CN114662394B - High-voltage alternating-current line protection behavior evaluation method and system based on YOLO V3 - Google Patents
High-voltage alternating-current line protection behavior evaluation method and system based on YOLO V3 Download PDFInfo
- Publication number
- CN114662394B CN114662394B CN202210312344.0A CN202210312344A CN114662394B CN 114662394 B CN114662394 B CN 114662394B CN 202210312344 A CN202210312344 A CN 202210312344A CN 114662394 B CN114662394 B CN 114662394B
- Authority
- CN
- China
- Prior art keywords
- network
- evaluation
- training
- sample
- protection
- 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
Links
- 238000011156 evaluation Methods 0.000 title claims abstract description 106
- 238000012549 training Methods 0.000 claims abstract description 87
- 239000011159 matrix material Substances 0.000 claims abstract description 46
- 238000004088 simulation Methods 0.000 claims abstract description 44
- 238000012360 testing method Methods 0.000 claims abstract description 34
- 238000010586 diagram Methods 0.000 claims abstract description 24
- 230000006399 behavior Effects 0.000 claims description 52
- 230000009471 action Effects 0.000 claims description 49
- 238000000034 method Methods 0.000 claims description 16
- 230000007704 transition Effects 0.000 claims description 12
- 230000005540 biological transmission Effects 0.000 claims description 11
- 230000016507 interphase Effects 0.000 claims description 3
- 238000005516 engineering process Methods 0.000 abstract description 6
- 238000004458 analytical method Methods 0.000 description 12
- 238000011161 development Methods 0.000 description 5
- 238000002372 labelling Methods 0.000 description 3
- 230000007257 malfunction Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000012937 correction Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- RZVAJINKPMORJF-UHFFFAOYSA-N Acetaminophen Chemical compound CC(=O)NC1=CC=C(O)C=C1 RZVAJINKPMORJF-UHFFFAOYSA-N 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013075 data extraction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000007123 defense Effects 0.000 description 1
- 230000002542 deteriorative effect Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000001172 regenerating effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/12—Symbolic schematics
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2113/00—Details relating to the application field
- G06F2113/04—Power grid distribution networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention provides a high-voltage alternating current line protection behavior evaluation method and system based on YOLO V3, comprising the following steps: presetting input ports and output ports of a YOLO V3 algorithm, wherein the number of the input ports and the output ports is not less than 2; constructing a double-end power supply system simulation model according to the input port, the output port and a preset double-end power supply system principle, generating a parameter matrix of the evaluated protection device, inputting the parameter matrix of a line where the evaluated protection device is arranged in batch by using the double-end power supply system simulation model to obtain data corresponding to a current result in a difference state, and generating a fault time sequence diagram as an evaluation sample and classifying the fault time sequence diagram; setting preset network parameters, acquiring from an evaluation sample and training a preset network by using a training set; and selecting an applicable test weight test preset network to acquire a network training criterion, and judging and acquiring an applicable evaluation network according to the network training criterion so as to evaluate the high-voltage alternating current line protection behavior by the applicable evaluation network. The problems of long time consumption, low accuracy and the like of relay protection behavior evaluation technology are solved.
Description
Technical Field
The invention relates to a relay protection action behavior evaluation method, in particular to a high-voltage alternating current line protection behavior evaluation method and system based on YOLO V3.
Background
In the period of rapid development of the electric power system in China, the electric energy has obvious changes in the aspects of production, transmission, application and the like. As the systems of the power system become larger and larger, the systems are more and more closely connected, and factors influencing the safe and stable operation of the system are more and more increased, particularly, accidents such as short circuit or disconnection and the like occur on a high-voltage-class power transmission line (such as an ultra-high voltage or ultra-high voltage line) or some core hub substations, which can generate great impact on the stability of the system, further influence the reliability of the power system and cause serious consequences for the daily life of various departments of national economy and people. The occurrence of various faults and abnormal operating conditions is unavoidable in the operation of the power system, and among these, various forms of short circuits are the most common and dangerous. Once the power transmission line has short circuit fault, the consequence of power failure is caused by light equipment, and the equipment is burnt down, thereby threatening the life safety of people, even causing large-scale power failure and disturbing the social order. The relay protection is used as a first defense line of the power system, the correct and rapid action of the relay protection can cut off faults, the fault range is reduced, the faults are prevented from further deteriorating, and the safe and stable operation of the power grid is ensured. The manufacturers of the relay protection device mainly comprise Nanrui, fang, nanzhong, xu Ji and the like in China and Siemens, ABB and the like in foreign countries. Although the protection function of the protection device of each manufacturer has become more and more standard and perfect along with the development of technology, in some unusual situations (such as rainy and snowy weather), the incorrect action of the relay protection device still exists. If the protection device is not properly operated, a great influence is caused on the power systems which are closely connected with each other, and the stable operation of the power systems is also damaged. Therefore, the method can analyze and evaluate the action behavior of the relay protection device in time, improve the reliability of relay protection and is very important for the stable operation of the power system.
Along with the wide application of the artificial intelligence algorithm in the power system, the research results of related software and algorithms in relay protection are also emerging. For example, the application patent with the application number of CN201310642248.3, namely a relay protection evaluation analysis method and a relay protection evaluation analysis system, analyzes the action condition of a protection device according to fixed value data provided by a fault recorder and the relay protection device, acquires and manages data information, and creates an information data database; creating an analysis model of the relay protection device, wherein the analysis model comprises an operation reason analysis model and a correction analysis model; searching the information database according to the current running state of the relay protection device, finding a corresponding analysis model, and calculating an evaluation result; the step of analyzing the action condition of the protection device according to the fixed value data provided by the fault recorder and the relay protection device, obtaining and managing data information and creating information data database specifically comprises the following steps: acquiring a fixed value state quantity, a fault state quantity and a protection action state quantity; and generating an operation module of each link by matching the logic principle of the relay protection device with the constant state quantity, the fault state quantity and the protection action state quantity, and storing the three state quantities and the operation module to create an information database. In the technical scheme disclosed in the prior patent, the analyzed parameters and types of the relay protection device are different from those of the application, and meanwhile, the patent adopts an operation reason analysis model and a correction analysis model, which are obviously different from the model adopted by the application.
The traditional relay protection action evaluation mainly uses information such as fault wave recording data, and the like, and combines the waveform diagrams of the analog quantity and the switching quantity to perform manual analysis and evaluation by redrawing the waveforms of the voltage and the current (analog quantity) and the protection action (switching quantity) during faults, so that the method is single in means, low in efficiency and long in time consumption. In addition, the correctness of the manual evaluation result is directly related to the experience of an expert, the experience of the expert needs to be accumulated through a large number of actual cases, but the actual cases of high-voltage communication faults are fewer, when the system has protection malfunction or refusal, the objectivity and comprehensiveness of the analysis result are difficult to ensure, and the actual requirements cannot be met by the traditional relay protection action evaluation. The scheme utilizes the excellent data extraction capability of the YOLO V3 algorithm and the capability of deep mining of data characteristics, jumps out of an original evaluation framework, avoids the influence of inherent defects of a protection principle and parameter configuration, improves the accuracy of protection behavior evaluation, reduces the difficulty of relay protection action evaluation, improves the evaluation efficiency, can assist operators on duty to process accidents, improves the emergency processing speed of faults, and has engineering practical significance. In summary, the prior art has the technical problems of long time consumption and low accuracy of relay protection behavior evaluation technology.
Disclosure of Invention
The technical problems to be solved by the invention are how to solve the technical problems of long time consumption, low accuracy and the like of relay protection behavior evaluation technology in the prior art.
The invention adopts the following technical scheme to solve the technical problems: a high-voltage alternating-current line protection behavior evaluation method based on YOLO V3 comprises the following steps:
s1, presetting input ports and output ports of the YOLO V3 algorithm, wherein the number of the input ports and the output ports is not less than 2;
S2, constructing a double-end power supply system simulation model according to the input port, the output port and a preset double-end power supply system principle, generating a parameter matrix of the evaluated protection device, inputting the parameter matrix of a line where the evaluated protection device is arranged in batch by using the double-end power supply system simulation model to obtain data corresponding to current results in different states, generating a fault time sequence diagram as an evaluation sample according to the data, and classifying the evaluation sample, wherein the step S2 further comprises:
S21, building a double-end power supply system simulation model in preset power system simulation software;
s22, taking values of parameters affecting the correctness of the line protection action in a variable range to obtain parameters of the protection device to be evaluated, traversing the parameters of the protection device to be evaluated according to a permutation and combination mode, and forming a parameter matrix;
S23, utilizing the double-end power supply system simulation model to input the parameter matrix of the line where the evaluated protection device is positioned in batch to obtain differential state current data and correspond to the differential evaluation results one by one so as to obtain data corresponding to the differential state current results;
S24, generating the fault time sequence diagram as an evaluation sample according to the data corresponding to the difference state current result, and classifying the evaluation sample by using a classification tag;
S3, setting network parameters and learning parameters of a preset network, acquiring from the evaluation sample, and training the preset network by using a training set;
And S4, selecting an applicable test weight to test the preset network to obtain a network training criterion, judging and obtaining an applicable evaluation network according to the network training criterion, and evaluating the high-voltage alternating current line protection behavior by using the applicable evaluation network.
The invention combines the current and the protection action waveform during the AC line fault by using the YOLO V3 algorithm, and the simulation expert analyzes the waveform to evaluate the protection action behavior. And simulating a large number of alternating current faults by adopting a simulation experiment to analyze, and judging the protection action condition by learning the characteristics of the waveform. The method is simple and direct, does not need to calculate various setting values, has accurate and reliable results, has no special requirements on equipment, and is convenient to implement.
In a more specific technical solution, the step S1 includes:
s11, acquiring protection behavior evaluation data of a YOLO V3 algorithm;
s12, setting the input item of the input port as the three-phase current quantity of the two ends of the line and the action condition of a relay protection device on a high-voltage alternating-current line according to the protection behavior evaluation data;
And S13, setting the output items of the output port to be correct actions of the fault protection in the area, non-actions of the fault protection outside the area, refusal actions of the protection due to large transition resistance and misoperation of the protection due to the influence of the distributed capacitance current of the long circuit according to the protection behavior evaluation data.
In a more specific technical solution, the preset power system simulation software in the step S2 includes MATLAB, and the simulation data in the double-ended power supply system simulation model includes: the power supply voltage at two ends of the line, the equivalent impedance of the power supply, the system frequency and the length of the line in the fault area.
According to the invention, by writing the MATLAB program, the parameter matrix is sequentially input into the model in a row unit and operated, so that the current data of the two ends of the power transmission line in various states are obtained, and meanwhile, the current data are in one-to-one correspondence with the evaluation results.
In a more specific embodiment, each row in the parameter matrix in step S22 represents a combination of all the variable parameters in the sequential failure, and each column represents a parameter.
According to the method, the circuit parameters corresponding to the four output types are respectively written with the program codes to generate the corresponding parameter matrixes, and the data are acquired more pertinently. Because each parameter is traversed when the training set is generated, the faults under different conditions can be correctly evaluated according to the generalization and learning capacity of the network, and parameter setting is not needed.
In a more specific aspect, the parameters of the protection device under evaluation in step S22 include: the power supply comprises the grade of power supplies at two ends, the system frequency, the length of a power transmission line, line parameters, fault positions, fault types, interphase fault resistances, ground fault resistances, phase angle differences and amplitude differences of the power supplies at two ends.
According to the invention, a simulation model is built to carry out traversal simulation on a series of parameters influencing the correct action of the relay protection device, such as power supply voltage, system frequency, transmission line parameters, fault positions, transition resistance, fault types and the like, so that massive double-end current data are obtained, the double-end current data are used as training samples, the trouble of sample lack in engineering practice is avoided, the powerful learning generalization capability of the YOLO V3 algorithm is relied on, accurate relay protection behavior evaluation on a local power grid and even a whole power grid can be realized, and the method has wide application prospects in the development of future intelligent power grids.
In a more specific aspect, the step S24 includes:
s241, generating training sample matrixes for all output ends;
s242, changing the value of the parameter affecting the correctness of the circuit protection action, and inputting a test value parameter matrix of the circuit where the evaluated protection device is positioned in batch by using the double-end power supply system simulation model to obtain differential test current data and corresponding to the differential evaluation result, so as to generate a test sample matrix for all the output ends;
s243, label classification is carried out on the training sample matrix and the test sample matrix.
According to the method, based on the presetting of the input port and the output port of the protection behavior evaluation result of the YOLO V3 algorithm, a double-end power supply system simulation model is built according to the principle of the double-end power supply system, and the labeling tool is utilized to label samples in a classified mode. Setting network parameters and learning parameters, and training the network by using a training set.
In a more specific technical solution, the step S3 includes:
s31, taking the current and protection condition graph line as a sample training set;
s32, setting the network parameters and the learning parameters of the preset network;
s33, converting the picture by automatic inversion, adjusting saturation, exposure and tone to expand the sample training set;
s34, multi-scale training of the preset network.
In a more specific aspect, the step S32 includes: momentum is set to 0.9, weight decay factor is set to 0.005, saturation is set to 1.5, exposure is set to 1.5, hue is set to 0.1, initial learning rate is set to 0.001, learning rate control parameter is set to 1000, learning rate variation step size is set to 40000, 45000, and learning rate variation factor is set to 0.1.
In a more specific technical solution, the step S4 includes:
S41, acquiring the error of the current network through training of the preset network;
s42, judging whether the error of the current network is within a preset application range;
S43, if yes, the network completes training, and the weight obtained by training is judged to be the prediction applicable weight;
And S44, if not, returning to the step S3.
According to the invention, the protection behavior evaluation is carried out through the trained network, the retraining is not needed, and the engineering requirement is met to a great extent. When the trained network is tested, the evaluation network can accurately evaluate the protection behavior under the condition that parameters such as transition resistance, fault position, system voltage level and the like are different from the training samples. When the training samples are enough, the YOLO V3 algorithm is used for evaluating relay protection action behaviors of the high-voltage alternating current line, and has good accuracy.
In a more specific technical scheme, the YOLO V3-based high-voltage alternating-current line protection behavior evaluation system comprises:
the port presetting module is used for presetting input ports and output ports of the YOLO V3 algorithm, and the number of the input ports and the output ports is not less than 2;
constructing a model generation sample module, constructing a double-end power supply system simulation model according to an input port, the output port and a preset double-end power supply system principle, generating a parameter matrix of an evaluated protection device, inputting the parameter matrix of a circuit where the evaluated protection device is positioned in batches by using the double-end power supply system simulation model to obtain data corresponding to current results in different states, generating a fault time sequence diagram as an evaluation sample, classifying the evaluation sample, and connecting the model generation sample module with the port preset module;
the build model generation sample module includes:
The simulation model building unit builds a double-end power supply system simulation model in preset power system simulation software;
The parameter matrix unit is used for taking the value of each factor in the variable range to obtain the parameters of the evaluated protection device, and traversing the parameters of the evaluated protection device in a permutation and combination mode to form a parameter matrix;
The current result corresponding data unit is used for inputting the parameter matrix of the circuit where the rated protection device is positioned in batches by using the double-end power supply system simulation model to obtain differential state current data and corresponds to the differential evaluation result one by one to obtain differential state current result corresponding data, and the current result corresponding data unit is connected with the simulation model building unit and the parameter matrix unit;
The label sample unit is used for generating a fault time sequence diagram as an evaluation sample according to the data corresponding to the current result in the difference state, classifying the label evaluation sample and connecting the label sample unit with the data unit corresponding to the current result;
the network training module is used for training and setting network parameters and learning parameters of a preset network, acquiring and training the preset network by using a training set from the evaluation sample, and connecting the network training module with the model building sample module;
The evaluation module is used for selecting the applicable test weight to test the preset network to obtain the network training criterion, judging and obtaining the applicable evaluation network according to the network training criterion, evaluating the high-voltage alternating current line protection behavior by the applicable evaluation network, and the evaluation module is connected with the network training module.
Compared with the prior art, the invention has the following advantages:
The invention combines the current and the protection action waveform during the AC line fault by using the YOLO V3 algorithm, and the simulation expert analyzes the waveform to evaluate the protection action behavior. And simulating a large number of alternating current faults by adopting a simulation experiment to analyze, and judging the protection action condition by learning the characteristics of the waveform. The method is simple and direct, does not need to calculate various setting values, has accurate and reliable results, has no special requirements on equipment, and is convenient to implement.
When the current data are acquired, the circuit parameters corresponding to the four output types are respectively written with the program codes to generate the corresponding parameter matrix, the parameters are traversed, the faults under different conditions can be correctly judged according to the generalization and learning capacity of the network, and the setting of the circuit parameters is not needed.
According to the invention, a simulation model is built to carry out traversal simulation on a series of parameters influencing the correct action of the relay protection device, such as power supply voltage, system frequency, transmission line parameters, fault positions, transition resistance, fault types and the like, so that massive double-end current data are obtained, the double-end current data are used as training samples, the trouble of sample lack in engineering practice is avoided, the powerful learning generalization capability of the YOLO V3 algorithm is relied on, accurate relay protection behavior evaluation on a local power grid and even a whole power grid can be realized, and the method has wide application prospects in the development of future intelligent power grids.
According to the method, based on the presetting of the input port and the output port of the protection behavior evaluation result of the YOLO V3 algorithm, a double-end power supply system simulation model is built according to the principle of the double-end power supply system, and the labeling tool is utilized to label samples in a classified mode. Setting network parameters and learning parameters, and training the network by using a training set. According to the invention, the protection behavior evaluation is carried out through the trained network, the retraining is not needed, and the engineering requirement is met to a great extent.
When the network after training is tested, the evaluation network can accurately evaluate the protection behavior under the condition that parameters such as transition resistance, fault position, system voltage level and the like are different from the training sample. When the training samples are enough, the YOLO V3 algorithm is used for evaluating relay protection action behaviors of the high-voltage alternating current line, and has high accuracy. The invention solves the problems of long time consumption and low accuracy of the existing relay protection behavior evaluation technology.
Drawings
FIG. 1 is a diagram of steps for generating a protection behavior evaluation network;
FIG. 2 is a block diagram of protection behavior evaluation based on the YOLO V3 algorithm;
FIG. 3 is a schematic diagram of a dual-ended power supply system of the present invention;
FIG. 4 is a diagram of a MATLAB simulation model of the present invention;
FIG. 5 is a timing diagram of the fault protection corrective action within a zone;
FIG. 6 is a timing diagram of out-of-zone fault protection correct inactivity;
FIG. 7 is a timing diagram of a large transition resistance rejection;
FIG. 8 is a timing diagram of distributed capacitive current malfunction;
FIG. 9 is a label schematic of a sample of the fault protection corrective action within a zone;
FIG. 10 is a schematic diagram showing the evaluation result as the correct action of the fault protection in the area;
FIG. 11 is a schematic diagram showing that the evaluation result is that the out-of-zone fault protection is correct;
FIG. 12 is a schematic diagram showing the evaluation result of large transition resistance rejection;
fig. 13 is an evaluation result of a schematic diagram of distributed capacitive current malfunction.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions in the embodiments of the present invention will be clearly and completely described in the following in conjunction with the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Example 1
The specific implementation steps are as follows:
Step 1: as shown in fig. 1, 2 and the following table, based on the presets of the input port and the output port of the protection behavior evaluation result of the convolutional neural network, 7 ports are input, namely, the three-phase current amount at the two ends of the line and the action condition of the relay protection device on the line, wherein the input of 1/0 is the protection action condition, the action of the protection device is represented by 1, and the action of the protection device is represented by 0, as shown in fig. 1. As shown in fig. 2, the output of 4 ports is the evaluation result of 4 types of protection actions, namely, the correct action of the fault protection in the area, the non-action of the fault protection outside the area, the refusal action of the protection due to the large transition resistance and the misoperation of the protection due to the influence of the distributed capacitance current of the long line.
Output port classification table
Step 2: as shown in fig. 3, according to a schematic diagram of the double-ended power supply system, a fault model is built, wherein,For the power supply voltage at two ends of the line, Z m、Zn is the power supply equivalent impedance, f is the system frequency, l 1 is the line length in the fault area, l 2 is the line length outside the fault area, x is the fault position, and i m、in、ia、ib is the sampling current at the protection installation position. The training samples and the test samples are obtained, and the implementation modes are as follows:
As shown in fig. 4, step a: building a double-end power supply system simulation model in electric power system simulation software MATLAB;
And (B) step (B): programming the program code to generate the parameter matrix. The parameters comprise a series of parameters which influence the correct action of the relay protection device, such as the grade of power supplies at two ends, the system frequency, the length of a transmission line, the line parameters, the fault position, the fault type, the interphase fault resistance, the grounding fault resistance, the phase angle difference and the amplitude difference of the power supplies at two ends, and the like, each factor is valued in a variable range of the relay protection device, the parameters are traversed in a permutation and combination mode to form a parameter matrix, each row in the matrix represents a combination of all variable parameters in the sequential faults, and each column represents a parameter. In order to obtain data more pertinently, the invention respectively writes the line parameters corresponding to the four output types into the program codes to generate the corresponding parameter matrix;
Step C: and carrying out batch input operation on the parameters. Programming a program code, sequentially inputting the parameter matrix generated in the step B into a model according to a row unit, and operating to obtain current data at two ends of the power transmission line in various states, wherein the current data corresponds to the evaluation result one by one;
Step D: training samples are generated. And C, drawing the current data of the two ends obtained in the step C and the corresponding protection action condition into a graph, and taking the graph as a sample, as shown in fig. 5 to 8. In step B, the present invention writes the line parameters corresponding to the 4 output types into the corresponding parameter matrix, and the variable parameter table and the sample number of each output training sample are as follows:
Table 1 is an output 1 training sample parameter traversal table, with all combinations of training sample parameters being common Therefore, the total number of training samples is 2400, the training samples are 101×7×2400 dimensional matrices, and the output is 5×2400 dimensional matrices.
TABLE 1 output 1 training sample parameter traversal table
Table 2 is an output 2 training sample parameter traversal table, with all combinations of training sample parameters being commonTherefore, the total number of training samples is 2400, the training samples are 101×7×2400 dimensional matrices, and the output is 5×2400 dimensional matrices.
TABLE 2 output 2 training sample parameter traversal table
Table 3 is an output 3 training sample parameter traversal table with 2230 sets of training samples.
TABLE 3 output 3 training sample parameter traversal table
Table 4 is an output 4 training sample parameter traversal table, with all combinations of training sample parameters being commonThe total number of training samples is 2178, the training samples are 101×7×2178 dimensional matrices, and the output is 5×2178 dimensional matrices.
TABLE 4 output 4 training parameter traversal table
The generation of test samples is consistent with training samples. Changing the value of each parameter affecting the evaluation result in the procedure of the step B, regenerating different parameter matrixes, and repeating the step C to obtain test samples after the processing of the step D, wherein the variable parameter table and the sample number of each output test sample are as follows:
Table 5 is an output 1 test sample parameter traverse table, all combined with The total number of samples is 1200, the test sample set is 101 multiplied by 7 multiplied by 1200 dimensional matrix, and the output is 5 multiplied by 1200 dimensional matrix.
TABLE 5 output 1 test sample parameter traversal table
Table 6 is an output 2 test sample parameter traverse table, all combined withThe total number of samples is 1200, the test sample set is 101 multiplied by 7 multiplied by 1200 dimensional matrix, and the output is 5 multiplied by 1200 dimensional matrix.
TABLE 6 output 2 test sample parameter traversal table
Table 7 is an output 3 test sample parameter traversal table with 1338 sets of test samples.
TABLE 7 output 3 test sample parameter traversal table
H Table 8 is an output 4 test sample parameter traversal table, all combined withThe total number of samples is 1188, the test sample set is a 101×7×1188 dimension matrix, and the output is a5×1188 dimension matrix.
TABLE 8 output 4 test sample parameter traversal table
Step E: labeling. The labeling tool is used for classifying and labeling samples, the tool used by the invention is labelImg, the labeling process is shown in fig. 9, and the label of the samples with correct action of fault protection in a region is shown in fig. 9.
Step 3: after the sample data set is generated, training and testing of the network can be performed. In the training stage, momentum is set to be 0.9 so as to avoid overfitting and improve training speed, a weight attenuation coefficient is set to be 0.005, saturation is set to be 1.5, exposure is set to be 1.5, tone is set to be 0.1, initial learning rate is set to be 0.001, learning rate control parameters are set to be 1000, learning rate change step sizes are set to be 40000 and 45000, learning rate change factors are set to be 0.1, multi-scale training is started, and diversity of sample pictures is increased by automatically reversing and converting pictures, adjusting saturation, exposure and tone. The proper iteration times have a larger influence on the loss value (loss) of the algorithm, the network cannot learn sufficiently due to the too small iteration times, excessive iteration times can generate an excessive fitting phenomenon, the generalization capability is lost, and the network has a better learning effect on specific features.
As shown in the following table, the average precision mAP of multiple categories at different iteration numbers is that, when the iteration number is 110000, the mAP is 0.84992, and the average processing time of a single image is 0.0408s.
Multi-class average precision table under different iteration times
Step 4: because each parameter is traversed when the training set is designed, the faults under different conditions can be correctly judged according to the generalization and learning capacity of the network, and the parameter setting is not needed. It can be found that when the trained network is tested, the evaluation network can perform correct evaluation on the protection behavior under the condition that parameters such as transition resistance, fault position, system voltage level and the like are different from the training samples, and fig. 10 to 13 are evaluation effect diagrams of 4 outputs.
In summary, the present invention combines the current and the protection action waveform during the ac line fault by using YOLO V3 algorithm, and simulates expert analysis waveform to evaluate the protection action behavior. And simulating a large number of alternating current faults by adopting a simulation experiment to analyze, and judging the protection action condition by learning the characteristics of the waveform. The method is simple and direct, does not need to calculate various setting values, has accurate and reliable results, has no special requirements on equipment, and is convenient to implement.
When the current data are acquired, the circuit parameters corresponding to the four output types are respectively written with the program codes to generate the corresponding parameter matrix, the parameters are traversed, the faults under different conditions can be correctly judged according to the generalization and learning capacity of the network, and the setting of the circuit parameters is not needed.
According to the invention, a simulation model is built to carry out traversal simulation on a series of parameters influencing the correct action of the relay protection device, such as power supply voltage, system frequency, transmission line parameters, fault positions, transition resistance, fault types and the like, so that massive double-end current data are obtained, the double-end current data are used as training samples, the trouble of sample lack in engineering practice is avoided, the powerful learning generalization capability of the YOLO V3 algorithm is relied on, accurate relay protection behavior evaluation on a local power grid and even a whole power grid can be realized, and the method has wide application prospects in the development of future intelligent power grids.
According to the method, based on the presetting of the input port and the output port of the protection behavior evaluation result of the YOLO V3 algorithm, a double-end power supply system simulation model is built according to the principle of the double-end power supply system, and the labeling tool is utilized to label samples in a classified mode. Setting network parameters and learning parameters, and training the network by using a training set. According to the invention, the protection behavior evaluation is carried out through the trained network, the retraining is not needed, and the engineering requirement is met to a great extent.
When the network after training is tested, the evaluation network can accurately evaluate the protection behavior under the condition that parameters such as transition resistance, fault position, system voltage level and the like are different from the training sample. When the training samples are enough, the YOLO V3 algorithm is used for evaluating relay protection action behaviors of the high-voltage alternating current line, and has high accuracy. The invention solves the problems of long time consumption and low accuracy of the existing relay protection behavior evaluation technology.
The above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (10)
1. A method for evaluating the protection behavior of a high-voltage alternating-current line based on YOLO V3 is characterized by comprising the following steps:
s1, presetting input ports and output ports of the YOLO V3 algorithm, wherein the number of the input ports and the output ports is not less than 2;
s2, constructing a double-end power supply system simulation model according to the input port, the output port and a preset double-end power supply system principle, generating a parameter matrix of an evaluated protection device, inputting the parameter matrix of a line where the evaluated protection device is arranged in batch by using the double-end power supply system simulation model to obtain data corresponding to a current result in a different state, generating a fault time sequence diagram as an evaluation sample according to the data, and classifying the evaluation sample, wherein the step S2 further comprises:
S21, building a double-end power supply system simulation model in preset power system simulation software;
s22, taking values of parameters affecting the correctness of the circuit protection action in a variable range to obtain parameters of the evaluated protection device, and traversing the parameters of the evaluated protection device in a permutation and combination mode to form a parameter matrix;
S23, utilizing the double-end power supply system simulation model to input the parameter matrix of the line where the evaluated protection device is positioned in batch to obtain differential state current data and correspond to the differential evaluation results one by one so as to obtain data corresponding to the differential state current results;
S24, generating the fault time sequence diagram as an evaluation sample according to the data corresponding to the difference state current result, and classifying the evaluation sample by using a classification tag;
S3, setting network parameters and learning parameters of a preset network, acquiring from the evaluation sample, and training the preset network by using a training set;
And S4, selecting an applicable test weight to test the preset network to obtain a network training criterion, judging and obtaining an applicable evaluation network according to the network training criterion, and evaluating the high-voltage alternating current line protection behavior by using the applicable evaluation network.
2. The YOLO V3-based high-voltage ac line protection behavior evaluation method according to claim 1, wherein said step S1 comprises:
s11, acquiring protection behavior evaluation data of a YOLO V3 algorithm;
s12, setting the input item of the input port as the three-phase current quantity of the two ends of the line and the action condition of a relay protection device on a high-voltage alternating-current line according to the protection behavior evaluation data;
And S13, setting the output items of the output port to comprise correct actions of the fault protection in the area, no actions of the fault protection outside the area, refusal of the protection due to weak feedback, refusal of the protection due to large transition resistance and misoperation due to influence of distributed capacitance current of a long line according to the protection behavior evaluation data.
3. The YOLO V3-based high-voltage ac line protection behavior evaluation method according to claim 1, wherein the preset power system simulation software in the step S21 includes MATLAB, and the simulation data in the double-ended power supply system simulation model includes: the power supply voltage at two ends of the line, the equivalent impedance of the power supply, the system frequency and the length of the line in the fault area.
4. The method according to claim 1, wherein each row in the parameter matrix in the step S22 represents a combination of all variable parameters in the sequential fault, and each column represents a parameter.
5. The YOLO V3-based high-voltage ac line protection behavior evaluation method according to claim 1, wherein the evaluation protection device parameters in step S22 include: the power supply comprises the grade of power supplies at two ends, the system frequency, the length of a power transmission line, line parameters, fault positions, fault types, interphase fault resistances, ground fault resistances, phase angle differences and amplitude differences of the power supplies at two ends.
6. The YOLO V3-based high-voltage ac line protection behavior evaluation method according to claim 1, wherein said step S24 comprises:
s241, generating training sample matrixes for all output ends;
s242, changing the value of the parameter affecting the correctness of the circuit protection action, and inputting a test value parameter matrix of the circuit where the evaluated protection device is positioned in batch by using the double-end power supply system simulation model to obtain differential test current data and corresponding to the differential evaluation result, so as to generate a test sample matrix for all the output ends;
s243, label classification is carried out on the training sample matrix and the test sample matrix.
7. The YOLO V3-based high-voltage ac line protection behavior evaluation method according to claim 1, wherein said step S3 comprises:
s31, taking the current and protection condition graph line as a sample training set;
s32, setting the network parameters and the learning parameters of the preset network;
s33, converting the picture by automatic inversion, adjusting saturation, exposure and tone to expand the sample training set;
s34, multi-scale training of the preset network.
8. The YOLO V3-based high-voltage ac line protection behavior evaluation method according to claim 7, wherein said step S32 comprises: momentum is set to 0.9, weight decay factor is set to 0.005, saturation is set to 1.5, exposure is set to 1.5, hue is set to 0.1, initial learning rate is set to 0.001, learning rate control parameter is set to 1000, learning rate variation step size is set to 40000, 45000, and learning rate variation factor is set to 0.1.
9. The YOLO V3-based high-voltage ac line protection behavior evaluation method according to claim 1, wherein said step S4 comprises:
S41, acquiring the error of the current network through training of the preset network;
s42, judging whether the error of the current network is within a preset application range;
S43, if yes, the network completes training, and the weight obtained by training is judged to be the prediction applicable weight;
And S44, if not, returning to the step S3.
10. A YOLO V3-based high voltage ac line protection behavior evaluation system, comprising:
the port presetting module is used for presetting input ports and output ports of the YOLO V3 algorithm, and the number of the input ports and the output ports is not less than 2;
The model generation sample module is used for constructing a double-end power supply system simulation model according to the input port, the output port and a preset double-end power supply system principle to generate a parameter matrix of the evaluated protection device, the parameter matrix of a line where the evaluated protection device is arranged is input in batches by using the double-end power supply system simulation model to obtain data corresponding to a current result in a different state, a fault time sequence chart is generated as an evaluation sample according to the data, the evaluation sample is classified, and the model generation sample module is connected with the port preset module;
the build model generation sample module includes:
the simulation model building unit is used for building a double-end power supply system simulation model in preset power system simulation software;
The parameter matrix unit is used for taking the value of each factor in the variable range to obtain the parameters of the evaluated protection device, and traversing the parameters of the evaluated protection device in a permutation and combination mode to form a parameter matrix;
The current result corresponding data unit is used for inputting the parameter matrix of the circuit where the evaluated protection device is positioned in batch by utilizing the double-end power supply system simulation model to obtain differential state current data and corresponds to the differential evaluation result one by one so as to obtain the differential state current result corresponding data, and the current result corresponding data unit is connected with the simulation model building unit and the parameter matrix unit;
a tag sample unit, configured to generate the fault timing diagram as an evaluation sample according to the differential state current result corresponding data, and classify the tag evaluation sample, where the tag sample unit is connected with the current result corresponding data unit;
The network training module is used for training and setting network parameters and learning parameters of a preset network, acquiring and training the preset network by using a training set from the evaluation sample, and connecting the network training module with the model building sample module;
and the evaluation module is used for selecting the applicable test weight to test the preset network to obtain the network training criterion, judging and obtaining the applicable evaluation network according to the network training criterion, evaluating the high-voltage alternating-current line protection behavior by using the applicable evaluation network, and is connected with the network training module.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210312344.0A CN114662394B (en) | 2022-03-28 | 2022-03-28 | High-voltage alternating-current line protection behavior evaluation method and system based on YOLO V3 |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210312344.0A CN114662394B (en) | 2022-03-28 | 2022-03-28 | High-voltage alternating-current line protection behavior evaluation method and system based on YOLO V3 |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114662394A CN114662394A (en) | 2022-06-24 |
| CN114662394B true CN114662394B (en) | 2024-07-09 |
Family
ID=82034285
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210312344.0A Active CN114662394B (en) | 2022-03-28 | 2022-03-28 | High-voltage alternating-current line protection behavior evaluation method and system based on YOLO V3 |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114662394B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117454231B (en) * | 2023-11-27 | 2024-07-02 | 广州得元电力科技有限公司 | Subway direct current protection test system applying intelligent simulation technology |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106786560A (en) * | 2017-02-14 | 2017-05-31 | 中国电力科学研究院 | A kind of power system stability characteristic automatic extraction method and device |
| CN106980073A (en) * | 2017-05-23 | 2017-07-25 | 武汉大学 | A kind of two ends of electric transmission line fault recorder data matching process based on convolutional neural networks |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113066042A (en) * | 2019-12-31 | 2021-07-02 | 英特尔公司 | Apparatus and method for assessing quality of 3D point cloud |
-
2022
- 2022-03-28 CN CN202210312344.0A patent/CN114662394B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106786560A (en) * | 2017-02-14 | 2017-05-31 | 中国电力科学研究院 | A kind of power system stability characteristic automatic extraction method and device |
| CN106980073A (en) * | 2017-05-23 | 2017-07-25 | 武汉大学 | A kind of two ends of electric transmission line fault recorder data matching process based on convolutional neural networks |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114662394A (en) | 2022-06-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109086889B (en) | Terminal fault diagnosis method, device and system based on neural network | |
| CN109635928B (en) | Voltage sag reason identification method based on deep learning model fusion | |
| CN113011481B (en) | Electric energy meter function abnormality assessment method and system based on decision tree algorithm | |
| CN114219147A (en) | Power distribution station fault prediction method based on federal learning | |
| CN107741952B (en) | Automatic scoring and examination system based on SCD configuration file comparison of intelligent substation | |
| CN107817404A (en) | A kind of Portable metering automatization terminal trouble-shooter and its diagnostic method | |
| CN109936113B (en) | Protection action intelligent diagnosis method and system based on random forest algorithm | |
| CN114662394B (en) | High-voltage alternating-current line protection behavior evaluation method and system based on YOLO V3 | |
| CN113219328B (en) | Intelligent fault diagnosis method for circuit breaker operating mechanism based on information fusion | |
| CN106649479A (en) | Probability graph-based transformer state association rule mining method | |
| CN112949874B (en) | Power distribution terminal defect characteristic self-diagnosis method and system | |
| CN110687473A (en) | Fault positioning method and system for relay protection test of intelligent substation | |
| CN110910026A (en) | Intelligent management and decision-making method and system for route loss of trans-provincial power transmission line | |
| CN108399579B (en) | Intelligent analysis system for monitoring data signals of substation equipment | |
| CN114091549A (en) | Equipment fault diagnosis method based on deep residual error network | |
| CN111027697A (en) | Genetic algorithm packaged feature selection power grid intrusion detection method | |
| CN113848471B (en) | Intelligent fault positioning method and system for relay protection system | |
| CN111091141B (en) | Photovoltaic backboard fault diagnosis method based on layered Softmax | |
| CN104267294A (en) | IGIS fault online diagnostic method based on rough set theory and Petri network | |
| CN110443481B (en) | Power distribution automation terminal state evaluation system and method based on hybrid K-nearest neighbor algorithm | |
| CN113740666B (en) | Method for positioning root fault of storm alarm in power system of data center | |
| CN114252716A (en) | Power grid fault diagnosis method and device and storage medium | |
| CN115586402B (en) | Power distribution network fault diagnosis and processing method | |
| CN117639228A (en) | Power distribution network running state prediction method and system based on digital twin | |
| CN116433018A (en) | Digital twinning-based power distribution network risk situation sensing method and device |
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 |