verify the effect of the invention, the following examples illustrate the advanced nature of the proposed protection method for the extra-high voltage direct current transmission line based on the generalized regression neural network.
Based on an electromagnetic transient simulation platform PSCAD/EMTDC, a four-terminal MMC direct current power grid model is built, protection research is conducted on a circuit and a bus, fault data samples are input into a generalized regression neural network optimized by a chaotic quantum particle swarm algorithm, the total number of sample data is 4000 groups, and a training set and a testing set are formed according to the proportion of 3:1. The fault initiation criteria threshold is set to 200kV/ms.
The iteration times of the network are continuously increased, and the recognition accuracy of faults and the network loss result are shown in fig. 2 and 3. From the graph, when the iteration number reaches 500, the method is stable, the fault identification accuracy is highest, and the network loss is minimum. The number of iterations of the network is set to 500.
Simulation verification is carried out on faults at the line, the outside area and the bus respectively, the transition resistance of the faults is set to be 0.01Ω, and the results are shown in table 1. The generalized regression neural network can quickly and accurately identify faults of the lines and the buses and can realize fault pole selection. The maximum detection time is 1.70ms, which shows that the method provided by the invention has the advantage of high detection speed and can meet the quick action requirement of the protection of the extra-high voltage direct current transmission line.
TABLE 1 generalized regression neural network output and protection action results
Figure SMS_125
In order to verify that the method provided by the invention has good effect on various faults, the identification effect of each fault is listed, and the result is shown in fig. 4. The ultra-high voltage direct current transmission line protection method based on the generalized regression neural network has high recognition rate and good practical significance on out-of-zone faults, bus positive faults, bus negative faults, bus bipolar faults, line positive faults, line negative faults and line bipolar faults.
To verify the superiority of the proposed method, the inventive method was compared with a mainstream convolutional neural network, a conventional generalized regression neural network, and a support vector machine, and the results are shown in table 2. The method has high identification accuracy and lower detection time, and meets the requirements of reliability and speed of ultra-high voltage direct current transmission line protection.
Table 2 comparison of different network results
Figure SMS_126
The above embodiments are only for illustrating the technical aspects of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the above embodiments, it should be understood by those of ordinary skill in the art that: modifications and equivalents may be made to the specific embodiments of the invention without departing from the spirit and scope of the invention, which are intended to be covered by the scope of the claims.

Claims (6)

1. The method for detecting the faults of the ultra-high voltage direct current transmission line based on the generalized regression neural network is characterized by comprising the following steps of: the method comprises the following steps:
step 1, constructing a protection starting criterion by using a fault starting element, and carrying out subsequent fault identification when the criterion is met;
step 2, extracting fault characteristic quantity under frequency domain characteristics by using generalized S transformation, and extracting transient energy of direct current voltage, bus voltage, positive reactance voltage and negative reactance voltage to form four-dimensional characteristic data as input of a generalized regression neural network;
step 3, carrying out normalization processing on the input four-dimensional characteristic data, and dividing the four-dimensional characteristic data into a test set and a training set;
step 4, optimizing parameters of the generalized regression neural network by using a chaotic vector particle swarm algorithm, finding out a smoothing factor under a minimum fitness function, and forming a classification model;
step 5, inputting a training set into the classification model obtained in the step 4, deeply learning fault characteristics of the ultra-high voltage direct current transmission line, inputting the learned fault characteristics into a Softmax classifier for classification, classifying faults into out-of-zone faults, bus faults and line faults, setting the fault classification into positive faults, negative faults and bipolar faults, setting parameters of the Softmax classifier, outputting a recognition result, and continuously iterating and optimizing parameters of a generalized regression neural network to finally form a network model;
step 6, verifying the validity of the network model obtained in the step 5 by using a test set;
step 1 is to set the bus voltage change rate as a fault starting element, and the constructed protection starting criterion is as follows:
Figure QLYQS_1
wherein: />
Figure QLYQS_2
For DC line voltage, ">
Figure QLYQS_3
Starting a threshold value for a direct current protection algorithm;
the step 5 is to use 1 and 0 to represent the out-of-zone fault
Figure QLYQS_4
Bus failure->
Figure QLYQS_5
Line fault->
Figure QLYQS_6
Positive electrode failure->
Figure QLYQS_7
Negative electrode failure->
Figure QLYQS_8
Bipolar failure->
Figure QLYQS_9
Whether or not the output failure category phasor is +.>
Figure QLYQS_10
。
2. Generalized regression neural network based on claim 1The fault detection method for the extra-high voltage direct current transmission line is characterized by comprising the following steps of: the generalized S transformation formula in the step 2 is as follows:
Figure QLYQS_12
wherein:Xdiscrete points that are discrete fourier transforms of the original signal;nin order to be able to take time,n=0,1,2,3,...N-1；/>
Figure QLYQS_14
the dimension is the number of sampling points;hrepresenting the quantity for the imaginary part; m is the dimension of the sampling point; />
Figure QLYQS_17
、/>
Figure QLYQS_18
As an adjustable factor, N is the total number of sampling points, and T is the sampling period; generalized S-transform transient energy and +.>
Figure QLYQS_19
The method comprises the following steps: />
Figure QLYQS_20
Wherein: />
Figure QLYQS_21
For complex time-frequency matrix, row vector +.>
Figure QLYQS_11
Column vector +.>
Figure QLYQS_13
Is the amplitude-frequency characteristic of the corresponding moment; p is the total number of rows of the complex time-frequency matrix; q is the total column number of the complex time-frequency matrix; />
Figure QLYQS_15
Is->
Figure QLYQS_16
Matrix element absolute value.
3. The generalized regression neural network-based ultra-high voltage direct current transmission line fault detection method according to claim 1, characterized by comprising the following steps: the data normalization processing in the step 3 is as follows:
Figure QLYQS_22
wherein: />
Figure QLYQS_23
For normalized sample numberdDimensional data; />
Figure QLYQS_24
Is the original firstdDimensional data; min, max are minimum and maximum functions.
4. The generalized regression neural network-based ultra-high voltage direct current transmission line fault detection method according to claim 1, characterized by comprising the following steps: the constructed generalized regression neural network consists of an input layer, a mode layer, a summation layer and an output layer, and outputszAt the position ofAThe above regression was:
Figure QLYQS_25
wherein: z is the actual output value +.>
Figure QLYQS_26
The predicted output value is the predicted output value of the generalized regression neural network; a is a generalized regression neural network input value; />
Figure QLYQS_27
Is a andzis a joint probability density function of (1);
if it is
Figure QLYQS_29
Satisfies the normal distribution, then: />
Figure QLYQS_31
Wherein: />
Figure QLYQS_33
In order to be the size of the sample,gfor the dimension of variable a, use +.>
Figure QLYQS_34
Substitute->
Figure QLYQS_35
The method comprises the following steps of:
Figure QLYQS_36
wherein: />
Figure QLYQS_37
Inputting a value for a generalized regression neural network; />
Figure QLYQS_28
Is->
Figure QLYQS_30
Learning samples corresponding to the neurons; />
Figure QLYQS_32
Is a smoothing factor of a generalized regression neural network.
5. The generalized regression neural network-based ultra-high voltage direct current transmission line fault detection method according to claim 1, characterized by comprising the following steps: the quantum particle swarm algorithm in the step 4 is as follows:X 1 in order to be the location of the particles,tfor the number of iterations, a wave function is used
Figure QLYQS_39
To represent the momentum and energy state quantity of particles, to obtain the probability of particles in a certain place by probability density function, wherein the probability of particles is determined by the potential field, and to obtain normalized probability distribution function by means of Schrodinger equationThe numbers are as follows: />
Figure QLYQS_40
In the method, in the process of the invention,Lto determine the search interval coefficients of the particles, the particles are searched according to the following iterative equation:
Figure QLYQS_41
Figure QLYQS_42
in (1) the->
Figure QLYQS_43
Is the particle ofjDimensional position numbertValues of +1 times,/->
Figure QLYQS_44
Is the particle ofjDimensional position numbertThe value of the secondary number of times,u ij is thatUniformly distributing random numbers between 0 and 1;kis a random number, and ranges from 0 to 1; />
Figure QLYQS_45
For the expansion factor, the convergence rate of the particles is adjusted, and the calculation formula is as follows: />
Figure QLYQS_38
In the method, in the process of the invention,T max for the maximum number of iterations to be performed,tfor the current iteration number
Figure QLYQS_46
Is the firstjThe average optimal position of each particle in the dimension is calculated as follows:
Figure QLYQS_47
in the method, in the process of the invention,Mis a group rule module; />
Figure QLYQS_48
Is->
Figure QLYQS_49
Particle NoDDimension intOptimal position at the time of iteration;
Figure QLYQS_50
is a local attractorPbestAndGbestrandom points in the space, and the calculation formula is as follows:
Figure QLYQS_51
in (1) the->
Figure QLYQS_52
Particle number t of iterationsjWeight coefficient of local optimal solution is maintained, +.>
Figure QLYQS_53
Particle number t at iteration number tjMaintaining the value of the local optimal solution, +.>
Figure QLYQS_54
The value of the j-th dimension global optimal solution in the t-th iteration;
optimizing parameters by using a method combining quantum behavior characteristics and chaos search, performing global search by using a quantum particle swarm algorithm to find an optimal value, adding micro disturbance by taking the value as a center, performing secondary optimization,
the secondary optimization adopts a classical chaotic system mapping model, and is as follows:
Figure QLYQS_55
in the method, in the process of the invention,cis chaos factor->
Figure QLYQS_56
Is the firsteIndividual variable->
Figure QLYQS_57
Value of sub-chaos search,/>
Figure QLYQS_58
For chaotic mapping parameters, the values are between 0, 4->
Figure QLYQS_59
Number of chaotic searches, ++>
Figure QLYQS_60
Is the sequence number of the variable which is,Gthe maximum chaos search times.
6. The generalized regression neural network-based ultra-high voltage direct current transmission line fault detection method according to claim 1, characterized by comprising the following steps: the step of optimizing parameters of the generalized regression neural network by the chaotic quantum particle swarm algorithm in the step 4 is as follows:
s1: initializing a population, and setting the population dimension, the population scale, the upper limit of iteration times and the maximum and minimum values of optimization parameters;
s2: the fitness function is constructed, and the expression is as follows:
Figure QLYQS_61
in (1) the->
Figure QLYQS_62
Is->
Figure QLYQS_63
Fitness value of individual particles; />
Figure QLYQS_64
Is->
Figure QLYQS_65
Predicted coordinates of the individual particles; (x i , y i ) Is->
Figure QLYQS_66
Actual coordinates of the individual particles, N being the total number of sampling points; calculation after setting fitness functionAfter initialization, the fitness function of each particle of all populations is set as an optimal value of the corresponding particle, then all the optimal values are compared to obtain a global optimal value, and the particle corresponding to the global optimal value is recorded as the current global optimal position;
s3: updating the particle position corresponding to the global optimal value by using the Schrodinger wave equation, and restraining the particle position;
s4: calculating the fitness value of the particles again to obtain a current global optimal value and optimal particles;
s5: the chaotic mapping model is used for chaotic search, the chaotic traversal space is enlarged, a chaotic sequence is generated on the basis of all optimal values of the quantum particle swarm algorithm, and if a position better than the current optimal position is found in a new interval, the current optimal position is replaced, so that the population is separated from the danger of local optimization;
s6: judging whether the maximum iteration times or the set searching precision are reached, if the conditions are met, finishing the optimizing, otherwise returning to the step S2 until the termination conditions are met;
s7: smoothing factor of the optimal generalized regression neural network obtained finally
Figure QLYQS_67
And (3) reconstructing a generalized regression neural network model by values, and inputting input data into the network model with trained parameters to obtain an optimal output value.
CN202310259217.3A 2023-03-17 2023-03-17 Ultra-high voltage direct current transmission line fault detection method based on generalized regression neural network Active CN116070151B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310259217.3A CN116070151B (en) 2023-03-17 2023-03-17 Ultra-high voltage direct current transmission line fault detection method based on generalized regression neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310259217.3A CN116070151B (en) 2023-03-17 2023-03-17 Ultra-high voltage direct current transmission line fault detection method based on generalized regression neural network

Publications (2)

Publication Number Publication Date
CN116070151A CN116070151A (en) 2023-05-05
CN116070151B true CN116070151B (en) 2023-06-20

Family

ID=86175219

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310259217.3A Active CN116070151B (en) 2023-03-17 2023-03-17 Ultra-high voltage direct current transmission line fault detection method based on generalized regression neural network

Country Status (1)

Country Link
CN (1) CN116070151B (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018072351A1 (en) * 2016-10-20 2018-04-26 北京工业大学 Method for optimizing support vector machine on basis of particle swarm optimization algorithm
CN111351668A (en) * 2020-01-14 2020-06-30 江苏科技大学 Diesel engine fault diagnosis method based on optimized particle swarm algorithm and neural network
CN112036296A (en) * 2020-08-28 2020-12-04 合肥工业大学 Motor bearing fault diagnosis method based on generalized S transformation and WOA-SVM
CN113158781A (en) * 2021-03-10 2021-07-23 国网安徽省电力有限公司电力科学研究院 Lightning trip type identification method
CN114414942A (en) * 2022-01-14 2022-04-29 重庆大学 Power transmission line fault identification classifier, identification method and system based on transient waveform image identification

Family Cites Families (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
GB2327553B (en) * 1997-04-01 2002-08-21 Porta Systems Corp System and method for telecommunications system fault diagnostics
GB0301707D0 (en) * 2003-01-24 2003-02-26 Rolls Royce Plc Fault diagnosis
WO2012009804A1 (en) * 2010-07-23 2012-01-26 Corporation De L'ecole Polytechnique Tool and method for fault detection of devices by condition based maintenance
US8756047B2 (en) * 2010-09-27 2014-06-17 Sureshchandra B Patel Method of artificial nueral network loadflow computation for electrical power system
CN107884706B (en) * 2017-11-09 2020-04-07 合肥工业大学 Analog circuit fault diagnosis method based on vector value regular kernel function approximation
WO2020087184A1 (en) * 2018-11-01 2020-05-07 University Of Manitoba Method for determining conductors involved in a fault on a power transmission line and fault location using local current measurements
EP3971791A4 (en) * 2019-05-13 2023-02-15 Samsung Electronics Co., Ltd. Classification result verifying method and classification result learning method which use verification neural network, and computing device for performing methods
CN113988136B (en) * 2021-10-29 2024-08-02 浙江工业大学 Medium voltage circuit breaker fault diagnosis method based on deep learning and intelligent optimization
CN114004263B (en) * 2021-12-29 2022-05-03 四川大学 Large-scale equipment working condition diagnosis and prediction method based on feature fusion conversion
CN115754713A (en) * 2022-07-29 2023-03-07 淮阴工学院 Motor fault diagnosis method based on frequency diagram and EFA-BP deep learning
CN115577290A (en) * 2022-09-28 2023-01-06 国网安徽省电力有限公司淮北供电公司 Distribution network fault classification and source positioning method based on deep learning

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018072351A1 (en) * 2016-10-20 2018-04-26 北京工业大学 Method for optimizing support vector machine on basis of particle swarm optimization algorithm
CN111351668A (en) * 2020-01-14 2020-06-30 江苏科技大学 Diesel engine fault diagnosis method based on optimized particle swarm algorithm and neural network
CN112036296A (en) * 2020-08-28 2020-12-04 合肥工业大学 Motor bearing fault diagnosis method based on generalized S transformation and WOA-SVM
CN113158781A (en) * 2021-03-10 2021-07-23 国网安徽省电力有限公司电力科学研究院 Lightning trip type identification method
CN114414942A (en) * 2022-01-14 2022-04-29 重庆大学 Power transmission line fault identification classifier, identification method and system based on transient waveform image identification

Also Published As

Publication number Publication date
CN116070151A (en) 2023-05-05

Similar Documents

Publication Publication Date Title
Tong et al. Detection and classification of transmission line transient faults based on graph convolutional neural network
CN110879917A (en) Electric power system transient stability self-adaptive evaluation method based on transfer learning
CN114169395B (en) Method for constructing dominant instability mode recognition model of power system and application
CN105354643A (en) Risk prediction evaluation method for wind power grid integration
CN111652478B (en) Umbrella algorithm-based power system voltage stability evaluation misclassification constraint method
CN115618249A (en) Low-voltage power distribution station area phase identification method based on LargeVis dimension reduction and DBSCAN clustering
Huang et al. Research on transformer fault diagnosis method based on GWO optimized hybrid kernel extreme learning machine
CN116131313A (en) Explanatory analysis method for association relation between characteristic quantity and transient power angle stability
CN115275990A (en) Evaluation method and system for broadband oscillation risk of regional power grid
CN113988558B (en) Power grid dynamic security assessment method based on blind area identification and electric coordinate system expansion
Xu et al. An improved ELM-WOA–based fault diagnosis for electric power
CN113298296A (en) Method for predicting day-ahead load probability of power transmission substation from bottom to top
CN116070151B (en) Ultra-high voltage direct current transmission line fault detection method based on generalized regression neural network
CN116933860A (en) Transient stability evaluation model updating method and device, electronic equipment and storage medium
Fang et al. Power distribution transformer fault diagnosis with unbalanced samples based on neighborhood component analysis and k-nearest neighbors
He et al. Partial discharge pattern recognition algorithm based on sparse self-coding and extreme learning machine
CN113435575B (en) Gate graph neural network transient stability evaluation method based on unbalanced data
CN116204771A (en) Power system transient stability key feature selection method, device and product
CN114841266A (en) Voltage sag identification method based on triple prototype network under small sample
Qiao et al. Transient stability assessment for ac-dc hybrid systems based on bayesian optimization xgboost
Miao et al. A Transformer District Line Loss Calculation Method Based on Data Mining and Machine Learning
CN113054653A (en) Power system transient stability evaluation method based on VGGNet-SVM
Qian et al. A Data Fusion Approach for Searching Fault Chains in Power Grid over Datasets from Multiple Systems
Zheng et al. DCSCF detection method of MMC-DC-grid based on statistical feature and DBN-SOFTMAX
Lu et al. Power System Transient Stability Assessment Based on Graph Convolutional Network

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant