CN107609769B - Intelligent power distribution network fault early warning method based on fault gene table - Google Patents

Intelligent power distribution network fault early warning method based on fault gene table Download PDF

Info

Publication number
CN107609769B
CN107609769B CN201710796634.6A CN201710796634A CN107609769B CN 107609769 B CN107609769 B CN 107609769B CN 201710796634 A CN201710796634 A CN 201710796634A CN 107609769 B CN107609769 B CN 107609769B
Authority
CN
China
Prior art keywords
distribution network
power distribution
fault
bus
intelligent power
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
CN201710796634.6A
Other languages
Chinese (zh)
Other versions
CN107609769A (en
Inventor
向敏
闵杰
屈琴芹
王在乾
高盼
陈诚
于祥春
许珑璋
孙永民
谭童
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.)
Chongqing University of Post and Telecommunications
Original Assignee
Chongqing University of Post and Telecommunications
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 Chongqing University of Post and Telecommunications filed Critical Chongqing University of Post and Telecommunications
Priority to CN201710796634.6A priority Critical patent/CN107609769B/en
Publication of CN107609769A publication Critical patent/CN107609769A/en
Application granted granted Critical
Publication of CN107609769B publication Critical patent/CN107609769B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Landscapes

  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Measuring Or Testing Involving Enzymes Or Micro-Organisms (AREA)

Abstract

The invention relates to a fault gene table-based intelligent power distribution network fault early warning method, and belongs to the field of intelligent power distribution network fault early warning. Firstly dividing the running state of the intelligent power distribution network into four states of excellent, good and medium, then adopting BP neural network algorithm to carry out state evaluation on the running historical data of the intelligent power distribution network and combining with corresponding faults to obtain the mapping relation between each section of state transfer time sequence and the faults, thereby constructing a fault gene table, then periodically obtaining the state transfer time sequence of the intelligent power distribution network on line, matching the state transfer time sequence with all the genes in the fault gene table through Smith-Waterman algorithm, and if the maximum matching value reaches a set threshold value, early warning the corresponding faults. The intelligent power distribution network fault early warning method can well face the problem that the intelligent power distribution network is more and more complex, improve the accuracy rate of intelligent power distribution network fault early warning, provide guidance and help for relevant managers to maintain the intelligent power distribution network, and effectively improve the scientificity and predictability of power grid operation decision.

Description

Intelligent power distribution network fault early warning method based on fault gene table
Technical Field
The invention belongs to the field of intelligent power distribution network fault early warning, and relates to an intelligent power distribution network fault early warning method based on a fault gene table, which can realize integral control of the running state of an intelligent power distribution network.
Background
In recent years, the research and construction of smart grids has risen globally. The intelligent power distribution network is used as an important component for connecting a main network and supplying power to users in the intelligent power grid, and whether the running state of the intelligent power distribution network is normal or not directly influences the power supply of thousands of households. Meanwhile, with the access of distributed power sources, the popularization of electric automobiles and the increase of user interaction electric power, the dynamic behavior of the power distribution network becomes complex, the operation risk is greatly increased, and once a power failure accident of the power distribution network occurs, huge influence and loss can be caused to social life. Therefore, a need exists for further research on fault early warning of the intelligent power distribution network, and guidance and help are provided for relevant management personnel to maintain the intelligent power distribution network.
At present, scholars at home and abroad propose various solutions from different angles aiming at the fault early warning of the intelligent power distribution network. Research shows that most faults of the power distribution network enter pathological operation before destructive faults occur, and the power distribution network has trend and cumulative effects. However, most of the existing solutions utilize local parameters of the power distribution network, such as harmonic current and short-circuit current, or local components, such as a transformer, or are associated with external factors, such as thunderstorm weather, to achieve the effect of early warning the fault of the intelligent power distribution network. The solutions fail to grasp the operation state of the intelligent power distribution network on the whole, and only consider a certain current factor to judge whether the early warning is needed or not in the early warning scheme, and the trend and the cumulative effect of the power distribution network faults cannot be fully utilized, so that the accuracy of fault early warning is slightly deficient in the face of more and more complex intelligent power distribution networks. Therefore, there is an urgent need for an early warning method that can realize the overall control of the operating state of the smart distribution network and can jointly determine whether a failure is about to occur from the current and past operating states.
Disclosure of Invention
In view of the above, the invention aims to provide a fault early warning method for an intelligent power distribution network based on a fault gene table, which makes full use of the trend and cumulative effect characteristics of faults of the intelligent power distribution network, solves the problem of fault early warning of the intelligent power distribution network, and improves the early warning accuracy rate.
In order to achieve the purpose, the invention provides the following technical scheme:
a fault gene table-based intelligent power distribution network fault early warning method specifically comprises the following steps:
s1: dividing the running states of the intelligent power distribution network, and constructing an intelligent power distribution network state evaluation index system;
s2: constructing a BP neural network model comprising an input layer, a hidden layer and an output layer;
s3: dividing the historical fault data of the smart power grid into two parts, and clustering the operation data of buses in the historical fault data source of the first part of the smart power grid to construct a training sample;
s4: inputting training samples of a first part of smart power grids and the running states of the power distribution network corresponding to the training samples into a BP neural network model for training to obtain a BP neural network state evaluation model;
s5: inputting operation data of buses in a second part of intelligent power distribution network historical fault data sources into a BP neural network state evaluation model to obtain a mapping relation between a state transition time sequence of the intelligent power distribution network and a fault, and constructing a fault gene table;
s6: periodically acquiring running state data of a bus of the current intelligent power distribution network, and inputting the running state data into a BP neural network state evaluation model to obtain a state transition time sequence of the current intelligent power distribution network;
s7: and matching the obtained state transition time sequence of the current smart grid with all genes in a fault gene table through a Smith-Waterman gene sequence comparison algorithm, solving a matching value, comparing the maximum matching value with a set threshold value, and if the maximum matching value exceeds the threshold value, early warning the fault corresponding to the gene.
Further, in step S1, the constructing of the intelligent distribution network state evaluation index system specifically includes:
determining the state evaluation index of the intelligent power distribution network as the state score of each bus of the intelligent power distribution network, wherein the calculation formula of the state score of each bus is as follows:
Figure GDA0002518201230000021
wherein G isbScore the bus state with a value of [0, 1%]Closer to 1 indicates better state, and closer to 0 indicates worse state; lambda [ alpha ]VWeight, g, representing the bus voltageVRepresents the bus voltage score, λPWeight, g, representing the active power of the busPRepresenting the bus active power score, λQWeight, g, representing the reactive power of the busQRepresenting the bus reactive power score.
Further, the bus voltage score calculation formula is as follows:
Figure GDA0002518201230000022
where V is the voltage value of the bus bar,
Figure GDA0002518201230000023
is the average value, V, of the corresponding bus voltage in the historical data setmaxFor the maximum value, V, of the corresponding bus voltage in the historical data setminIs the minimum value of the corresponding bus voltage in the historical data set.
Further, the bus active power score calculation formula is as follows:
Figure GDA0002518201230000024
where P is the value of the active power of the bus,
Figure GDA0002518201230000025
is the average value, P, of the active power of the corresponding bus in the historical data setmaxIs the maximum value, P, of the active power of the corresponding bus in the historical data setminFor the minimum value of the active power of the corresponding bus in the historical data set。
Further, the bus reactive power score calculation formula is as follows:
Figure GDA0002518201230000031
where Q is the reactive power value of the bus,
Figure GDA0002518201230000032
is the average value, Q, of the reactive power of the corresponding bus in the historical data setmaxFor maximum value, Q, of reactive power of corresponding bus in historical data setminThe minimum value of the reactive power of the corresponding bus in the historical data set.
Further, in step S3, the constructing a training sample specifically includes: clustering the operation data of the buses in the first part of historical data sources of the intelligent power distribution network into four types of samples with different states,
Figure GDA0002518201230000033
wherein T isn×m(k) For the training sample matrix, k is 1, 2, 3, 4 for different classes of training samples, n for the number of training samples, m for the number of bus bars, tnmThe state score of the mth bus of the nth training sample is represented.
Further, in step S4, the training process specifically includes:
s41: let the connection weight from the input layer to the hidden layer be wijThe connection weight from the hidden layer to the output layer is wjThe threshold from the input layer to the hidden layer is gammajThe threshold from the hidden layer to the output layer is
Figure GDA0002518201230000034
The number of nodes in the hidden layer is N, the learning rate is η, and the expected output is
Figure GDA0002518201230000035
S42: selecting a sigmoid-type function asActivating a function Aperture for an hidden layer1And output layer activation function air entrainer2；
Figure GDA0002518201230000036
S43: calculating the input and output of each unit of the hidden layer by using the input t of the input layernmInput layer to hidden layer connection weight wijAnd inputting layer to hidden layer threshold gammajCalculating the input h of each unit of the hidden layerjIs then reused hjBy activating a function air entrainer1Computing the output b of each unit of the hidden layerj；
Figure GDA0002518201230000037
Wherein r 1, 2.., n, indicates that the r-th sample is trained;
s44: output b from the hidden layerjThe connection weight w from the hidden layer to the output layerjAnd hidden layer to output layer threshold
Figure GDA0002518201230000039
Calculating an output result y;
Figure GDA0002518201230000038
s45: calculating an error;
Figure GDA0002518201230000041
Figure GDA0002518201230000042
ehj＝wj×eo×bj×(1-bj)
where e is the output error, eo is the output layer generalized error, ehjGeneralizing errors for each unit of the hidden layer;
S46: adjusting the connection weight w from hidden layer to output layerjAnd a threshold value
Figure GDA0002518201230000043
Figure GDA0002518201230000044
W 'of'jAnd
Figure GDA0002518201230000045
the adjusted connection weight and threshold value from the hidden layer to the output layer;
s47: adjusting the connection weight w from the input layer to the hidden layerijAnd a threshold value gammaj；
Figure GDA0002518201230000046
W 'of'ijAnd gamma'jConnecting the adjusted input layer to the hidden layer by the weight and the threshold;
s48: taking the variable r from 1 to n, finishing training all the training samples, then accumulating the error E of each training sample to calculate a global error E, judging whether the error E reaches a specified error range, and if so, finishing training and recording the current connection weight and threshold; if not, the global error E is set to zero, and the process goes to step S43 to repeat the learning training.
Further, the step S5 specifically includes:
s51: and (3) constructing an evaluation matrix by using historical fault data sources of the second part of intelligent power distribution network:
Figure GDA0002518201230000047
wherein
Figure GDA00025182012300000410
To evaluate the matrix, n*Indicating the number of samples to be evaluated,
Figure GDA00025182012300000411
denotes the n-th*The evaluation score of the m-th bus of each evaluation sample;
s52: will evaluate the matrix
Figure GDA00025182012300000412
And inputting the state evaluation model of the BP neural network to obtain the mapping between the state transfer time sequence of each section of the intelligent power distribution network and the fault, thereby constructing a fault gene table.
Further, the step S7 specifically includes:
s71: setting the gene sequences to be compared as
Figure GDA0002518201230000048
Length of lSThe sequence in the fault gene table is
Figure GDA0002518201230000049
Figure GDA0002518201230000051
Length of lU；
S72: setting a substitution matrix of gene sequence alignment;
s73: the size of the structure is (l)S+1)×(lU+1) a matrix D for storing the comparison result;
Figure GDA0002518201230000052
wherein D (l)i，lj) Denotes the gene sequence to be matched, D (l)i，0)＝D(0，lj)＝0，1≤li≤lS，1≤lj≤lU；
Figure GDA0002518201230000053
Denotes the l-th in the Gene sequence SiThe state of each element is set to be,
Figure GDA0002518201230000054
denotes the l-th of the gene sequence UjA state of the individual element;
s74: finding l in matrix Di *And lj *And satisfies the following conditions:
Figure GDA0002518201230000055
wherein D (l)i *，lj *) Represents the maximum alignment score for sequences S and U;
s75: until the gene sequence to be compared is compared with all the gene sequences of the fault gene table, all the maximum comparison scores are obtained, the maximum comparison scores are compared with a set threshold, and if the maximum comparison scores exceed the threshold, the fault corresponding to the gene is warned.
Further, the Smith-Waterman gene sequence alignment algorithm of step S7 satisfies:
Figure GDA0002518201230000056
where σ represents the importance of the state.
The invention has the beneficial effects that: the method provided by the invention integrates a BP neural network algorithm and a Smith-Waterman gene sequence comparison algorithm, fully excavates the fault characteristics of the intelligent power distribution network, and provides a new thought and solution for fault early warning of the intelligent power distribution network. The method has strong universality and applicability, different fault gene tables can be established to correspond to the intelligent distribution network with different scales, and the different fault gene tables have mutual reference significance. The method can well face the problem that the intelligent power distribution network is more and more complex, the accuracy rate of fault early warning is improved, guidance and help are provided for relevant managers to maintain the intelligent power distribution network, and the scientificity and predictability of power grid operation decision making are effectively improved.
Drawings
In order to make the object, technical scheme and beneficial effect of the invention more clear, the invention provides the following drawings for explanation:
FIG. 1 is a schematic diagram showing the construction process of a fault gene table according to the present invention;
FIG. 2 is a diagram of a three-layer BP neural network model according to the present invention;
FIG. 3 is a representation of a fault gene of the distribution network according to the present invention;
FIG. 4 is a fault pre-warning block diagram based on a fault gene table according to the present invention;
FIG. 5 is a schematic view of the overall process of the present invention.
Detailed Description
Preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
1. Stage of constructing fault gene table
Considering that a certain mapping relation exists between the fault occurring in the intelligent power distribution network and the operation data before the fault, the operation data in a period of time before the fault can be used as a gene for representing the fault on the basis. The method constructs four bases in human genes in an analogy manner, divides the running state of the intelligent power distribution network into a good state, a good state and a neutral state and a poor state, and then converts the running data of the intelligent power distribution network into an ordered state transition time sequence by adopting a BP neural network evaluation model, namely the genes of the intelligent power distribution network, wherein the input of the BP neural network is the state score value of each bus of the intelligent power distribution network. The mapping relation between the transfer time sequence of each segment and the fault can be obtained from a fault historical operation data source of the intelligent power distribution network, so that a fault gene table is constructed.
As shown in FIG. 1, the process of the fault gene table construction of the present invention is:
s1, determining the operation state division of the intelligent power distribution network, and constructing an intelligent power distribution network state evaluation index system.
The operation states of the intelligent power distribution network are divided into four states, namely a good state, a medium state and a poor state, and are marked as E, G, M and B, as shown in the table 1.
TABLE 1 operating status partitioning for smart distribution networks
Operating state of intelligent power distribution network Identification
Superior food E
Good wine G
In M
Difference (D) B
Determining the state evaluation index of the intelligent power distribution network as the state score of each bus of the intelligent power distribution network, wherein the calculation formula of the state score of each bus is as shown in formula (1):
Figure GDA0002518201230000061
wherein G isbScore the bus state with a value of [0, 1%]Closer to 1 indicates better state, and closer to 0 indicates worse state; lambda [ alpha ]VWeight, g, representing the bus voltageVRepresents the bus voltage score, λPWeight, g, representing the active power of the busPRepresenting the bus active power score, λQWeight, g, representing the reactive power of the busQRepresenting the bus reactive power score.
The calculation formula of the voltage score is shown as formula (2):
Figure GDA0002518201230000071
where V is the voltage value of the bus bar,
Figure GDA0002518201230000072
is the average value, V, of the corresponding bus voltage in the historical data setmaxFor the maximum value, V, of the corresponding bus voltage in the historical data setminIs the minimum value of the corresponding bus voltage in the historical data set.
The calculation formula of the active power score is shown as formula (3):
Figure GDA0002518201230000073
where P is the value of the active power of the bus,
Figure GDA0002518201230000074
is the average value, P, of the active power of the corresponding bus in the historical data setmaxIs the maximum value, P, of the active power of the corresponding bus in the historical data setminThe minimum value of the active power of the corresponding bus in the historical data set.
The calculation formula of the reactive power score is shown as formula (4):
Figure GDA0002518201230000075
where Q is the reactive power value of the bus,
Figure GDA0002518201230000076
is the average value, Q, of the reactive power of the corresponding bus in the historical data setmaxFor maximum value, Q, of reactive power of corresponding bus in historical data setminThe minimum value of the reactive power of the corresponding bus in the historical data set.
S2 designs a 3-layer BP neural network model that contains an input layer, a hidden layer, and an output layer, as shown in FIG. 2.
The number of the input layers is equal to the number of buses of the power distribution network, the number of the hidden layers can be the most appropriate number of nodes of the hidden layers according to the Mathlab simulation effect, the output layers are state scores of the power distribution network, and the running state of the power distribution network is determined according to the state scores, as shown in Table 2.
TABLE 2BP neural network output results and State partition rules
Outputting the result State partitioning
0≤y＜0.25 Difference (B)
0.25≤y＜0.5 Middle (M)
0.5≤y＜0.75 Liang (G)
0.75≤y≤1 You (E)
S3, clustering the operation data of each bus of the intelligent power distribution network in part of historical fault data sources to construct training samples.
Clustering the state score data of all buses in the historical data source of the intelligent power distribution network into four types of samples with different states, namely a training sample matrix Tn×m(k) As shown in formula (5):
Figure GDA0002518201230000081
wherein T isn×m(k) For training sample matrix, k is 1, 2, 3, 4 represents training samples of different state classes, n represents the number of training samples, m represents the number of bus bars, t representsnmThe state score of the mth bus of the nth training sample is represented.
S4, inputting the training samples and the corresponding running state of the power distribution network into the BP neural network for training to obtain a BP neural network state evaluation model.
Will train the sample matrix Tn×m(k) Inputting the BP neural network for training, determining the weight and the threshold of each neuron, wherein the training process is as follows:
a. let the connection weight from the input layer to the hidden layer be wijThe connection weight from the hidden layer to the output layer is wjThe threshold from the input layer to the hidden layer is gammajThe threshold from the hidden layer to the output layer is
Figure GDA0002518201230000082
The number of nodes in the hidden layer is N, the learning rate is η, and the expected output is
Figure GDA0002518201230000083
b. Selecting sigmoid type function as hidden layer activation function air entrainer1And output layer activation function air entrainer2As shown in formula (6):
Figure GDA0002518201230000084
c. calculating the input and output of each unit of the hidden layer by using the input t of the input layernmInput layer to hidden layer connection weight wijAnd inputting layer to hidden layer threshold gammaj. Calculating input h of each unit of hidden layerjIs then reused hjBy activating a function air entrainer1Computing the output b of each unit of the hidden layerjAs shown in formula (7):
Figure GDA0002518201230000085
where r 1, 2.., n, indicates that the r-th sample is trained.
d. Output b from the hidden layerjThe connection weight w from the hidden layer to the output layerjAnd hidden layer to output layer threshold
Figure GDA0002518201230000089
And calculating an output result y as shown in equation (8):
Figure GDA0002518201230000086
e. calculating an error, calculating an error e of each training sample according to the formula (9), calculating a generalized error of an output layer, as shown in the formula (10), calculating a generalized error of each unit of the hidden layer, as shown in the formula (11):
Figure GDA0002518201230000087
Figure GDA0002518201230000088
ehj＝wj×eo×bj×(1-bj) (11)
where e is the output error, eo is the output layer generalized error, ehjThe error is generalized for the cells of the hidden layer.
f. Adjusting the connection weight w from hidden layer to output layerjAnd a threshold value
Figure GDA0002518201230000091
As shown in equation (12):
Figure GDA0002518201230000092
w 'of'jAnd
Figure GDA0002518201230000093
the adjusted connection weight value and threshold value from the hidden layer to the output layer.
g. Adjusting the connection weight w from the input layer to the hidden layerijAnd a threshold value gammajAs shown in formula (13):
Figure GDA0002518201230000094
w 'of'ijAnd gamma'jThe adjusted connection weight value and threshold value from the input layer to the hidden layer are obtained.
h. Taking the variable r from 1 to n, finishing training all the training samples, then accumulating the error E of each training sample to calculate a global error E, judging whether the error E reaches a specified error range, and if so, finishing training and recording the current connection weight and threshold; if not, setting the global error E to be zero, and turning to the step c to repeat the learning training.
And S5, inputting the operation data of each bus in the rest historical fault data sources into the trained BP neural network to obtain the mapping relation between the state transition time sequence of the power distribution network and the fault, thereby constructing a fault gene table.
Constructing the rest historical fault data source into an evaluation matrix
Figure GDA0002518201230000096
As shown in equation (14):
Figure GDA0002518201230000095
wherein
Figure GDA0002518201230000097
To evaluate the matrix, n*Indicating the number of samples to be evaluated,
Figure GDA0002518201230000098
denotes the n-th*The evaluation score of the m-th bus of each evaluation sample.
Will evaluate the matrix
Figure GDA0002518201230000099
Inputting the trained BP neural network for evaluation to obtain the mapping between the state transition time sequence of each section of the power distribution network and the fault, thereby constructing a fault gene table, as shown in FIG. 3.
2. Fault early warning stage based on fault gene table
After a fault gene table of the intelligent power distribution network is constructed, periodically acquiring operation data of the intelligent power distribution network in real time, converting the operation data into genes through a BP neural network, matching the genes with the genes in the fault gene table through a Smith-Waterman gene sequence comparison algorithm, obtaining a matching value, and if the matching value reaches a set threshold value, beginning early warning on corresponding faults to be generated.
When the Smith-Waterman algorithm is used for gene sequence matching in organisms, the importance degrees of four bases are the same, so that the score design of a substitution matrix is the same when the substitution matrix is matched with the same bases, and when the substitution matrix is applied to the field of fault early warning of a smart distribution network, the operation state of the smart distribution network is divided into four states with sequentially decreasing performance, namely E, G, M and B, so that the importance degrees of the four states are different in the gene sequence matching process.
In order to meet the online fault early warning requirement of the intelligent power distribution network, the traditional Smith-Waterman algorithm needs to be improved, and two principles need to be followed when a replacement matrix is designed:
when the two states are matched, the worse the performance corresponding to the two states is, the higher the obtained score is;
II when the two states are not matched, the larger the performance difference between the two states is, the more scores are deducted.
The first principle can match the running states with poor performance as much as possible when the states are matched, highlight the importance degree of the running states and improve the accuracy and speed of early warning; the second principle can reduce the negative influence on the accuracy and speed of early warning when the states are not matched.
According to the importance degree of each state and the design principle of the substitution matrix, the design formula of the substitution matrix can be obtained as shown in formula (15):
Figure GDA0002518201230000101
where, σ represents the importance of the state,
Figure GDA0002518201230000102
denotes the l-th in the S Gene sequence mentioned lateriThe state of each element is set to be,
Figure GDA0002518201230000103
denotes the l-th in the sequence of the U Gene mentioned laterjThe state of the individual elements.
As shown in fig. 4, the fault early warning of the present invention is divided into the following steps:
s1, periodically obtaining running state scores of each bus of the current power distribution network, and inputting the running state scores into a BP neural network to obtain a state transition time sequence;
the method comprises the steps of periodically acquiring operation data of the intelligent power distribution network in real time, inputting the operation data into a BP neural network to be converted into a state transition time sequence, namely a gene to be compared, and gradually increasing the length of the gene along with the time.
S2, periodically matching the obtained state transition time sequence with the genes in the fault gene table through a Smith-Waterman gene sequence comparison algorithm to obtain a matching value.
Periodically matching the genes to be compared with the genes in the fault gene table through an improved Smith-Waterman algorithm, wherein the matching process is as follows:
a. setting the gene sequences to be compared as
Figure GDA0002518201230000104
Length of lSThe sequence in the fault gene table is
Figure GDA0002518201230000105
Figure GDA0002518201230000106
Length of lU。
b. Setting a substitution matrix of gene sequence alignment, setting the importance degrees of the four states E, G, M and B of the power distribution network as 1, 2, 3 and 4 respectively, and calculating the substitution matrix according to the formula (15) as shown in Table 3.
TABLE 3 substitution matrix for gene sequence alignment
E G M B
E 1 -1 -2 -3
G -1 4 -1 -2
M -2 -1 9 -1
B -3 -2 -1 16
c. According to the method of dynamic programming, the structure size is (l)S+1)×(lU+1) size matrix D for storing the comparison results, which can be calculated by equations (16) and (17).
D(li，0)＝D(0，lj)＝0 (16)
Figure GDA0002518201230000111
Wherein 1 is less than or equal to li≤ls，1≤lj≤lU；
Figure GDA0002518201230000112
Can be obtained from the equation (15),
Figure GDA0002518201230000113
is composed of
Figure GDA0002518201230000114
The penalty when taking a gap is taken,
Figure GDA0002518201230000115
is composed of
Figure GDA0002518201230000116
Penalty when gaps are taken.
Thus D (l)i，lj) It means the gene sequence to be matched
Figure GDA0002518201230000117
And sequences in the Fault Gene Table
Figure GDA0002518201230000118
Figure GDA0002518201230000119
All possible alignment scores in between.
d. Finding l in dynamic programming matrix Di *And lj *And satisfies the following conditions:
Figure GDA00025182012300001110
wherein D (l)i *，lj *) Represents the maximum alignment score for sequences S and U.
e. Until the gene sequence to be compared is compared with all the gene sequences of the fault gene table, all the maximum comparison scores are obtained, the maximum comparison scores are compared with a set threshold, and if the maximum comparison scores exceed the threshold, the fault corresponding to the gene is warned.
The early warning threshold value is set to be a percentage value of full scores of fault genes, although the threshold values are the same for different fault genes in a gene table, scores corresponding to the threshold values are different, therefore, the early warning threshold value has better adaptability, different percentage values are selected for simulation, and the percentage value when the early warning accuracy is optimal is selected as the early warning threshold value.
Finally, it is noted that the above-mentioned preferred embodiments illustrate rather than limit the invention, and that, although the invention has been described in detail with reference to the above-mentioned preferred embodiments, it will be understood by those skilled in the art that various changes in form and detail may be made therein without departing from the scope of the invention as defined by the appended claims.

Claims (4)

1. A fault gene table-based intelligent power distribution network fault early warning method is characterized by comprising the following steps: the method specifically comprises the following steps:
s1: dividing the running states of the intelligent power distribution network, and constructing an intelligent power distribution network state evaluation index system;
s2: constructing a BP neural network model comprising an input layer, a hidden layer and an output layer;
s3: dividing the historical fault data of the smart power grid into two parts, and clustering the operation data of buses in the historical fault data source of the first part of the smart power grid to construct a training sample;
s4: inputting training samples of a first part of smart power grids and the running states of the power distribution network corresponding to the training samples into a BP neural network model for training to obtain a BP neural network state evaluation model;
s5: inputting operation data of buses in a second part of intelligent power distribution network historical fault data sources into a BP neural network state evaluation model to obtain a mapping relation between a state transition time sequence of the intelligent power distribution network and a fault, and constructing a fault gene table;
s6: periodically acquiring running state data of a bus of the current intelligent power distribution network, and inputting the running state data into a BP neural network state evaluation model to obtain a state transition time sequence of the current intelligent power distribution network;
s7: matching the obtained state transition time sequence of the current smart grid with all genes in a fault gene table through a Smith-Waterman gene sequence comparison algorithm, solving a matching value, comparing the maximum matching value with a set threshold value, and if the maximum matching value exceeds the threshold value, early warning the fault corresponding to the gene;
in S1, the method for constructing the state evaluation index system of the smart distribution network specifically includes:
determining the state evaluation index of the intelligent power distribution network as the state score of each bus of the intelligent power distribution network, wherein the calculation formula of the state score of each bus is as follows:
Figure FDA0002423501320000011
wherein G isbScore the bus state with a value of [0, 1%]Closer to 1 indicates better state, and closer to 0 indicates worse state; lambda [ alpha ]VWeight, g, representing the bus voltageVRepresents the bus voltage score, λPWeight, g, representing the active power of the busPRepresenting the bus active power score, λQWeight, g, representing the reactive power of the busQRepresenting a bus reactive power score;
in S3, constructing a training sample specifically includes: clustering the operation data of the buses in the first part of historical data sources of the intelligent power distribution network into four types of samples with different states,
Figure FDA0002423501320000012
wherein T isn×m(k) For the training sample matrix, k is 1, 2, 3, 4 for different classes of training samples, n for the number of training samples, m for the number of bus bars, tnmRepresenting the state score of the mth bus of the nth training sample;
in S4, the training process specifically includes:
s41: let the connection weight from the input layer to the hidden layer be wijThe connection weight from the hidden layer to the output layer is wjThe threshold from the input layer to the hidden layer is gammajThe threshold from the hidden layer to the output layer is
Figure FDA0002423501320000021
The number of nodes in the hidden layer is N, the learning rate is η, and the expected output is
Figure FDA0002423501320000022
S42: selecting sigmoid type function as hidden layer activation function f1And output layer activation function f2；
Figure FDA0002423501320000023
S43: calculating the input and output of each unit of the hidden layer by using the input t of the input layernmInput layer to hidden layer connection weight wijAnd inputting layer to hidden layer threshold gammajCalculating the input h of each unit of the hidden layerjIs then reused hjBy activating a function f1Computing the output b of each unit of the hidden layerj；
Figure FDA0002423501320000024
Wherein r is 1, 2, …, n, indicating that the r-th sample is trained;
s44: output b from the hidden layerjThe connection weight w from the hidden layer to the output layerjAnd hidden layer to output layer threshold
Figure FDA00024235013200000211
Calculating an output result y;
Figure FDA0002423501320000025
s45: calculating an error;
Figure FDA0002423501320000026
Figure FDA0002423501320000027
ehj＝wj×eo×bj×(1-bj)
where e is the output error, eo is the output layer generalized error, ehjGeneralizing errors for each unit of the hidden layer;
s46: adjusting the connection weight w from hidden layer to output layerjAnd a threshold value
Figure FDA0002423501320000028
Figure FDA0002423501320000029
W 'of'jAnd
Figure FDA00024235013200000210
the adjusted connection weight and threshold value from the hidden layer to the output layer;
s47: adjusting the connection weight w from the input layer to the hidden layerijAnd a threshold value gammaj；
Figure FDA0002423501320000031
W 'of'ijAnd gamma'jConnecting the adjusted input layer to the hidden layer by the weight and the threshold;
s48: taking the variable r from 1 to n, finishing training all the training samples, then accumulating the error E of each training sample to calculate a global error E, judging whether the error E reaches a specified error range, and if so, finishing training and recording the current connection weight and threshold; if not, setting the global error E to be zero, and turning to the step S43 to repeat the learning training;
the S5 specifically includes:
s51: and (3) constructing an evaluation matrix by using historical fault data sources of the second part of intelligent power distribution network:
Figure FDA0002423501320000032
wherein
Figure FDA0002423501320000033
To evaluate the matrix, n*Indicating the number of samples to be evaluated,
Figure FDA0002423501320000034
denotes the n-th*The evaluation score of the m-th bus of each evaluation sample;
s52: will evaluate the matrix
Figure FDA0002423501320000035
Inputting a BP neural network state evaluation model to obtain mapping between state transition time sequences of all sections of the intelligent power distribution network and faults, and accordingly constructing a fault gene table;
the S7 specifically includes:
s71: setting the gene sequences to be compared as
Figure FDA0002423501320000036
Length of lSThe sequence in the fault gene table is
Figure FDA0002423501320000037
Figure FDA0002423501320000038
Length of lU；
S72: setting a substitution matrix of gene sequence alignment;
s73: the size of the structure is (l)S+1)×(lU+1) a matrix D for storing the comparison result;
Figure FDA0002423501320000039
wherein D (l)i，lj) Denotes the gene sequence to be matched, D (l)i，0)＝D(0，lj)＝0，1≤li≤lS，1≤lj≤lU；
Figure FDA00024235013200000310
Denotes the l-th in the Gene sequence SiThe state of each element is set to be,
Figure FDA00024235013200000311
denotes the l-th of the gene sequence UjA state of the individual element;
s74: finding l in matrix Di *And lj *And satisfies the following conditions:
Figure FDA00024235013200000312
wherein D (l)i *，lj *) Represents the maximum alignment score for sequences S and U;
s75: until the gene sequence to be compared is compared with all gene sequences of a fault gene table, all maximum comparison scores are obtained, the maximum comparison scores are compared with a set threshold, and if the maximum comparison scores exceed the threshold, a fault corresponding to the gene is early warned;
in S7, the Smith-Waterman gene sequence alignment algorithm needs to follow two principles when replacing the matrix:
when the two states are matched, the worse the performance corresponding to the two states is, the higher the obtained score is;
II, when the two states are not matched, the larger the performance difference between the two states is, the more the points are deducted;
namely, satisfies:
Figure FDA0002423501320000041
where σ represents the importance of the state.
2. The intelligent power distribution network fault early warning method based on the fault gene table as claimed in claim 1, wherein: the bus voltage score calculation formula is as follows:
Figure FDA0002423501320000042
where V is the voltage value of the bus bar,
Figure FDA0002423501320000043
is the average value, V, of the corresponding bus voltage in the historical data setmaxFor the maximum value, V, of the corresponding bus voltage in the historical data setminIs the minimum value of the corresponding bus voltage in the historical data set.
3. The intelligent power distribution network fault early warning method based on the fault gene table as claimed in claim 1, wherein: the bus active power score calculation formula is as follows:
Figure FDA0002423501320000044
where P is the value of the active power of the bus,
Figure FDA0002423501320000045
is the average value, P, of the active power of the corresponding bus in the historical data setmaxIs the maximum value, P, of the active power of the corresponding bus in the historical data setminThe minimum value of the active power of the corresponding bus in the historical data set.
4. The intelligent power distribution network fault early warning method based on the fault gene table as claimed in claim 1, wherein: the bus reactive power score calculation formula is as follows:
Figure FDA0002423501320000046
where Q is the reactive power value of the bus,
Figure FDA0002423501320000047
is the average value, Q, of the reactive power of the corresponding bus in the historical data setmaxFor maximum value, Q, of reactive power of corresponding bus in historical data setminThe minimum value of the reactive power of the corresponding bus in the historical data set.
CN201710796634.6A 2017-09-06 2017-09-06 Intelligent power distribution network fault early warning method based on fault gene table Active CN107609769B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710796634.6A CN107609769B (en) 2017-09-06 2017-09-06 Intelligent power distribution network fault early warning method based on fault gene table

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710796634.6A CN107609769B (en) 2017-09-06 2017-09-06 Intelligent power distribution network fault early warning method based on fault gene table

Publications (2)

Publication Number Publication Date
CN107609769A CN107609769A (en) 2018-01-19
CN107609769B true CN107609769B (en) 2020-07-31

Family

ID=61057457

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710796634.6A Active CN107609769B (en) 2017-09-06 2017-09-06 Intelligent power distribution network fault early warning method based on fault gene table

Country Status (1)

Country Link
CN (1) CN107609769B (en)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108537415A (en) * 2018-03-20 2018-09-14 深圳市泰和安科技有限公司 A kind of distribution method, the apparatus and system of online safety utilization of electric power
CN112649118A (en) * 2020-11-10 2021-04-13 许继集团有限公司 Transmission line temperature sensing system based on edge calculation
CN114692480A (en) * 2020-12-30 2022-07-01 上海宏力达信息技术股份有限公司 Intelligent switch reliability analysis method and system based on radial basis function neural network
CN115048965B (en) * 2022-08-15 2022-12-13 南方医科大学珠江医院 Method, device, equipment and medium for detecting fault of ultrasonic medical equipment
CN115372759B (en) * 2022-10-19 2023-04-11 国网浙江省电力有限公司温州供电公司 Cable fault diagnosis system and method for secondary circuit of transformer substation

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105225020A (en) * 2015-11-11 2016-01-06 国家电网公司 A kind of running status Forecasting Methodology based on BP neural network algorithm and system
CN106483947A (en) * 2016-09-21 2017-03-08 国网江苏省电力公司南通供电公司 Distribution Running State assessment based on big data and method for early warning
CN106874963A (en) * 2017-03-17 2017-06-20 南京邮电大学 A kind of Fault Diagnosis Method for Distribution Networks and system based on big data technology
CN106991531A (en) * 2017-03-30 2017-07-28 国家电网公司 Power distribution network project construction effect Post-assessment Method based on BP neural network

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105225020A (en) * 2015-11-11 2016-01-06 国家电网公司 A kind of running status Forecasting Methodology based on BP neural network algorithm and system
CN106483947A (en) * 2016-09-21 2017-03-08 国网江苏省电力公司南通供电公司 Distribution Running State assessment based on big data and method for early warning
CN106874963A (en) * 2017-03-17 2017-06-20 南京邮电大学 A kind of Fault Diagnosis Method for Distribution Networks and system based on big data technology
CN106991531A (en) * 2017-03-30 2017-07-28 国家电网公司 Power distribution network project construction effect Post-assessment Method based on BP neural network

Also Published As

Publication number Publication date
CN107609769A (en) 2018-01-19

Similar Documents

Publication Publication Date Title
CN107609769B (en) Intelligent power distribution network fault early warning method based on fault gene table
CN112217202B (en) Distributed new energy, energy storage and power distribution network planning method considering flexibility investment
CN109086913B (en) Power system transient stability assessment method and system based on deep learning
CN110929918B (en) 10kV feeder fault prediction method based on CNN and LightGBM
CN103914735B (en) A kind of fault recognition method and system based on Neural Network Self-learning
CN105303262A (en) Short period load prediction method based on kernel principle component analysis and random forest
CN111445010B (en) Distribution network voltage trend early warning method based on evidence theory fusion quantum network
CN111798095B (en) Power cable state evaluation method based on neural network
CN105226638A (en) Based on the electric power system Calculation of Available Transfer Capability method improving learning aid algorithm
CN111736084A (en) Valve-regulated lead-acid storage battery health state prediction method based on improved LSTM neural network
Barzola-Monteses et al. Energy consumption of a building by using long short-term memory network: a forecasting study
CN108563863B (en) Energy consumption calculation and scheduling method for urban rail transit system
CN110390461B (en) Nonlinear fuzzy language power distribution network node vulnerability evaluation method based on complex network
CN106296044A (en) power system risk scheduling method and system
CN105303466A (en) Intelligent power grid engineering project comprehensive evaluation method based on AHP-GRA
CN103544546A (en) Method for online prediction of quiescent voltage stability limit of electric system
CN115115127A (en) Low-voltage transformer area intelligent regulation and control method and system based on deep learning
CN114777192A (en) Secondary network heat supply autonomous optimization regulation and control method based on data association and deep learning
CN114971090A (en) Electric heating load prediction method, system, equipment and medium
CN114757330A (en) Urban instantaneous water consumption prediction method based on LSTM
CN114429248A (en) Transformer apparent power prediction method
CN117034762A (en) Composite model lithium battery life prediction method based on multi-algorithm weighted sum
CN111612227A (en) Load prediction method based on K-means clustering and bat optimization neural network
CN111756049A (en) Data-driven reactive power optimization method considering loss of real-time measurement information of power distribution network
CN106296444A (en) A kind of annual whole society based on particle cluster algorithm and support vector machine power predicating method

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