CN116008733A - Single-phase grounding fault diagnosis method based on integrated deep neural network - Google Patents

Single-phase grounding fault diagnosis method based on integrated deep neural network Download PDF

Info

Publication number
CN116008733A
CN116008733A CN202310273460.0A CN202310273460A CN116008733A CN 116008733 A CN116008733 A CN 116008733A CN 202310273460 A CN202310273460 A CN 202310273460A CN 116008733 A CN116008733 A CN 116008733A
Authority
CN
China
Prior art keywords
layer
input
fault diagnosis
data
neural network
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202310273460.0A
Other languages
Chinese (zh)
Inventor
张葛祥
马松
杨强
刘启虞
胡洋珩
朱明�
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.)
Chengdu University of Information Technology
Original Assignee
Chengdu University of Information Technology
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 Chengdu University of Information Technology filed Critical Chengdu University of Information Technology
Priority to CN202310273460.0A priority Critical patent/CN116008733A/en
Publication of CN116008733A publication Critical patent/CN116008733A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
    • Y04S10/52—Outage or fault management, e.g. fault detection or location

Landscapes

  • Supply And Distribution Of Alternating Current (AREA)

Abstract

The invention provides a single-phase earth fault diagnosis method based on an integrated deep neural network, which belongs to the technical field of single-phase earth fault diagnosis and comprises the following steps: s1, acquiring historical data and online real-time data of a fault line and a non-fault line; s2, performing missing value processing and standardization processing; s3, dividing the historical data into a training set and a testing set; defining the online real-time data as a verification set; s4, establishing a single-phase grounding fault diagnosis model, inputting a training set and a testing set into the single-phase grounding fault diagnosis model, and performing model training; s5, after model training is finished, storing optimal model parameters; s6, inputting the verification set into a single-phase grounding fault diagnosis model with optimal model parameters, and performing single-phase grounding fault diagnosis to complete fault line selection. The invention can rapidly and effectively judge single-phase earth faults, and can obviously improve the fault line selection accuracy and ensure the reliability of fault line selection results relative to a single neural network model.

Description

Single-phase grounding fault diagnosis method based on integrated deep neural network
Technical Field
The invention relates to the technical field of single-phase grounding fault diagnosis, in particular to a single-phase grounding fault diagnosis method based on an integrated deep neural network.
Background
The 3-35 kV distribution network system generally adopts a neutral point ungrounded system and a neutral point arc suppression coil grounded system. According to statistics, the probability of single-phase earth faults is highest in the distribution network system in China, and the probability of single-phase earth faults is about more than 80% of the distribution network faults. The power company in China can keep running for 1-2 hours when the power distribution network system has single-phase earth fault. However, when the power distribution network system operates in a long-time single-phase earth fault, the fault condition is easily increased, and if the power distribution network system is not maintained in time, the normal operation of the power system is affected, so that the life safety of people is endangered. Therefore, the fault line needs to be rapidly and accurately positioned and cut off, and the method has very important significance.
At present, fault line selection methods of a low-current grounding system are mainly divided into two types, one type is a traditional method, fault characteristics are extracted from fault data by utilizing manual analysis, fault line selection standards are set manually, and fault line selection work is completed. The other type is to automatically learn fault characteristics from a large number of fault data samples and classify the fault characteristics by using a machine learning method, so as to detect a fault line. Most fault line selection methods in the low-current grounding system adopt recording data for analysis, and have the problems of low line selection precision, high cost and the like.
Disclosure of Invention
The invention provides a single-phase grounding fault diagnosis method based on an integrated deep neural network, which aims to solve at least one problem.
The embodiment of the invention discloses a single-phase grounding fault diagnosis method based on an integrated deep neural network, which comprises the following steps of:
s1, acquiring historical data of a fault line and a non-fault line of a power dispatching automation system and online real-time data of the fault line and the non-fault line;
s2, carrying out missing value processing and standardization processing on the historical data and the online real-time data;
s3, dividing 80% of the historical data processed by the S2 into training sets and 20% into test sets; defining the online real-time data processed by the S2 as a verification set;
s4, establishing a single-phase grounding fault diagnosis model based on a Convolutional Neural Network (CNN) combined long-short term memory network (LSTM) combined Attention mechanism (Attention) integrated deep neural network, inputting the training set and the testing set into the single-phase grounding fault diagnosis model, and ending model training after the calculation loss error determination model converges;
the model training process adopts an Early stop method (Early stop) mechanism and a learning rate self-updating strategy;
s5, after model training is finished, storing optimal model parameters;
s6, inputting the verification set into the single-phase grounding fault diagnosis model of the optimal model parameters, and performing single-phase grounding fault diagnosis to complete fault line selection.
In some embodiments, in step S1, the historical data and the online real-time data each include active power, reactive power, three-phase current, and three-phase voltage of the faulty line and the non-faulty line collected by the power dispatching automation system as characteristic data.
In some embodiments, in step S2, the normalization formula of the normalization process is:
Figure SMS_1
；
in the method, in the process of the invention,
Figure SMS_3
is the first
Figure SMS_7
Sample and the first
Figure SMS_9
Characteristic data of the individual parameters;
Figure SMS_2
；
Figure SMS_5
the number of the data samples;
Figure SMS_8
feature dimension for the original sample;
Figure SMS_10
and
Figure SMS_4
respectively the first of the data samples
Figure SMS_6
Average and standard deviation of the individual feature quantities.
In some embodiments, in step S2, the missing value processing uses a linear interpolation method, where the linear interpolation formula is:
Figure SMS_11
；
in the method, in the process of the invention,
Figure SMS_12
missing values for the data samples;
Figure SMS_13
and
Figure SMS_14
data point values adjacent to the missing value;
Figure SMS_15
index that is a missing value;
Figure SMS_16
and
Figure SMS_17
indexing data points adjacent to each other.
In some embodiments, in step S4, the single-phase ground fault diagnosis model includes one input layer, one intermediate layer, and one output layer.
In some embodiments, the intermediate layer comprises:
a convolution layer having a neuron count of 108, a convolution kernel size of 3*3, and an activation function of Relu;
a pooling layer having a filter size of 3*3;
four long-short-term memory network (LSTM) layers with neuron numbers of 64, 128 and 256, respectively;
an Attention mechanism (Attention) layer;
a fully connected layer with a neuron count of 32 and an activation function of normalized exponential function (Softmax).
In some embodiments, the convolution layer is configured to perform sequence feature extraction on input feature data, perform a convolution operation by using convolution check feature data, and perform optimization on the sequence feature extracted by the convolution layer through maximum pooling, where a calculation formula of the convolution layer is:
Figure SMS_18
；
in the method, in the process of the invention,
Figure SMS_19
in order to input the dimensions of the feature data,
Figure SMS_20
in order to fill the value of the filling,
Figure SMS_21
in order to filter the size of the core,
Figure SMS_22
in order to be a step size,
Figure SMS_23
for outputting the dimension of the feature data.
In some embodiments, the LSTM layer is configured to extract a timing characteristic, where the LSTM layer is configured by three gate structures including a forgetting gate, an input gate, and an output gate, the first layer is a forgetting gate, the forgetting gate determines how much useless information in a cell state is discarded, and a calculation formula is as follows:
Figure SMS_24
；
Figure SMS_25
for the output of the forgetting gate,
Figure SMS_26
and
Figure SMS_27
the weight and bias of the forgetting gate are respectively;
Figure SMS_28
for the hidden layer output at the time t-1,
Figure SMS_29
the hidden layer is input at the moment t;
the second layer is an input door, and the input door determines how much information input at present needs to be stored in the current cell state and updates the cell state;
the third layer is an output gate which is multiplied by the cell state at the current moment after the cell state is subjected to nonlinear change of a tangent activation function to obtain a result which needs to be output at present, and the calculation formula is as follows:
Figure SMS_30
；
Figure SMS_31
；
Figure SMS_32
；
Figure SMS_33
；
Figure SMS_34
；
in the method, in the process of the invention,
Figure SMS_38
output for forget gate; o (O) t An output for the output gate; h is a t Outputting hidden layers at the moment t; w (W) C Is the weight; b C To bias, I t For the output of the input gate,
Figure SMS_39
and
Figure SMS_41
the weight and bias of the forgetting gate are respectively;
Figure SMS_36
and
Figure SMS_40
the weight and bias of the input gate are respectively;
Figure SMS_42
and
Figure SMS_43
the weight and bias of the output gate respectively;
Figure SMS_35
for the hidden layer output at the time t-1,
Figure SMS_44
the hidden layer is input at the moment t; tan h is tangentActivating a function;
Figure SMS_45
and
Figure SMS_46
cell states at times t-1 and t, respectively;
Figure SMS_37
is the current candidate input vector.
In some embodiments, the Attention mechanism (Attention) layer is configured to input the sequence features extracted by the LSTM layer into an Attention mechanism (Attention), and perform Attention distribution calculation on the input feature data by the Attention mechanism (Attention), and calculate a weighted average of the input feature data according to the Attention distribution, where the Attention distribution calculation formula is:
Figure SMS_47
；
wherein p is a correlation; z is
Figure SMS_48
Is a distribution of (3); m is m groups of input characteristic information;
Figure SMS_49
inputting characteristic information;
Figure SMS_50
is a query vector;
Figure SMS_51
is the attention distribution;
Figure SMS_52
the function converts the value to [0,1 ]]Probability distribution of (2);
Figure SMS_53
the attention scoring function is calculated by the following formula:
Figure SMS_54
；
wherein x is input characteristic information; x is x T Is the transpose of x;
Figure SMS_55
as a parameter that can be learned,
Figure SMS_56
is a query vector;
the weight weighted average calculation formula of the input characteristic data is as follows;
Figure SMS_57
；
in the method, in the process of the invention,
Figure SMS_58
is the attention distribution;
Figure SMS_59
is the m-th characteristic information; q is a query vector;
Figure SMS_60
for inputting the characteristic information.
In some embodiments, in step S4, the calculation formula of the learning rate self-update strategy is:
Figure SMS_61
；
in the method, in the process of the invention,
Figure SMS_62
is the initial learning rate;
Figure SMS_63
the updated learning rate;
Figure SMS_64
the factor is self-updated for the learning rate.
In summary, the invention has at least the following advantages:
according to the invention, a single-phase grounding fault diagnosis model based on a CNN-LSTM-Attention integrated deep neural network is established, and historical data and online real-time data of fault lines and non-fault lines of a power dispatching automation system are collected to diagnose single-phase grounding faults in a power distribution network system. The fault line can be detected more accurately through online real-time data verification of the power dispatching automation system, and the power dispatching automation system has good practicability; by collecting data in the power dispatching automation system, the data is sourced from a real dispatching automation platform, no additional data collection equipment is needed, and investment cost is reduced; meanwhile, the invention can rapidly and effectively judge single-phase earth faults, and compared with a single neural network model, the invention can obviously improve the fault line selection accuracy and ensure the reliability of fault line selection results.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic diagram of the steps of a CNN-LSTM-Attention based integrated deep neural network based single-phase earth fault diagnosis method according to the present invention.
FIG. 2 is a schematic diagram of a CNN-LSTM-Attention integrated deep neural network in accordance with the present invention.
FIG. 3 is a flow chart of a single-phase grounding fault diagnosis method based on a CNN-LSTM-Attention integrated deep neural network.
FIG. 4 is a schematic diagram of a loss iteration of the CNN-LSTM-Attention integrated deep neural network training process involved in the present invention.
Detailed Description
Hereinafter, only certain exemplary embodiments are briefly described. As will be recognized by those of skill in the pertinent art, the described embodiments may be modified in numerous different ways without departing from the spirit or scope of the embodiments of the present invention. Accordingly, the drawings and description are to be regarded as illustrative in nature and not as restrictive.
The following disclosure provides many different implementations, or examples, for implementing different configurations of embodiments of the invention. In order to simplify the disclosure of embodiments of the present invention, components and arrangements of specific examples are described below. Of course, they are merely examples and are not intended to limit embodiments of the present invention. Furthermore, embodiments of the present invention may repeat reference numerals and/or letters in the various examples, which are for the purpose of brevity and clarity, and which do not themselves indicate the relationship between the various embodiments and/or arrangements discussed.
Embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
As shown in fig. 1 and 3, the embodiment of the invention discloses a single-phase grounding fault diagnosis method based on an integrated deep neural network, which comprises the following steps:
s1, historical data of a fault line and a non-fault line of the power dispatching automation system and online real-time data of the fault line and the non-fault line are obtained.
The historical data and the online real-time data comprise active power, reactive power, three-phase current and three-phase voltage of a fault line and a non-fault line which are acquired by the power dispatching automation system, and the active power, the reactive power, the three-phase current and the three-phase voltage are used as characteristic data for subsequent steps.
S2, performing missing value processing and standardization processing on the historical data and the online real-time data, so that the follow-up neural network can be ensured to be effectively trained, and the model training efficiency is improved.
The method comprises the steps of carrying out standardization processing on active power, reactive power, three-phase current and three-phase voltage values, wherein a standardized formula of the standardization processing is as follows:
Figure SMS_65
；
in the method, in the process of the invention,
Figure SMS_66
is the first
Figure SMS_69
Sample and the first
Figure SMS_72
Characteristic data of the individual parameters;
Figure SMS_68
；
Figure SMS_71
the number of the data samples;
Figure SMS_73
feature dimension for the original sample;
Figure SMS_74
and
Figure SMS_67
respectively the first of the data samples
Figure SMS_70
Average and standard deviation of the individual feature quantities.
Carrying out missing value processing on original data (collected historical data and online real-time data), carrying out numerical value estimation according to two adjacent data points on the left and right of a point needing interpolation in a one-dimensional data sequence by adopting a linear interpolation method, wherein a linear interpolation formula is as follows:
Figure SMS_75
；
in the method, in the process of the invention,
Figure SMS_76
missing values for the data samples;
Figure SMS_77
and
Figure SMS_78
data point values adjacent to the missing value;
Figure SMS_79
index that is a missing value;
Figure SMS_80
and
Figure SMS_81
indexing data points adjacent to each other.
S3, dividing 80% of the historical data processed by the S2 into training sets and 20% into test sets. And (5) defining the online real-time data processed by the S2 as a verification set.
For the collected historical data and online real-time data, one sample is formed by every 50 lines of one-dimensional time sequence data, and 5200 groups of fault and non-fault samples are collected together.
S4, establishing a single-phase grounding fault diagnosis model based on the CNN-LSTM-Attention integrated deep neural network, inputting a training set and a testing set into the single-phase grounding fault diagnosis model, calculating loss errors, optimizing model parameters, and ending model training after model convergence is determined. If the model does not converge, model training is continued until the model converges. As shown in fig. 2 and 4.
And constructing a CNN-LSTM-Attention integrated deep neural network through a Pytorch framework.
The model training process adopts an Early stop method (Early stop) mechanism and a learning rate self-updating strategy;
the single-phase earth fault diagnosis model comprises an input layer, one or more intermediate layers and an output layer.
The intermediate layer comprises:
a convolution layer having a neuron count of 108, a convolution kernel size of 3*3, and an activation function of Relu;
a pooling layer having a filter size of 3*3;
four LSTM layers with neuron numbers of 64, 128 and 256, respectively;
an Attention mechanism (Attention) layer;
a fully connected layer with a neuron count of 32 and an activation function of Softmax.
The number of the neurons of the output layer is 2, and the neurons are used for outputting the single-phase grounding fault line selection result.
The convolution layer is used for extracting sequence features of the input feature data, carrying out convolution operation by utilizing the convolution check feature data, optimizing the sequence features extracted by the convolution layer through maximum pooling, and the calculation formula of the convolution layer is as follows:
Figure SMS_82
；
in the method, in the process of the invention,
Figure SMS_83
in order to input the dimensions of the feature data,
Figure SMS_84
in order to fill the value of the filling,
Figure SMS_85
in order to filter the size of the core,
Figure SMS_86
in order to be a step size,
Figure SMS_87
for outputting the dimension of the feature data.
The LSTM layer is used for extracting time sequence characteristics, and is composed of three gate structures of a forgetting gate, an input gate and an output gate, wherein the first layer is the forgetting gate, the forgetting gate determines how much useless information in the cell state is discarded, and the calculation formula is as follows:
Figure SMS_88
；
Figure SMS_89
for the output of the forgetting gate,
Figure SMS_90
and
Figure SMS_91
the weight and bias of the forgetting gate are respectively;
Figure SMS_92
for the hidden layer output at the time t-1,
Figure SMS_93
the hidden layer is input at the moment t;
the second layer is an input door, and the input door determines how much information input at present needs to be stored in the current cell state and updates the cell state;
the third layer is an output gate which is multiplied by the cell state at the current moment after the cell state is subjected to nonlinear change of a tangent activation function to obtain a result which needs to be output at present, and the calculation formula is as follows:
Figure SMS_94
；
Figure SMS_95
；
Figure SMS_96
；
Figure SMS_97
；
Figure SMS_98
；
in the method, in the process of the invention,
Figure SMS_102
output for forget gate; o (O) t An output for the output gate; h is a t Outputting hidden layers at the moment t; w (W) C Is the weight; b C To bias, I t For the output of the input gate,
Figure SMS_104
and
Figure SMS_107
the weight and bias of the forgetting gate are respectively;
Figure SMS_100
and
Figure SMS_103
the weight and bias of the input gate are respectively;
Figure SMS_106
and
Figure SMS_109
the weight and bias of the output gate respectively;
Figure SMS_99
for the hidden layer output at the time t-1,
Figure SMS_105
the hidden layer is input at the moment t; tanh is the tangent activation function;
Figure SMS_108
and
Figure SMS_110
cell states at times t-1 and t, respectively;
Figure SMS_101
is the current candidate input vector.
The Attention mechanism (Attention) layer is used for inputting the sequence features extracted by the LSTM layer into the Attention mechanism (Attention) for processing, the Attention mechanism (Attention) layer performs Attention distribution calculation on the input feature data, and weight weighted average of the input feature data is calculated according to the Attention distribution, and an Attention distribution calculation formula is as follows:
Figure SMS_111
；
wherein p is a correlation; z is
Figure SMS_112
Is a distribution of (3); m is m groups of inputsCharacteristic information;
Figure SMS_113
inputting characteristic information;
Figure SMS_114
is a query vector;
Figure SMS_115
is the attention distribution;
Figure SMS_116
the function converts the value to [0,1 ]]Probability distribution of (2);
Figure SMS_117
the attention scoring function is calculated by the following formula:
Figure SMS_118
；
wherein x is input characteristic information; x is x T Is the transpose of x;
Figure SMS_119
as a parameter that can be learned,
Figure SMS_120
is a query vector;
the weight weighted average calculation formula of the input characteristic data is as follows:
Figure SMS_121
；
in the method, in the process of the invention,
Figure SMS_122
is the attention distribution;
Figure SMS_123
is the m-th characteristic information; q is a query vector;
Figure SMS_124
for inputting the characteristic information.
The parameters in Early stop (Early stop) mechanism are set to 50, and delta is set to
Figure SMS_125
The learning rate self-updating parameter is set to be 0.35, the parameter is 20, and the min_lr is 0.0001. The calculation formula of the learning rate self-updating strategy is as follows:
Figure SMS_126
；
in the method, in the process of the invention,
Figure SMS_127
is the initial learning rate;
Figure SMS_128
the updated learning rate;
Figure SMS_129
the factor is self-updated for the learning rate.
S5, storing optimal model parameters after model training is finished.
In the model training process, the loss function is a cross entropy loss function, and the calculation formula is as follows:
Figure SMS_130
；
wherein N is the number of training samples;
Figure SMS_131
a probability value representing that the nth sample is in the c-th class;
Figure SMS_132
is the best model parameter.
S6, inputting the verification set into a single-phase grounding fault diagnosis model with optimal model parameters, and performing single-phase grounding fault diagnosis to complete fault line selection.
And evaluating the single-phase grounding fault diagnosis method based on the integrated deep neural network according to three evaluation indexes of Accuracy (Accuracy), precision and Recall. The calculation formulas of the accuracy rate, the precision rate and the recall rate are as follows:
Figure SMS_133
；
Figure SMS_134
；
Figure SMS_135
；
in the method, in the process of the invention,
Figure SMS_136
representing the number of faulty samples in all sample data;
Figure SMS_137
representing the number of normal samples in all sample data;
Figure SMS_138
representing predicting the failure samples as the number of failure samples;
Figure SMS_139
representing the number of normal samples predicted as normal samples;
Figure SMS_140
indicating the number of normal samples predicted as faulty samples.
The comparison of the Accuracy (Accuracy), precision (Precision) and Recall (Recall) of the different methods is shown in table 1 and fig. 4. As can be seen from Table 1, the values of the Accuracy (Accuracy), precision (Precision) and Recall (Recall) are close to 1, which shows that the single-phase earth fault diagnosis method of the invention has high Precision and can more accurately detect the fault line.
Figure SMS_141
The above-described embodiments are provided to illustrate the present invention, not to limit the present invention,alterations of the example values or substitutions of equivalent elements are intended to be within the scope of the present invention. From the foregoing detailed description, it will be apparent to those skilled in the art that the present invention can be practiced without these specific details, and that the present invention meets the requirements of the patent statutes.
While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. It is therefore intended that the following claims be interpreted as including the preferred embodiments and all such alterations and modifications as fall within the scope of the invention. The foregoing description of the preferred embodiment of the invention is not intended to be limiting, but rather to cover all modifications, equivalents, and alternatives falling within the spirit and principles of the invention.
It should be noted that the above description of the flow is only for the purpose of illustration and description, and does not limit the application scope of the present specification. Various modifications and changes to the flow may be made by those skilled in the art under the guidance of this specification. However, such modifications and variations are still within the scope of the present description.
While the basic concepts have been described above, it will be apparent to those of ordinary skill in the art after reading this application that the above disclosure is by way of example only and is not limiting of the present application. Although not explicitly described herein, various modifications, improvements, and adaptations of the present application are possible for those of ordinary skill in the art. Such modifications, improvements, and modifications are intended to be suggested within this application, and are therefore within the spirit and scope of the exemplary embodiments of this application.
Meanwhile, the present application uses specific words to describe embodiments of the present application. For example, "one embodiment," "an embodiment," and/or "some embodiments" means a particular feature, structure, or characteristic in connection with at least one embodiment of the present application. Thus, it should be emphasized and should be appreciated that two or more references to "an embodiment" or "one embodiment" or "an alternative embodiment" in various positions in this specification are not necessarily referring to the same embodiment. Furthermore, certain features, structures, or characteristics of one or more embodiments of the present application may be combined as suitable.
Furthermore, those of ordinary skill in the art will appreciate that aspects of the invention may be illustrated and described in terms of several patentable categories or circumstances, including any novel and useful processes, machines, products, or materials, or any novel and useful improvements thereof. Thus, aspects of the present application may be implemented entirely in hardware, entirely in software (including firmware, resident software, pseudo-code, etc.) or a combination of hardware and software. The above hardware or software may be referred to as a "unit," module, "or" system. Furthermore, aspects of the present application may take the form of a computer program product embodied in one or more computer-readable media, wherein the computer-readable program code is embodied therein.
Computer program code required for operation of portions of the present application may be written in any one or more programming languages, including an object oriented programming language such as Java, scala, smalltalk, eiffel, JADE, emerald, C ++, c#, vb.net, python, etc., a conventional programming language such as C programming language, visualBasic, fortran2103, perl, COBOL2102, PHP, ABAP, a dynamic programming language such as Python, ruby, and Groovy, or other programming languages, etc. The program code may execute entirely on the user's computer, or as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer may be connected to the user's computer through any form of network, such as a Local Area Network (LAN) or a Wide Area Network (WAN), or the connection may be made to an external computer (for example, through the Internet), or the use of services such as software as a service (SaaS) in a cloud computing environment.
Furthermore, the order in which the elements and sequences are presented, the use of numerical letters, or other designations are used in the application and are not intended to limit the order in which the processes and methods of the application are performed unless explicitly recited in the claims. While certain presently useful inventive embodiments have been discussed in the foregoing disclosure, by way of various examples, it is to be understood that such details are merely illustrative and that the appended claims are not limited to the disclosed embodiments, but, on the contrary, are intended to cover all modifications and equivalent arrangements included within the spirit and scope of the embodiments of the present application. For example, while the implementation of the various components described above may be embodied in a hardware device, it may also be implemented as a purely software solution, e.g., an installation on an existing server or mobile device.
Likewise, it should be noted that in order to simplify the presentation disclosed herein and thereby aid in understanding one or more inventive embodiments, various features are sometimes grouped together in a single embodiment, figure, or description thereof for the purpose of streamlining the disclosure. This method of disclosure, however, is not to be interpreted as reflecting an intention that the claimed subject matter requires more features than are expressly recited in each claim. Rather, the inventive subject matter should be provided with fewer features than the single embodiments described above.

Claims (10)

1. The single-phase earth fault diagnosis method based on the integrated deep neural network is characterized by comprising the following steps of:
s1, acquiring historical data of a fault line and a non-fault line of a power dispatching automation system and online real-time data of the fault line and the non-fault line;
s2, carrying out missing value processing and standardization processing on the historical data and the online real-time data;
s3, dividing 80% of the historical data processed by the S2 into training sets and 20% into test sets; defining the online real-time data processed by the S2 as a verification set;
s4, establishing a single-phase grounding fault diagnosis model based on a CNN-LSTM-Attention integrated deep neural network, inputting the training set and the testing set into the single-phase grounding fault diagnosis model, and ending model training after calculating loss error determination model convergence;
the model training process adopts an early-stop method mechanism and a learning rate self-updating strategy;
s5, after model training is finished, storing optimal model parameters;
s6, inputting the verification set into the single-phase grounding fault diagnosis model of the optimal model parameters, and performing single-phase grounding fault diagnosis to complete fault line selection.
2. The integrated deep neural network-based single-phase earth fault diagnosis method according to claim 1, wherein in step S1, the history data and the online real-time data each include active power, reactive power, three-phase current and three-phase voltage of a fault line and a non-fault line collected by a power dispatching automation system, as characteristic data.
3. The integrated deep neural network-based single-phase earth fault diagnosis method according to claim 1, wherein in step S2, the normalization formula of the normalization process is:
Figure QLYQS_1
；
in the method, in the process of the invention,
Figure QLYQS_3
is->
Figure QLYQS_7
Sample and->
Figure QLYQS_8
Characteristic data of the individual parameters; />
Figure QLYQS_4
；/>
Figure QLYQS_6
The number of the data samples; />
Figure QLYQS_9
Feature dimension for the original sample; />
Figure QLYQS_10
And->
Figure QLYQS_2
Respectively the +.>
Figure QLYQS_5
Average and standard deviation of the individual feature quantities.
4. The method for diagnosing a single-phase earth fault based on an integrated deep neural network according to claim 1, wherein in step S2, the missing value processing adopts a linear interpolation method, and a linear interpolation formula is:
Figure QLYQS_11
；
in the method, in the process of the invention,
Figure QLYQS_12
missing values for the data samples; />
Figure QLYQS_13
And->
Figure QLYQS_14
Data point values adjacent to the missing value; />
Figure QLYQS_15
Index that is a missing value; />
Figure QLYQS_16
And->
Figure QLYQS_17
Indexing data points adjacent to each other.
5. The integrated deep neural network-based single-phase ground fault diagnosis method according to claim 1, wherein in step S4, the single-phase ground fault diagnosis model includes one input layer, one intermediate layer, and one output layer.
6. The integrated deep neural network based single phase earth fault diagnosis method of claim 5, wherein the intermediate layer comprises:
a convolution layer having a neuron count of 108, a convolution kernel size of 3*3, and an activation function of Relu;
a pooling layer having a filter size of 3*3;
four LSTM layers with neuron numbers of 64, 128 and 256, respectively;
an Attention layer;
a fully connected layer with a neuron count of 32 and an activation function of Softmax.
7. The integrated deep neural network-based single-phase earth fault diagnosis method according to claim 6, wherein the convolution layer is configured to perform sequence feature extraction on input feature data, perform convolution operation by using convolution check feature data, and perform optimization on the sequence feature extracted by the convolution layer through maximum pooling, and the calculation formula of the convolution layer is as follows:
Figure QLYQS_18
；
in the method, in the process of the invention,
Figure QLYQS_19
for the dimension of the input feature data +.>
Figure QLYQS_20
For filling value, ++>
Figure QLYQS_21
For the size of the filter kernel +.>
Figure QLYQS_22
For step size->
Figure QLYQS_23
For outputting the dimension of the feature data.
8. The method for diagnosing a single-phase earth fault based on an integrated deep neural network according to claim 6, wherein the LSTM layer is used for extracting timing characteristics, the LSTM layer is composed of three gate structures of a forgetting gate, an input gate and an output gate, the first layer is the forgetting gate, the forgetting gate determines how much useless information in a cell state is discarded, and a calculation formula is as follows:
Figure QLYQS_24
；
Figure QLYQS_25
output for forgetting gate, ++>
Figure QLYQS_26
And->
Figure QLYQS_27
The weight and bias of the forgetting gate are respectively; />
Figure QLYQS_28
For the hidden layer output at t-1 time, +.>
Figure QLYQS_29
The hidden layer is input at the moment t;
the second layer is an input door, and the input door determines how much information input at present needs to be stored in the current cell state and updates the cell state;
the third layer is an output gate which is multiplied by the cell state at the current moment after the cell state is subjected to nonlinear change of a tangent activation function to obtain a result which needs to be output at present, and the calculation formula is as follows:
Figure QLYQS_30
；
Figure QLYQS_31
；
Figure QLYQS_32
；
Figure QLYQS_33
；
Figure QLYQS_34
；
in the method, in the process of the invention,
Figure QLYQS_36
output for forget gate; o (O) t An output for the output gate; h is a t Outputting hidden layers at the moment t; w (W) C Is the weight; b C To bias, I t For the output of the input gate, +.>
Figure QLYQS_40
And->
Figure QLYQS_43
The weight and bias of the forgetting gate are respectively; />
Figure QLYQS_35
And->
Figure QLYQS_39
The weight and bias of the input gate are respectively; />
Figure QLYQS_42
And->
Figure QLYQS_45
The weight and bias of the output gate respectively; />
Figure QLYQS_38
For the hidden layer output at t-1 time, +.>
Figure QLYQS_41
The hidden layer is input at the moment t; tanh is the tangent activation function; />
Figure QLYQS_44
And->
Figure QLYQS_46
Cell states at times t-1 and t, respectively; />
Figure QLYQS_37
Is the current candidate input vector.
9. The integrated deep neural network-based single-phase earth fault diagnosis method according to claim 7, wherein the Attention layer is configured to input the sequence features extracted by the LSTM layer into an Attention for processing, the Attention layer performs Attention distribution calculation on input feature data, and calculates a weighted average of the input feature data according to the Attention distribution, where the Attention distribution calculation formula is:
Figure QLYQS_47
；
wherein p is a correlation; z is
Figure QLYQS_48
Is a distribution of (3); m is m groups of input characteristic information; />
Figure QLYQS_49
Inputting characteristic information; />
Figure QLYQS_50
Is a query vector; />
Figure QLYQS_51
Is the attention distribution; />
Figure QLYQS_52
The function converts the value to [0,1 ]]Probability distribution of (2); />
Figure QLYQS_53
The attention scoring function is calculated by the following formula:
Figure QLYQS_54
；
wherein x is input characteristic information; x is x T Is the transpose of x;
Figure QLYQS_55
for a learnable parameter->
Figure QLYQS_56
Is a query vector;
the weight weighted average calculation formula of the input characteristic data is as follows:
Figure QLYQS_57
；
in the method, in the process of the invention,
Figure QLYQS_58
is the attention distribution; />
Figure QLYQS_59
Is the m-th characteristic information; q is a query vector; />
Figure QLYQS_60
For inputting the characteristic information.
10. The method for diagnosing a single-phase earth fault based on an integrated deep neural network according to claim 1, wherein in step S4, a calculation formula of the learning rate self-update strategy is:
Figure QLYQS_61
；
in the method, in the process of the invention,
Figure QLYQS_62
is the initial learning rate; />
Figure QLYQS_63
The updated learning rate; />
Figure QLYQS_64
The factor is self-updated for the learning rate. />
CN202310273460.0A 2023-03-21 2023-03-21 Single-phase grounding fault diagnosis method based on integrated deep neural network Pending CN116008733A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310273460.0A CN116008733A (en) 2023-03-21 2023-03-21 Single-phase grounding fault diagnosis method based on integrated deep neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310273460.0A CN116008733A (en) 2023-03-21 2023-03-21 Single-phase grounding fault diagnosis method based on integrated deep neural network

Publications (1)

Publication Number Publication Date
CN116008733A true CN116008733A (en) 2023-04-25

Family

ID=86035857

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310273460.0A Pending CN116008733A (en) 2023-03-21 2023-03-21 Single-phase grounding fault diagnosis method based on integrated deep neural network

Country Status (1)

Country Link
CN (1) CN116008733A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117826019A (en) * 2024-03-06 2024-04-05 国网吉林省电力有限公司长春供电公司 Line single-phase grounding fault area and type detection method of neutral point ungrounded system

Citations (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103760440A (en) * 2014-01-16 2014-04-30 广东电网公司电力科学研究院 Power system fault monitoring method and system
CN104537908A (en) * 2014-12-17 2015-04-22 国电南瑞科技股份有限公司 Multi-stage scheduling integrated simulation system based on model sharing and method
CN106771846A (en) * 2016-11-08 2017-05-31 西华大学 Power transmission line fault phase selection based on fuzzy reasoning pulse nerve membranous system
CN106940411A (en) * 2017-04-21 2017-07-11 国网江苏省电力公司宿迁供电公司 A kind of single-phase ground fault line selecting method of small-electric current grounding system
CN107907792A (en) * 2017-11-03 2018-04-13 国网江苏省电力公司新沂市供电公司 Neutral by arc extinction coil grounding ring distribution system single-phase grounding selecting method
CN109978134A (en) * 2019-02-26 2019-07-05 华中科技大学 A kind of failure prediction method based on fast integration convolutional neural networks
CN111898446A (en) * 2020-09-04 2020-11-06 国电南瑞科技股份有限公司 Single-phase earth fault studying and judging method based on multi-algorithm normalization analysis
CN111985155A (en) * 2020-08-10 2020-11-24 武汉大学 Circuit health state prediction method and system based on integrated deep neural network
CN112149554A (en) * 2020-09-21 2020-12-29 广东电网有限责任公司清远供电局 Fault classification model training method, fault detection method and related device
CN112200032A (en) * 2020-09-28 2021-01-08 辽宁石油化工大学 Attention mechanism-based high-voltage circuit breaker mechanical state online monitoring method
CN112364973A (en) * 2020-08-05 2021-02-12 华侨大学 Irrelevant multi-source frequency domain load identification method based on neural network and model transfer learning
CN112396234A (en) * 2020-11-20 2021-02-23 国网经济技术研究院有限公司 User side load probability prediction method based on time domain convolutional neural network
CN112541233A (en) * 2019-09-20 2021-03-23 宫文峰 Rotary machine fault diagnosis method based on improved convolutional neural network
CN112633317A (en) * 2020-11-02 2021-04-09 国能信控互联技术有限公司 CNN-LSTM fan fault prediction method and system based on attention mechanism
US20210190882A1 (en) * 2019-12-10 2021-06-24 Wuhan University Transformer failure identification and location diagnosis method based on multi-stage transfer learning
CN113625115A (en) * 2021-08-16 2021-11-09 广西电网有限责任公司 Low-current ground fault line selection system based on scheduling data
CN114563671A (en) * 2022-03-03 2022-05-31 海南电网有限责任公司屯昌供电局 High-voltage cable partial discharge diagnosis method based on CNN-LSTM-Attention neural network
CN115267428A (en) * 2022-07-20 2022-11-01 福州大学 LCC-MMC single-pole grounding fault positioning method based on VMD-ET feature selection

Patent Citations (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103760440A (en) * 2014-01-16 2014-04-30 广东电网公司电力科学研究院 Power system fault monitoring method and system
CN104537908A (en) * 2014-12-17 2015-04-22 国电南瑞科技股份有限公司 Multi-stage scheduling integrated simulation system based on model sharing and method
CN106771846A (en) * 2016-11-08 2017-05-31 西华大学 Power transmission line fault phase selection based on fuzzy reasoning pulse nerve membranous system
CN106940411A (en) * 2017-04-21 2017-07-11 国网江苏省电力公司宿迁供电公司 A kind of single-phase ground fault line selecting method of small-electric current grounding system
CN107907792A (en) * 2017-11-03 2018-04-13 国网江苏省电力公司新沂市供电公司 Neutral by arc extinction coil grounding ring distribution system single-phase grounding selecting method
CN109978134A (en) * 2019-02-26 2019-07-05 华中科技大学 A kind of failure prediction method based on fast integration convolutional neural networks
CN112541233A (en) * 2019-09-20 2021-03-23 宫文峰 Rotary machine fault diagnosis method based on improved convolutional neural network
US20210190882A1 (en) * 2019-12-10 2021-06-24 Wuhan University Transformer failure identification and location diagnosis method based on multi-stage transfer learning
CN112364973A (en) * 2020-08-05 2021-02-12 华侨大学 Irrelevant multi-source frequency domain load identification method based on neural network and model transfer learning
CN111985155A (en) * 2020-08-10 2020-11-24 武汉大学 Circuit health state prediction method and system based on integrated deep neural network
US20220043955A1 (en) * 2020-08-10 2022-02-10 Wuhan University Circuit health state prediction method and system based on integrated deep neural network
CN111898446A (en) * 2020-09-04 2020-11-06 国电南瑞科技股份有限公司 Single-phase earth fault studying and judging method based on multi-algorithm normalization analysis
CN112149554A (en) * 2020-09-21 2020-12-29 广东电网有限责任公司清远供电局 Fault classification model training method, fault detection method and related device
CN112200032A (en) * 2020-09-28 2021-01-08 辽宁石油化工大学 Attention mechanism-based high-voltage circuit breaker mechanical state online monitoring method
CN112633317A (en) * 2020-11-02 2021-04-09 国能信控互联技术有限公司 CNN-LSTM fan fault prediction method and system based on attention mechanism
CN112396234A (en) * 2020-11-20 2021-02-23 国网经济技术研究院有限公司 User side load probability prediction method based on time domain convolutional neural network
CN113625115A (en) * 2021-08-16 2021-11-09 广西电网有限责任公司 Low-current ground fault line selection system based on scheduling data
CN114563671A (en) * 2022-03-03 2022-05-31 海南电网有限责任公司屯昌供电局 High-voltage cable partial discharge diagnosis method based on CNN-LSTM-Attention neural network
CN115267428A (en) * 2022-07-20 2022-11-01 福州大学 LCC-MMC single-pole grounding fault positioning method based on VMD-ET feature selection

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
姜文晖: "《物体检索与定位》", 中国铁道出版社有限公司, pages: 72 - 13 *
柳秀 等: "面向轴承故障诊断的深度学习方法", 《哈尔滨理工大学学报》, vol. 27, no. 5, pages 118 - 124 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117826019A (en) * 2024-03-06 2024-04-05 国网吉林省电力有限公司长春供电公司 Line single-phase grounding fault area and type detection method of neutral point ungrounded system

Similar Documents

Publication Publication Date Title
CN103914064B (en) Based on the commercial run method for diagnosing faults that multi-categorizer and D-S evidence merge
CN109344517A (en) A kind of high-voltage isulation method for diagnosing faults of new-energy automobile
CN111780800B (en) Method and system for monitoring, isolating and reconstructing sensor fault
CN113298297B (en) Wind power output power prediction method based on isolated forest and WGAN network
CN105137354B (en) One kind is based on neutral net electrical fault detection method
CN109948194B (en) High-voltage circuit breaker mechanical defect integrated learning diagnosis method
CN113642754A (en) Complex industrial process fault prediction method based on RF noise reduction self-coding information reconstruction and time convolution network
CN116008733A (en) Single-phase grounding fault diagnosis method based on integrated deep neural network
CN113538353B (en) Five-phase asynchronous motor rolling bearing fault diagnosis method based on single-channel diagram data enhancement and migration training residual error network
CN116226646A (en) Method, system, equipment and medium for predicting health state and residual life of bearing
CN110865924A (en) Health degree diagnosis method and health diagnosis framework for internal server of power information system
CN111967620A (en) Photovoltaic module diagnosis method, device, equipment and readable storage medium
CN113987294A (en) CVT (continuously variable transmission) online fault diagnosis method based on genetic optimization GRU (generalized regression Unit) neural network
CN114611372A (en) Industrial equipment health prediction method based on Internet of things edge calculation
CN113011248A (en) Nuclear power plant fault diagnosis method based on LSTM
CN105488270B (en) A kind of more attribute synthesis methods of transformer device structure fault diagnosis
CN112595918A (en) Low-voltage meter reading fault detection method and device
CN105741184A (en) Transformer state evaluation method and apparatus
CN111797944A (en) Vehicle door abnormity diagnosis method and device
CN111459697A (en) Excitation system fault monitoring method based on deep learning network
CN114137915A (en) Fault diagnosis method for industrial equipment
CN113033845B (en) Construction method and device for power transmission resource co-construction and sharing
CN114564877A (en) Rolling bearing service life prediction method, system, equipment and readable storage medium
CN113705637A (en) Method, system and equipment for detecting mechanical fault of circuit breaker and readable storage medium
CN115556099B (en) Sustainable learning industrial robot fault diagnosis system and 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