CN118040679B - Disturbance and fault identification method for in-phase power supply system - Google Patents
Disturbance and fault identification method for in-phase power supply system Download PDFInfo
- Publication number
- CN118040679B CN118040679B CN202410430549.8A CN202410430549A CN118040679B CN 118040679 B CN118040679 B CN 118040679B CN 202410430549 A CN202410430549 A CN 202410430549A CN 118040679 B CN118040679 B CN 118040679B
- Authority
- CN
- China
- Prior art keywords
- attribute
- power supply
- phase power
- disturbance
- fault
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 47
- 238000000605 extraction Methods 0.000 claims abstract description 54
- 238000012549 training Methods 0.000 claims abstract description 39
- 230000001052 transient effect Effects 0.000 claims abstract description 29
- 239000002131 composite material Substances 0.000 claims abstract description 27
- 230000002159 abnormal effect Effects 0.000 claims abstract description 21
- 238000013507 mapping Methods 0.000 claims abstract description 8
- 230000005012 migration Effects 0.000 claims abstract description 4
- 238000013508 migration Methods 0.000 claims abstract description 4
- 239000013598 vector Substances 0.000 claims description 77
- 230000004913 activation Effects 0.000 claims description 22
- 230000008569 process Effects 0.000 claims description 16
- 238000011176 pooling Methods 0.000 claims description 13
- 230000009977 dual effect Effects 0.000 claims description 12
- 238000010606 normalization Methods 0.000 claims description 12
- 238000012360 testing method Methods 0.000 claims description 11
- 238000007476 Maximum Likelihood Methods 0.000 claims description 8
- 230000006870 function Effects 0.000 claims description 7
- 238000006243 chemical reaction Methods 0.000 claims description 6
- 230000005284 excitation Effects 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- 150000001875 compounds Chemical class 0.000 claims description 4
- 230000009286 beneficial effect Effects 0.000 abstract description 3
- 230000008878 coupling Effects 0.000 description 8
- 238000010168 coupling process Methods 0.000 description 8
- 238000005859 coupling reaction Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 4
- 238000010276 construction Methods 0.000 description 3
- 230000009471 action Effects 0.000 description 2
- 238000003745 diagnosis Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000003860 storage Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000010355 oscillation Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/096—Transfer learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02H—EMERGENCY PROTECTIVE CIRCUIT ARRANGEMENTS
- H02H1/00—Details of emergency protective circuit arrangements
- H02H1/0092—Details of emergency protective circuit arrangements concerning the data processing means, e.g. expert systems, neural networks
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J13/00—Circuit arrangements for providing remote indication of network conditions, e.g. an instantaneous record of the open or closed condition of each circuitbreaker in the network; Circuit arrangements for providing remote control of switching means in a power distribution network, e.g. switching in and out of current consumers by using a pulse code signal carried by the network
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
- H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Business, Economics & Management (AREA)
- Power Engineering (AREA)
- Economics (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Human Resources & Organizations (AREA)
- Molecular Biology (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Computing Systems (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention relates to the technical field of disturbance and fault identification of an in-phase power supply system, and particularly discloses a disturbance and fault identification method of the in-phase power supply system, which comprises the following steps: constructing a data set of transient disturbance of the in-phase power supply system and fault conditions of the in-phase power supply converter device; constructing a TimeNet-based feature extraction network and an attribute tag capable of describing various features in the dataset; training the constructed TimeNet-based feature extraction network by utilizing the constructed dataset, and learning the features of single disturbance or fault; constructing an attribute learner, and learning the mapping from the features to the attribute labels to realize the knowledge migration from disturbance or fault types to the composite working condition; and predicting the sample type based on the extracted attribute of the attribute learner, so as to realize the identification of the transient disturbance abnormal working condition of the in-phase power supply system and the fault abnormal working condition of the converter of the in-phase power supply device and the identification of the composite working condition. And the method provides guidance for the operation of the in-phase power supply converter, and is beneficial to the normal and safe operation of the in-phase power supply system.
Description
Technical Field
The invention relates to the technical field of disturbance and fault identification of an in-phase power supply system, in particular to a disturbance and fault identification method of the in-phase power supply system.
Background
The in-phase power supply system realizes in-phase power supply and negative sequence, reactive power and partial harmonic compensation by using the in-phase power supply technical device, thereby solving the problem of electric split phase and improving the working efficiency and the working stability of the traction power supply system. The core equipment of the in-phase power supply device is an in-phase power supply converter, and stable and reliable operation is a precondition of normal operation of an in-phase power supply system. The reliability of the power electronic device is low, and the power electronic device belongs to a weak link of an in-phase power supply system. On the other hand, in the train operation process, the abnormal electrical phenomenon of the original train network coupling is likely to cause the abnormal working condition of the in-phase power supply device, and according to the actual measurement data of the in-phase power supply device of a certain power supply section, when a large excitation surge current occurs at the network side, a large direct current component can cause the bias of a matching transformer, so that magnetic saturation occurs at the zero crossing point of the voltage of the matching transformer, and further, the abnormal triggering of the output current of the converter is caused to trigger the over-current protection.
Aiming at the abnormal working conditions caused by the vehicle network coupling disturbance of the in-phase power supply system and the power electronic device fault, no method is available at present for simultaneously considering the vehicle network coupling disturbance and the power electronic device fault, and the direct cause of the protection action of the in-phase power supply device cannot be analyzed.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a disturbance and fault identification method of an in-phase power supply system, which can simultaneously consider two main abnormal working conditions of vehicle network coupling disturbance and in-phase power supply converter power electronic device faults so as to realize the analysis of the reasons of the abnormal working conditions of the in-phase power supply device and solve the problems in the background art.
In order to achieve the above purpose, the present invention provides the following technical solutions: a disturbance and fault identification method of an in-phase power supply system comprises the following steps:
S1, constructing a transient disturbance and in-phase power supply device converter fault data set of a single-phase combined in-phase power supply system, and dividing the data set into a training set and a testing set;
s2, constructing a feature extraction network based on TimeNet, adding a Softmax layer at the tail end of the feature extraction network to form a classification network, training the feature extraction network by using a training set, learning the features of transient disturbance and a fault sample of the converter, retaining network parameters of the feature extraction network after training, and realizing feature extraction of the sample by using the feature extraction network after training;
S3, constructing attribute labels describing various characteristics of transient disturbance and converter faults based on current and voltage signals of different types of samples;
S4, constructing a fully-connected network as an attribute learner, and mapping learning features to attribute labels to realize knowledge migration from transient disturbance or converter fault types to composite working conditions;
S5, outputting an attribute vector based on the trained attribute learner, and predicting sample types by adopting maximum likelihood estimation, so that identification of transient disturbance abnormal working conditions of the in-phase power supply system and fault abnormal working conditions of the in-phase power supply device converter and identification of composite working conditions are realized.
Preferably, in step S1, the method specifically includes the following steps: collecting an output current signal of an in-phase power supply converter and a network side 27.5kV voltage signal in a single-phase combined in-phase power supply system, dividing the collected current signal and the collected voltage signal to form independent data samples with the same size, wherein a single data sample contains the current signal and the voltage signal in the same time period; the data samples are tensors with the size of 2 multiplied by 6400, and comprise current waveform data output by the 1s internal converter module and network side 27.5kV voltage waveform data.
Preferably, the training set does not include a composite working condition of transient disturbance of the in-phase power supply system and simultaneous occurrence of converter faults of the in-phase power supply device, namely the composite working condition is of an unobserved type;
The transient disturbance type comprises harmonic resonance, excitation surge current and grounding short circuit; the converter fault type comprises an inverter IGBT dual alpha type open circuit fault, a dual beta type open circuit fault and a non-dual double-tube open circuit fault in the back-to-back converter module.
Preferably, in step S2, the feature extraction network based on TimeNet includes a voltage channel and a current channel, each channel includes a time sequence dimension-increasing module and a feature extraction module, and performs feature extraction on the voltage and current waveforms and merges the extracted voltage and current features, where an inference process of the feature extraction network is expressed asWherein/>Represents the/>Sample number,/>Represents the/>The individual samples correspond to feature vectors.
Preferably, the time sequence dimension-increasing module is used for one-dimensional time sequenceFFT conversion is carried out to obtain the period and amplitude information of different frequency components, a period P corresponding to the frequency component with the largest amplitude is selected, signals are sheared by taking P as a unit and spliced in a first dimension, and a two-dimensional tensor/>, is obtainedAnd finishing the conversion of the one-dimensional time sequence problem into two-dimensional image recognition.
Preferably, the feature extraction module is formed by stacking a two-dimensional convolution layer, a batch normalization layer, a nonlinear activation layer and a maximum pooling layer, and adopts Inception network structure to improve generalization capability of the feature extraction module, and comprises a first Inception module, a second Inception module and a third Inception module;
The stacking sequence is a two-dimensional convolution layer, a batch normalization layer (BN layer), a nonlinear activation layer, a maximum pooling layer, a two-dimensional convolution layer, a first Inception module, a second Inception module and a third Inception module, wherein the first Inception module, the second Inception module and the third Inception module have the same structure, and each Inception module comprises the two-dimensional convolution layer, the Inception convolution layer, the batch normalization layer, the nonlinear activation layer and the maximum pooling layer.
Preferably, in step S3, the attribute tags contain direct descriptions of different types of transient and steady-state features in the sample, and the sample attribute vectors are defined manuallyThe attribute vector contains descriptions of n attributes of a single sample;
attribute vector of composite working condition XY formed by combination of disturbance type X and fault type Y Expressed as: Wherein/> And/>Attribute vectors corresponding to the disturbance type X and the fault type Y respectively;
attribute vectors of different classes form an Attribute description matrix Where k represents the total number of categories of fault category, disturbance category and compound condition category.
Preferably, in step S4, the attribute learner is formed by stacking a full connection layer, a BN layer, and a nonlinear activation layer, where the stacking order is in turn: a full connection layer, a BN layer, a nonlinear activation layer, a full connection layer, a BN layer, a nonlinear activation layer and a full connection layer;
The input of the attribute learner is the output feature vector of the feature extraction network after training Output as attribute vector/>, of corresponding sampleThe mapping relation between the features and the attribute vectors is learned in a supervised mode, and the reasoning process of the attribute learner is expressed as/>Wherein/>Represents the/>Corresponding feature vector of each sample,/>An attribute vector representing the ith sample generated by the attribute learner.
Preferably, the loss function when the attribute learner is trained is:
；
wherein the method comprises the steps of Representing the number of input feature vectors,/>Representing the artificially defined/>The attribute vector of the individual samples is used,An attribute vector representing the ith sample generated by the attribute learner.
Preferably, the maximum likelihood estimation formula is:
；
Wherein M is the number of test set categories; extracting the attribute by the expression algorithm through a feature extraction network and an attribute learner to generate an inference process of an attribute vector; /(I) Representing samples,/>Representing attribute learner outputting attribute vectors, i.e./>；/>Representing the attribute vector of the i-th sample defined manually.
The beneficial effects of the invention are as follows:
1) According to the disturbance and fault identification method for the single-phase combined type in-phase power supply system, provided by the invention, the output current of the converter of the single-phase combined type in-phase power supply system and the 27.5kV voltage on the network side are sampled and analyzed, the faults of power electronic devices in the in-phase power supply converter and the abnormal electrical phenomena outside the vehicle network coupling system can be considered at the same time, and the disturbance and fault identification for the in-phase power supply system can be realized; the invention can solve the problem that the fault diagnosis method of the current converter can not identify and classify the external disturbance and the internal power electronic device fault of the in-phase power supply device, judge the position (traction network side and in-phase power supply converter side) of the in-phase power supply system where the fault/disturbance source is located, further check the occurrence reason of abnormal working conditions, provide guidance for the operation of the in-phase power supply converter and be beneficial to the normal and safe operation of the in-phase power supply system.
2) The composite working condition is various in category and high in data set construction difficulty, and the disturbance and fault identification method provided by the invention adopts a zero sample learning method, and under the condition that no composite working condition sample exists, the corresponding semantic information of the fault characteristics is summarized through learning the voltage and current characteristics when single fault or disturbance occurs. The model can realize the identification of the composite working condition by utilizing the attribute information of the learned fault characteristics, and realize the zero sample diagnosis of the composite working condition while avoiding the huge workload of manufacturing the composite working condition data set.
Drawings
FIG. 1 is a schematic diagram of a single-phase in-phase power supply system in an embodiment of the invention;
Fig. 2 is a schematic diagram of a converter module of an in-phase power supply device according to an embodiment of the present invention;
FIG. 3 is a diagram illustrating an algorithm for identifying disturbance and fault of a single-phase combined in-phase power supply system according to an embodiment of the present invention;
FIG. 4 is a flowchart of an algorithm construction method for disturbance and fault identification of a single-phase combined in-phase power supply system according to an embodiment of the present invention;
Fig. 5 is a schematic diagram of a Inception network structure in an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Example 1:
A disturbance and fault identification method of an in-phase power supply system comprises the following steps:
S1, constructing a transient disturbance and in-phase power supply device converter fault data set of a single-phase combined in-phase power supply system, and dividing the data set into a training set and a testing set;
in the step S1, the method specifically comprises the steps of collecting an output current signal of an in-phase power supply converter and a network side 27.5kV voltage signal in a single-phase combined in-phase power supply system, dividing the collected current signal and the collected voltage signal to form independent data samples with the same size, wherein a single data sample contains the current signal and the voltage signal in the same time period; the data samples are tensors with the size of 2 multiplied by 6400 and comprise 1s internal converter module output current waveform data and network side 27.5kV voltage waveform data;
The complexity and the bulkiness of the required workload are established by considering a composite working condition data set, wherein the training set does not comprise a composite working condition of transient disturbance of an in-phase power supply system and simultaneous occurrence of converter faults of the in-phase power supply device, namely the composite working condition is of an unobserved type; the recognition problem under the condition of the composite working condition sample missing is solved based on the zero sample learning thought at the algorithm level, and the problems of huge category number and difficult acquisition of the composite working condition data set are solved by adopting zero sample learning.
Transient disturbance types include harmonic resonance, excitation surge current and grounding short circuit; the converter fault type comprises an inverter IGBT dual alpha type open circuit fault, a dual beta type open circuit fault and a non-dual double-tube open circuit fault in the back-to-back converter module;
s2, constructing a feature extraction network based on TimeNet, adding a Softmax layer at the tail end of the feature extraction network to form a classification network, training the feature extraction network by using a training set, learning the features of transient disturbance and a fault sample of the converter, retaining network parameters of the feature extraction network after training, and realizing feature extraction of the sample by using the feature extraction network after training;
The TimeNet-based feature extraction network comprises a voltage channel and a current channel, each channel comprises a time sequence dimension-increasing module and a feature extraction module, the voltage and current waveforms are respectively subjected to feature extraction, and the extracted voltage and current features are combined, wherein the reasoning process of the feature extraction network is expressed as follows
Wherein/>Represents the/>Sample number,/>Represents the/>The individual samples correspond to feature vectors;
The time sequence dimension-increasing module is used for one-dimensional time sequence FFT conversion is carried out to obtain the period and amplitude information of different frequency components, a period P corresponding to the frequency component with the largest amplitude is selected, signals are sheared by taking P as a unit and spliced in a first dimension, and a two-dimensional tensor/>, is obtained. Therefore, the one-dimensional time sequence problem is converted into two-dimensional image recognition without loss;
The feature extraction module is formed by stacking a two-dimensional convolution layer, a batch normalization layer, a nonlinear activation layer and a maximum pooling layer, improves the generalization capability of the feature extraction module by adopting a Inception network structure, and comprises a first Inception module, a second Inception module and a third Inception module;
The stacking sequence is a two-dimensional convolution layer, a batch normalization layer (BN layer), a nonlinear activation layer, a maximum pooling layer, a two-dimensional convolution layer, a first Inception module, a second Inception module and a third Inception module, wherein the first Inception module, the second Inception module and the third Inception module have the same structure, and each Inception module comprises the two-dimensional convolution layer, the Inception convolution layer, the batch normalization layer, the nonlinear activation layer and the maximum pooling layer.
S3, constructing attribute labels describing various characteristics of transient disturbance and converter faults based on current and voltage signals of different types of samples; the attribute tag comprises direct description of different types of transient and steady state characteristics in the sample, and sample attribute vector is defined manuallyThe attribute vector contains descriptions of n attributes of a single sample. Attribute vector/>Intermediate element/>Representing whether the voltage-current waveform signal contains a representation of a characteristic,/>The value is 0 or 1.
The attribute label of the composite working condition where the fault and the disturbance occur simultaneously can be constructed by the mode that the fault attribute label and the disturbance attribute label are combined.
Attribute vector of composite working condition XY formed by combination of disturbance type X and fault type YExpressed as: Wherein/> And/>Attribute vectors corresponding to the disturbance type X and the fault type Y respectively;
attribute vectors of different classes form an Attribute description matrix Where k represents the total number of categories of fault category, disturbance category and compound condition category.
S4, constructing a fully-connected network as an attribute learner, and mapping learning features to attribute labels to realize knowledge migration from transient disturbance or converter fault types to composite working conditions;
the attribute learner is formed by stacking a full connection layer, a BN layer and a nonlinear activation layer, wherein the stacking sequence is as follows: a full connection layer, a BN layer, a nonlinear activation layer, a full connection layer, a BN layer, a nonlinear activation layer and a full connection layer;
The input of the attribute learner is the output feature vector of the feature extraction network after training Output as attribute vector/>, of corresponding sampleThe mapping relation between the features and the attribute vectors is learned in a supervised mode, and the reasoning process of the attribute learner is expressed as/>Wherein/>Represents the/>Corresponding feature vector of each sample,/>An attribute vector representing the ith sample generated by the attribute learner. And (3) realizing the reasoning process of the attribute learner by constructing a fully connected network.
The loss function of the attribute learner during training is as follows:
；
wherein the method comprises the steps of Representing the number of input feature vectors,/>Attribute vector representing artificially defined ith sample,/>An attribute vector representing the ith sample generated by the attribute learner.
S5, outputting an attribute vector by an attribute learner based on training completionAnd predicting the sample category by adopting maximum likelihood estimation to obtain a sample prediction category label, thereby realizing the identification of transient disturbance abnormal working conditions of the in-phase power supply system and fault abnormal working conditions of the in-phase power supply device converter and the identification of composite working conditions.
The maximum likelihood estimation formula is:
；
Wherein M is the number of test set categories; extracting the attribute by the expression algorithm through a feature extraction network and an attribute learner to generate an inference process of an attribute vector; /(I) Representing samples,/>Representing attribute learner outputting attribute vectors, i.e./>；/>Representing the attribute vector of the i-th sample defined manually.
Example 2:
Firstly, the invention is introduced to the applicable scene, and the invention can be applied to the identification of the vehicle network coupling disturbance and the power electronic device fault of the in-phase power supply converter in the single-phase combined in-phase power supply system.
As shown in figure 1, the single-phase combined type in-phase power supply system is an AC-DC-AC converter, and is connected with a 27.5kV traction network side through a matching transformer, and in the train operation process, the in-phase power supply system can experience transient and quasi-steady states such as harmonic resonance, low-frequency oscillation, excitation surge current and the like to influence the normal operation of the in-phase power supply converter. The in-phase power supply converter is formed by cascading or connecting a plurality of back-to-back H-bridge converter modules in parallel, and the structure of a single converter module is shown in figure 2. The inverter side of the converter is connected with a traction network through a matching transformer, the port voltage of the matching transformer is U M, and the output current of the port of the inverter is Ialpha; the rectifier side is connected with the traction transformer through a matching transformer, the port voltage of the matching transformer is U T, and the output current of the rectifier port is I beta. The open circuit fault of the IGBT in the converter can have a great influence on the output of the in-phase power supply device, and even the converter can trigger overcurrent protection to stop operation.
Aiming at the abnormal working conditions caused by the vehicle network coupling disturbance of the in-phase power supply system and the power electronic device fault, no method is available at present for simultaneously considering the vehicle network coupling disturbance and the power electronic device fault, and the direct cause of the protection action of the in-phase power supply device cannot be analyzed.
In conclusion, the disturbance and the fault in the in-phase power supply system are identified, the abnormal working condition of the in-phase power supply converter can be analyzed from the level of the in-phase power supply system, and further the occurrence reason of the abnormal working condition is checked, so that the normal and safe operation of the in-phase power supply system is facilitated.
Based on this, the embodiment of the invention provides a disturbance and fault identification method of an in-phase power supply system, a framework is shown in fig. 3, a construction flow is shown in fig. 4, and the method comprises the following steps:
s1, constructing a data set, and considering that disturbance types comprise harmonic resonance, excitation surge current and grounding short circuit; the converter fault type comprises an inverter IGBT dual alpha type open circuit fault, a dual beta type open circuit fault and a non-dual double-tube open circuit fault in the back-to-back converter module.
Building training data sets based thereonAnd a test dataset. Wherein/>Represents the/>The number of (data) samples and their corresponding tags; /(I)Representing the samples of disturbances or faults of the type seen in the training set and in the test set, respectively, and there are/>；Representing the unoccupied type in the test set, namely the compound working condition; /(I)Representing the visible type tags in the training set,/>Representing the visible type tags and the invisible type tags in the test set, respectively. The sampling frequency of the voltage and current signals is 6400Hz, and the samples in the data set/>The tensors are all tensors with the size of 2 multiplied by 6400, and comprise waveform data of the output current Ialpha of the 1s internal converter module and waveform data of the network side 27.5kV voltage U M.
S2, constructing a TimeNet-based feature extraction network, wherein the input of the network is different types of faults or disturbance samples, and the output is a feature vector. The characteristic extraction network comprises a voltage channel and a current channel, each channel comprises a time sequence dimension increasing module and a characteristic extraction module, and the structures of the two channels are identical. Respectively extracting characteristics of voltage and current waveforms and extracting the extracted current characteristicsVoltage characteristics/>Merging to obtain final features/>. The reasoning process of the feature extraction network is expressed asWherein/>Represents the/>Sample number,/>Represents the/>The individual samples correspond to feature vectors.
Taking a voltage channel as an example, a time sequence dimension-increasing module in the channel performs one-dimensional time sequence on voltage waveformsFFT conversion is carried out to obtain the period and amplitude information of different frequency components, a period P corresponding to the frequency component with the largest amplitude is selected, the signal is sheared by taking P as a unit and is spliced in the first dimension, and the two-dimensional tensor/> isobtained. Therefore, the one-dimensional time sequence problem is converted into the two-dimensional image recognition field without loss.
The feature extraction module in the channel is formed by stacking a two-dimensional convolution layer, a batch normalization layer, a nonlinear activation layer and a maximum pooling layer, wherein the stacking sequence is the two-dimensional convolution layer, a BN layer, the nonlinear activation layer and the maximum pooling layer, the two-dimensional convolution layer, a first Inception module, a second Inception module and a third Inception module, and the first Inception module, the second Inception module and the third Inception module have the same structure as each other, as shown in FIG. 5.
Each Inception module contains a two-dimensional convolution layer, a Inception convolution layer, a BN layer, a nonlinear activation layer, and a max-pooling layer. The stack of Inception convolution layers and pooling layers fully extracts the high-dimensional characteristics of the data on the basis of guaranteeing the generalization capability of the model, reduces the calculated amount and improves the data processing speed.
Setting super parameters, loss functions and optimization strategies of the feature extraction network, and training the feature extraction network by using a training set.
When the feature extraction network is trained, a softmax classifier is added at the end of the network, and the feature extraction network is subjected to supervised training in the form of classification subtasks. In the training process, an Adam algorithm is selected as a model optimization strategy, and a cross entropy loss function is selected as a model loss function.
S3, constructing an attribute description capable of describing fault or disturbance characteristics based on current-voltage signal characteristics of samples of different categories, wherein the attribute description is shown in the following table:
TABLE 1 disturbance or failure attribute information
The attribute tag contains direct description of different types of transient and steady state features in the sample, and sample attribute vector is defined manuallyThe attribute vector contains descriptions of 9 attributes of a single sample. Attribute vectors of different classes make up an attribute description matrix/>. The attribute description matrix is shown in table 2.
Table 2 attribute description matrix
Performing attribute description on each sample in the training set to form an attribute vector, and combining the feature vectors of the samples output by the feature extractor to form a second training set. For training the attribute learner. /(I)Representing the corresponding eigenvector of the ith sample,/>Attribute vector representing artificially defined ith sample,/>Representation utilizing training set sample set/>A feature vector set generated by the middle sample through a feature extractor; /(I)Representing a set of attribute vectors corresponding to different types of data.
S4, constructing a fully-connected network as an attribute learner, wherein the attribute learner is formed by stacking a fully-connected layer, a BN layer and a nonlinear activation layer, and the stacking sequence is as follows: full tie layer, BN layer, nonlinear activation layer, full tie layer. The input of the attribute learner is the output feature vector of the feature extraction network after trainingOutput as attribute vector/>, of corresponding sampleThe mapping relationship between the feature and the attribute vector is learned in a supervised manner. The reasoning process of the attribute learner is denoted/>Wherein/>Represents the/>Samples. In order to make the attribute vector generated by the attribute learner to be as close to the artificial attribute vector as possible, the loss function of the attribute learner during training is set as follows:
；
wherein the method comprises the steps of Representing the number of input feature vectors,/>Representing the artificially defined/>The attribute vector of the individual samples is used,Representing the second/>, generated by the attribute learnerAttribute vector of individual samples.
S5, outputting an attribute vector by an attribute learner based on training completionAnd predicting the sample category by adopting maximum likelihood estimation to obtain a sample prediction category label, thereby realizing the identification of transient disturbance abnormal working conditions of the in-phase power supply system and fault abnormal working conditions of the in-phase power supply device converter and the identification of composite working conditions. The maximum likelihood estimation formula is:
；
Wherein M is the number of test set categories; extracting the attribute by the expression algorithm through a feature extraction network and an attribute learner to generate an inference process of an attribute vector; /(I) Representing samples,/>Representing attribute learner outputting attribute vectors, i.e./>；/>Representing the attribute vector of the i-th sample defined manually.
The embodiment of the specification also discloses a computer readable storage medium, wherein the storage medium stores computer instructions, and when the computer reads the computer instructions, the computer can execute the steps of the in-phase power supply system disturbance and fault identification method.
Although the present invention has been described with reference to the foregoing embodiments, it will be apparent to those skilled in the art that modifications may be made to the embodiments described, or equivalents may be substituted for elements thereof, and any modifications, equivalents, improvements and changes may be made without departing from the spirit and principles of the present invention.
Claims (8)
1. The disturbance and fault identification method of the in-phase power supply system is characterized by comprising the following steps of:
S1, constructing a transient disturbance and in-phase power supply device converter fault data set of a single-phase combined in-phase power supply system, and dividing the data set into a training set and a testing set;
s2, constructing a feature extraction network based on TimeNet, adding a Softmax layer at the tail end of the feature extraction network to form a classification network, training the feature extraction network by using a training set, learning the features of transient disturbance and a fault sample of the converter, retaining network parameters of the feature extraction network after training, and realizing feature extraction of the sample by using the feature extraction network after training;
S3, constructing attribute labels describing various characteristics of transient disturbance and converter faults based on current and voltage signals of different types of samples; the attribute tag comprises direct description of different types of transient and steady-state characteristics in a sample, a sample attribute vector alpha= { a 1,a2,...,an } is manually defined, and the attribute vector comprises description of n attributes of a single sample;
The attribute vector α XY of the composite condition XY formed by the combination of the disturbance type X and the fault type Y is expressed as: α XY＝αX∪αY, wherein α X and α Y are attribute vectors corresponding to the disturbance type X and the fault type Y respectively;
The attribute vectors of different classes form an attribute description matrix A= { alpha 1,α2,...,αk }, wherein k represents the total class number of fault classes, disturbance classes and compound working condition classes;
s4, constructing a fully-connected network as an attribute learner, and mapping learning features to attribute labels to realize knowledge migration from transient disturbance or converter fault types to composite working conditions; the attribute learner is formed by stacking a full connection layer, a batch normalization layer and a nonlinear activation layer, wherein the stacking sequence is as follows: full connection layer, batch normalization layer, nonlinear activation layer, full connection layer;
The input of the attribute learner is the output feature vector f of the feature extraction network after training, the output is the attribute vector alpha '= { a 1',a2',...,an' } of the corresponding sample, the mapping relation between the feature and the attribute vector is learned in a supervised mode, the reasoning process of the attribute learner is expressed as G: f i→αi ', wherein f i represents the corresponding feature vector of the ith sample, and alpha i' represents the attribute vector of the ith sample generated by the attribute learner;
S5, outputting an attribute vector based on the trained attribute learner, and predicting sample types by adopting maximum likelihood estimation, so that identification of transient disturbance abnormal working conditions of the in-phase power supply system and fault abnormal working conditions of the in-phase power supply device converter and identification of composite working conditions are realized.
2. The method for identifying disturbance and failure of in-phase power supply system according to claim 1, wherein: in step S1, the method specifically includes the following steps: collecting an output current signal of an in-phase power supply converter and a network side 27.5kV voltage signal in a single-phase combined in-phase power supply system, dividing the collected current signal and the collected voltage signal to form independent data samples with the same size, wherein a single data sample contains the current signal and the voltage signal in the same time period; the data samples are tensors with the size of 2 multiplied by 6400, and comprise current waveform data output by the 1s internal converter module and network side 27.5kV voltage waveform data.
3. The method for identifying disturbance and failure of in-phase power supply system according to claim 1, wherein: the training set does not comprise a composite working condition of transient disturbance of the in-phase power supply system and simultaneous occurrence of converter faults of the in-phase power supply device, namely the composite working condition type is an unobserved type;
The transient disturbance type comprises harmonic resonance, excitation surge current and grounding short circuit; the converter fault type comprises an inverter IGBT dual alpha type open circuit fault, a dual beta type open circuit fault and a non-dual double-tube open circuit fault in the back-to-back converter module.
4. The method for identifying disturbance and failure of in-phase power supply system according to claim 1, wherein: in step S2, the feature extraction network based on TimeNet includes a voltage channel and a current channel, each channel includes a time sequence dimension raising module and a feature extraction module, and features are extracted from the voltage waveform and the current waveform respectively, and the extracted voltage and current features are combined, wherein an reasoning process of the feature extraction network is expressed as F: x i→fi, where x i represents an ith sample, and F i represents a feature vector corresponding to the ith sample.
5. The method for identifying disturbance and failure of in-phase power supply system according to claim 4, wherein: the time sequence dimension-increasing module performs FFT conversion on the one-dimensional time sequence X 1∈R1×6400 to obtain the period and amplitude information of different frequency components, selects a period P corresponding to the frequency component with the largest amplitude, cuts signals by taking P as a unit and splices the signals in a first dimension to obtain a two-dimensional tensor X 2∈RP×(6400/P), and completes the conversion of the one-dimensional time sequence problem to two-dimensional image recognition.
6. The method for identifying disturbance and failure of in-phase power supply system according to claim 4, wherein: the feature extraction module is formed by stacking a two-dimensional convolution layer, a batch normalization layer, a nonlinear activation layer and a maximum pooling layer, improves the generalization capability of the feature extraction module by adopting a Inception network structure, and comprises a first Inception module, a second Inception module and a third Inception module;
The stacking sequence is two-dimensional convolution layers, batch normalization layers, nonlinear activation layers, maximum pooling layers, two-dimensional convolution layers, a first Inception module, a second Inception module and a third Inception module, wherein the first Inception module, the second Inception module and the third Inception module have the same structure, and each Inception module comprises two-dimensional convolution layers, inception convolution layers, batch normalization layers, nonlinear activation layers and maximum pooling layers.
7. The method for identifying disturbance and failure of in-phase power supply system according to claim 1, wherein: the loss function of the attribute learner during training is as follows:
Where N S represents the number of input feature vectors, a i represents the attribute vector of the i-th sample that is manually defined, and a i' represents the attribute vector of the i-th sample that is generated by the attribute learner.
8. The method for identifying disturbance and failure of in-phase power supply system according to claim 1, wherein: the maximum likelihood estimation formula is:
Wherein M is the number of test set categories; g (F ()) represents an inference process of extracting the attribute by the algorithm through the feature extraction network and the attribute learner to generate an attribute vector; x represents a sample, G (F (x)) represents an attribute learner output attribute vector, i.e., α i';αi represents an attribute vector of an artificially defined i-th sample.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410430549.8A CN118040679B (en) | 2024-04-11 | 2024-04-11 | Disturbance and fault identification method for in-phase power supply system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410430549.8A CN118040679B (en) | 2024-04-11 | 2024-04-11 | Disturbance and fault identification method for in-phase power supply system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN118040679A CN118040679A (en) | 2024-05-14 |
| CN118040679B true CN118040679B (en) | 2024-06-18 |
Family
ID=90989721
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410430549.8A Active CN118040679B (en) | 2024-04-11 | 2024-04-11 | Disturbance and fault identification method for in-phase power supply system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118040679B (en) |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2013538543A (en) * | 2010-07-16 | 2013-10-10 | ザ トラスティーズ オブ コロンビア ユニバーシティ イン ザ シティー オブ ニューヨーク | Machine learning for power grids |
| CN113933757B (en) * | 2020-06-29 | 2022-09-16 | 株洲中车时代电气股份有限公司 | Traction transformer overcurrent diagnosis protection device and method |
| CN113283550A (en) * | 2021-07-22 | 2021-08-20 | 西南交通大学 | Abnormal identification model training method for vehicle network electric coupling data |
| CN114414942B (en) * | 2022-01-14 | 2024-08-16 | 重庆大学 | Power transmission line fault identification classifier, identification method and system based on transient waveform image identification |
| CN116106672B (en) * | 2023-04-13 | 2023-08-18 | 西南交通大学 | Vehicle network resonance detection method and device based on data driving and engineering knowledge |
-
2024
- 2024-04-11 CN CN202410430549.8A patent/CN118040679B/en active Active
Non-Patent Citations (2)
| Title |
|---|
| "Identification of Abnormal Electrical Phenomena in Train-grid System in Open Set Environment";Fulin Zhou 等;《 IEEE Transactions on Transportation Electrification ( Early Access )》;20240220;1-12 * |
| "基于MEEMD与概率神经网络的贯通式同相牵引直接供电系统故障识别";蔡潇 等;《电力科学与工程 》;20200131;第36卷(第1期);63-70 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN118040679A (en) | 2024-05-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109635928B (en) | Voltage sag reason identification method based on deep learning model fusion | |
| Haque et al. | Deep learning based micro-grid fault detection and classification in future smart vehicle | |
| CN111914320A (en) | No-sample turnout fault diagnosis method based on deep learning | |
| CN109782091A (en) | Multi-electrical level inverter method for diagnosing faults based on depth small echo extreme learning machine | |
| CN113093058A (en) | NPC three-level inverter open-circuit fault diagnosis method | |
| CN110068776A (en) | Three-level inverter open-circuit fault diagnostic method based on Support Vector Machines Optimized | |
| Jacob et al. | Fault diagnostics in shipboard power systems using graph neural networks | |
| CN111999591A (en) | Method for identifying abnormal state of primary equipment of power distribution network | |
| CN113030789A (en) | Series arc fault diagnosis and line selection method based on convolutional neural network | |
| Tikariha et al. | Fault classification in an ieee 30 bus system using convolutional neural network | |
| CN116484299A (en) | Charging pile fault diagnosis method based on integration of gradient lifting tree and multi-layer perceptron | |
| CN107037278A (en) | A kind of substandard intelligent substation method for diagnosing faults of IEC61850 | |
| CN118040679B (en) | Disturbance and fault identification method for in-phase power supply system | |
| Xu et al. | An improved ELM-WOA–based fault diagnosis for electric power | |
| Moreira et al. | Evaluation of pattern recognition algorithms for applications on power factor compensation | |
| Boorgula et al. | Incipient Fault Diagnosis in Stator Winding of Synchronous Generator: A CMFFLC Technique | |
| Mestha et al. | Support vector machine based fault detection in inverter‐fed electric vehicle | |
| Zeng et al. | Fault Detection of Flexible DC Distribution Network Based on GAF and Improved Deep Residual Network | |
| Thapa et al. | A machine learning-based approach to detection of fault locations in transmission networks | |
| Liu et al. | Circuit Fault Diagnosis Method of Wind Power Converter with Wavelet-DBN | |
| Li et al. | Fault diagnosis of three-level inverters based on ensemble empirical mode decomposition and deep neural network | |
| Wang et al. | Turn-to-turn short circuit of motor stator fault diagnosis using dropout rate improved deep sparse autoencoder | |
| Wei et al. | Fault detection method for flexible DC grid based on CEEMDAN multiscale entropy and GA-SVM | |
| CN118535953A (en) | Transient disturbance and fault identification method, system and equipment for vehicle network electrical system and storage medium | |
| Zhang et al. | A data-driven approach for electromagnetic transient recognition using transfer learning |
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 |