CN117572161A - Arc fault detection method and device and related components - Google Patents
Arc fault detection method and device and related components Download PDFInfo
- Publication number
- CN117572161A CN117572161A CN202210931972.7A CN202210931972A CN117572161A CN 117572161 A CN117572161 A CN 117572161A CN 202210931972 A CN202210931972 A CN 202210931972A CN 117572161 A CN117572161 A CN 117572161A
- Authority
- CN
- China
- Prior art keywords
- data
- arc fault
- effnet
- arc
- training
- 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
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 54
- 238000012549 training Methods 0.000 claims abstract description 103
- 238000000034 method Methods 0.000 claims abstract description 26
- 238000001228 spectrum Methods 0.000 claims description 25
- 238000011176 pooling Methods 0.000 claims description 18
- 230000008569 process Effects 0.000 claims description 17
- 238000012545 processing Methods 0.000 claims description 16
- 238000012795 verification Methods 0.000 claims description 15
- 238000010606 normalization Methods 0.000 claims description 11
- 238000007781 pre-processing Methods 0.000 claims description 9
- 238000012360 testing method Methods 0.000 claims description 9
- 230000002159 abnormal effect Effects 0.000 claims description 7
- 238000004590 computer program Methods 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 2
- 238000004364 calculation method Methods 0.000 description 17
- 238000013528 artificial neural network Methods 0.000 description 11
- 238000010586 diagram Methods 0.000 description 7
- 230000009471 action Effects 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 3
- 238000006243 chemical reaction Methods 0.000 description 3
- 230000000630 rising effect Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000004140 cleaning Methods 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000010355 oscillation Effects 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 206010000369 Accident Diseases 0.000 description 1
- 230000032683 aging Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000002035 prolonged effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/12—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing
- G01R31/1227—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing of components, parts or materials
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Testing Of Short-Circuits, Discontinuities, Leakage, Or Incorrect Line Connections (AREA)
Abstract
The application discloses an arc fault detection method, an arc fault detection device and related components, and relates to the field of arc detection, wherein the arc fault detection method comprises the following steps: acquiring current data of electrical equipment in a current detection period; extracting characteristic data in the current data; inputting the characteristic data into an arc fault identification model to obtain an arc fault identification result of the current detection period; the arc fault recognition model is a recognition model obtained by training the EffNet network based on arc fault current data under various working conditions. The method and the device can effectively identify arc faults of the electrical equipment under various working conditions, have fewer parameter amounts, do not need a large amount of storage space and computing resources, and are convenient to apply to the embedded platform.
Description
Technical Field
The present disclosure relates to the field of arc detection, and in particular, to an arc fault detection method, apparatus, and related components.
Background
With the continuous rise of electricity consumption, electrical fire accidents caused by electric equipment faults, electrical circuit aging and the like are also increasing year by year, wherein more than 60% of electrical fires are caused by arc faults of a medium-low voltage system. The traditional arc fault identification method is based on the time-frequency domain characteristics of the manually extracted arc current, and the arc fault is identified by setting a characteristic quantity threshold value according to human experience, but the arc fault current is influenced by factors such as load type, current grade and the like, so that the method has strong randomness, the time-frequency domain characteristics of the arc fault are different, and the fixed characteristic quantity threshold value is difficult to effectively identify the arc faults under various working conditions.
Therefore, how to provide a solution to the above technical problem is a problem that a person skilled in the art needs to solve at present.
Disclosure of Invention
The purpose of the application is to provide an arc fault detection method, an arc fault detection device and related components, which can effectively identify arc faults of electrical equipment under various working conditions, have fewer parameter amounts, do not need a large amount of storage space and calculation resources, and are convenient to apply to an embedded platform.
In order to solve the above technical problems, the present application provides an arc fault detection method, including:
acquiring current data of electrical equipment in a current detection period;
extracting characteristic data in the current data;
inputting the characteristic data into an arc fault identification model to obtain an arc fault identification result of the current detection period;
the arc fault recognition model is a recognition model obtained by training an EffNet network based on arc fault current data under various working conditions, the EffNet network comprises n EffNet modules, each EffNet module comprises an input layer, an intermediate layer and an output layer, the input layer is a one-dimensional convolution layer with the width of 1, the intermediate layer comprises a space separable convolution layer with the width of 1 and a maximum pooling layer, the output layer is a one-dimensional convolution layer with the width of 1, and n is a positive integer.
Optionally, the process of training the affnet network based on the arc fault current data under various working conditions to obtain the arc fault recognition model includes:
arc fault current data under various working conditions are obtained;
preprocessing all the arc fault current data to obtain a data set meeting training requirements;
dividing the data set into a training set, a verification set and a test set;
training the EffNet network through the training set to obtain an initial recognition model;
inputting the verification set into the initial recognition model, and judging whether the initial recognition model meets arc fault recognition requirements or not based on an arc fault recognition result output by the initial recognition model;
if not, adjusting network super parameters of the EffNet network, and repeating the step of training the EffNet network through the training set to obtain an initial recognition model until an arc fault recognition model meeting the arc fault recognition requirement is obtained.
Optionally, the process of preprocessing all the arc fault current data to obtain a data set meeting training requirements includes:
cutting all the arc fault current data according to a preset time window to obtain a plurality of first data samples with preset lengths;
performing FFT processing on each first data sample to obtain energy spectrum data, and selecting the energy spectrum data of a target frequency band as a second data sample;
carrying out normalization processing on each second data sample to obtain a third data sample;
removing abnormal data in each third data sample to obtain a fourth data sample;
adding a label to each fourth data sample to obtain a fifth data sample;
a data set is obtained based on all of the fifth data samples.
Optionally, the target frequency band is 20kHz-100 kHz.
Optionally, the process of extracting the characteristic data in the current data includes:
performing FFT (fast Fourier transform) on the current data so as to convert the current data in the time domain into energy spectrum data in the frequency domain;
and carrying out normalization processing on the energy spectrum data to obtain characteristic data.
Optionally, the input layer is a 1×1 one-dimensional convolution layer, the intermediate layer includes a 3×1 spatially separable convolution layer and a step size of 2 max-pooling layer, and the output layer is a 2×1 one-dimensional convolution layer.
Optionally, the EffNet network is formed by stacking 3 EffNet modules.
Optionally, the batch size of the affnet network is 64.
For solving the technical problem, the application also provides an arc fault detection device, which comprises:
the acquisition module is used for acquiring current data of the electrical equipment in the current detection period;
the extraction module is used for extracting characteristic data in the current data;
the detection module is used for inputting the characteristic data into an arc fault recognition model to obtain an arc fault recognition result of the current detection period;
the arc fault recognition model is a recognition model obtained by training an EffNet network based on arc fault current data under various working conditions, the EffNet network comprises n EffNet modules, each EffNet module comprises an input layer, an intermediate layer and an output layer, the input layer is a one-dimensional convolution layer with the width of 1, the intermediate layer comprises a space separable convolution layer with the width of 1 and a maximum pooling layer, the output layer is a one-dimensional convolution layer with the width of 1, and n is a positive integer.
To solve the above technical problem, the present application further provides a computer readable storage medium, on which a computer program is stored, which when executed by a processor, implements the steps of the arc fault detection method as described in any one of the above.
The invention provides an arc fault detection method, which comprises the steps of firstly improving an EffNet network based on current data characteristics so that the EffNet network can be suitable for arc data, then detecting whether arc faults exist in the current data of electric equipment through an arc fault identification model obtained by training the EffNet network based on the arc fault current data under various working conditions, effectively identifying the arc faults of the electric equipment under various working conditions, and meanwhile, the EffNet network is a lightweight neural network, has fewer parameter amounts, does not need a large amount of storage space and calculation resources and is convenient to apply to an embedded platform. The application also provides an arc fault detection device and a computer readable storage medium, which have the same beneficial effects as the arc fault detection method.
Drawings
For a clearer description of the embodiments of the present application, the drawings that are needed in the embodiments will be briefly described, it being apparent that the drawings in the following description are only some embodiments of the present application, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of steps of a method for arc fault detection provided herein;
fig. 2 is a schematic structural diagram of a conventional affnet module;
fig. 3 is a schematic structural diagram of an arc_affnet module provided in the present application;
fig. 4 is a schematic diagram of an affnet network including different numbers of arc_affnet modules according to the present application;
FIG. 5 is a graph of model training set accuracy including different numbers of arc_EffNet modules;
FIG. 6 is a schematic diagram of model accuracy including different numbers of arc_EffNet modules;
FIG. 7 is a schematic diagram of model parameters and training time including different numbers of arc_EffNet modules;
FIG. 8 is a graph of training set accuracy at different Baschsize as provided herein;
FIG. 9 is a graph of accuracy and training time for different Baschsize provided herein;
FIG. 10 is a graph showing loss and accuracy in model training
FIG. 11 is a graph of learning rate change during model training provided herein;
fig. 12 is a schematic structural diagram of an arc fault detection device provided in the present application.
Detailed Description
The core of the application is to provide an arc fault detection method, an arc fault detection device and related components, which can effectively identify arc faults of electrical equipment under various working conditions, have fewer parameter amounts, do not need a large amount of storage space and calculation resources, and are convenient to apply to an embedded platform.
For the purposes of making the objects, technical solutions and advantages of the embodiments of the present application more clear, the technical solutions of the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is apparent that the described embodiments are some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are within the scope of the present disclosure.
Referring to fig. 1, fig. 1 is a flowchart illustrating steps of an arc fault detection method provided in the present application, where the arc fault detection method includes:
s101: acquiring current data of electrical equipment in a current detection period;
specifically, the collecting device can collect current data of the electrical equipment according to a preset collecting period, the step is to obtain a plurality of current data collected by the collecting device according to a preset detecting period, and it is understood that the detecting period is larger than the collecting period, and the detecting period can be specifically determined according to the length of input data required by the EffNet network in the application.
The electrical device may specifically be a photovoltaic dc inverter.
S102: extracting characteristic data in the current data;
specifically, considering that the acquired plurality of current data of the electrical equipment are data on a time domain, performing time-frequency conversion on the plurality of current data through FFT (fast Fourier transform) to convert the current data on the time domain into energy spectrum data on a frequency domain, and considering that the actually sampled current data has the problem of overlarge data difference, failure characteristics can not be accurately extracted, performing normalization processing on the energy spectrum data after time-frequency conversion, so that the range of all the energy spectrum data is between [0,1], and taking the energy spectrum data after normalization processing as characteristic data.
S103: inputting the characteristic data into an arc fault identification model to obtain an arc fault identification result of the current detection period;
the arc fault recognition model is a recognition model obtained by training an EffNet network based on arc fault current data under various working conditions, the EffNet network comprises n EffNet modules, each EffNet module comprises an input layer, an intermediate layer and an output layer, the input layer is a one-dimensional convolution layer with the width of 1, the intermediate layer comprises a space separable convolution layer with the width of 1 and a maximum pooling layer, the output layer is a one-dimensional convolution layer with the width of 1, and n is a positive integer.
Before executing the step, a preset recognition model constructed in advance is firstly explained, in order to facilitate accurate recognition of arc faults under various working conditions, an arc generating platform is firstly built according to UL1699B standard so as to generate arc faults under various working conditions and collect arc fault current data. And then training the preset neural network based on arc fault current data under various working conditions as a data set to obtain a preset identification model.
Specifically, in order to reduce the number of network parameters and the calculated amount while improving the accuracy and the reliability of the arc fault identification method, the preset neural network can be a lightweight neural network EffNet network. Specifically, the affnet network is formed by stacking a plurality of affnet modules, and referring to fig. 2, the conventional affnet module includes a convolution kernel with a size of 1×3, which means that the input data should have a length of at least 1 and a width of at least 3, so that the affnet network model is suitable for two-dimensional data with a width greater than 3. The current data is one-dimensional time series data, and cannot be directly applied to arc fault identification by the EffNet. Therefore, the application improves the EffNet module, and provides a lightweight fault Arc identification model arc_EffNet capable of training by adopting one-dimensional current data, the structure of the arc_EffNet module is shown in fig. 3, compared with the traditional EffNet module, the arc_EffNet module omits a 1×3 convolution kernel and a subsequent pooling layer, the maximum pooling calculation is added after the 3×1 convolution calculation, and the arc_EffNet module keeps the depth separable convolution function of the traditional EffNet module and is applicable to the one-dimensional current data.
Specifically, the arc_EffNet module comprises an input layer, an intermediate layer and an output layer, wherein the input layer is a one-dimensional convolution layer with the width of 1, the intermediate layer comprises a space separable convolution layer with the width of 1 and a maximum pooling layer, and the output layer is a one-dimensional convolution layer with the width of 1. As a preferred embodiment, referring to fig. 3, the input layer may be a 1×1 one-dimensional convolution layer, and the intermediate layer includes a 3×1 spatially separable convolution layer and a maximum pooling layer with a step size of 2, and the last layer is a 2×1 one-dimensional convolution layer with a step size of 2. In order to eliminate the bottleneck of the data stream, the number of convolution kernel channels of the last layer needs to be twice that of the other two layers, and specific structural parameters of the arc_EffNet module are shown in table 1.
Table 1arc_EffNet module parameter table
| Network layer | Output feature map shape | Data volume |
| One-dimensional convolution 1×1×32 | 800×32 | 2560 |
| Depth separable convolution 3 x 1 | 800×32 | 2560 |
| Maximum pooling | 400×32 | 1280 |
| One-dimensional convolution 2×1×64 | 200×64 | 1280 |
It can be understood that, in order to increase the network depth and enhance the learning capability, the affnet network corresponding to the Arc fault recognition model is formed by stacking a plurality of arc_affnet modules. Considering that the shallower the depth of the arc fault recognition model is, the smaller the parameter is, which is more beneficial to the application of the arc fault recognition model on embedded equipment, but the shallow model has limited representation capability and low recognition accuracy. As the depth of the network model increases, the representation ability and recognition accuracy of the model are higher, but the number of parameters and calculation amount are also increased. Therefore, the proper number of arc_EffNet modules is selected, so that the parameter quantity and the calculated quantity of the Arc fault recognition model can be minimized while the recognition accuracy is ensured.
Specifically, assume that the length of each sample data in the data set is 800. Referring to table 1, the length of the output data after passing through one arc_affnet module is 200, and is reduced to 1/4 of the length of the input data. The length of the output data is 800/4n after passing through n arc_affnet modules. If arc_affnet modules in the Arc fault recognition model are too many, the length of input data of the arc_affnet modules of the last layers is too short, and therefore difficulty is brought to training of the neural network.
Based on the above, three types of Arc_EffNet modules, the number of which is 2, 3 and 4, are selected for training, and are respectively marked as Arc_EffNet_2blocks, arc_EffNet_3blocks and Arc_EffNet_4blocks. The structure of the three models is shown in fig. 4. In fig. 4, the convolution computation in the arc_effnet module all uses the "same" mode. After convolution calculation, the data length is the input data length divided by the step length, the pooling calculation adopts a "valid" mode, when the data length is odd, a one-bit numerical value is abandoned, the number of arc_EffNet modules is recorded as n, when n is 3, the pooling layer output of the third layer arc_EffNet module is 25, and when the final convolution operation is performed, 0 is automatically supplemented to ensure that the data length is unchanged, and after the convolution calculation with the step length of 2, the output length is changed to (25+1)/2=13. When n is 4, the length of the input data of the fourth layer arc_affnet module is 13, when pooling calculation is carried out, one-bit numerical value is abandoned, the length of pooled output data is (13-1)/2=6, and after convolution calculation with the step length of 2, the length of the output data of the fourth layer arc_affnet module is 3 finally. It can be seen that when n is 4, the output length of the fourth layer arc_effnet module is reduced to 3, if the fourth layer arc_effnet module is continuously overlapped, the input dimension of the fifth layer arc_effnet module is lower than the convolution kernel, a large amount of 0 data needs to be supplemented during convolution to maintain the data length, and excessive 0 data can cause the network to be difficult to train or even fail to train. The number of modules n should be 4 or less.
Specifically, model training accuracy curves containing different numbers of arc_affnet modules are shown in fig. 5. The accuracy of the arc_EffNet_2blocks is lower, the accuracy is 87.883% after 150-generation training, the arc_EffNet_3blocks are 81.190% after 10-generation training, and the accuracy curve still keeps rising trend. After 150 generations of training, the final accuracy is 97.214%. And the arc_EffNet_4blocks have the operations of convolutional complement 0, pooling layer data discarding and the like, so that the network accuracy remains unchanged basically in the first 7 generations of training, and the network identification accuracy rises rapidly after 7 generations. After 60 generations, the accuracy rate curve is basically coincident with the accuracy rate curve of arc_EffNet_3blocks, and after 150 generations of training, the accuracy rate reaches 96.960%.
The training results, the number of parameters and the training time of the three models are shown in fig. 6 and 7. The arc fault identification model needs to be compatible with the parameter quantity and the accuracy. From the parameter quantity, the Arc_EffNet_2blocks have the lowest parameter quantity, which is only 81608, the Arc_EffNet_3blocks have the parameter quantity 74624 which is smaller than twice that of the Arc_EffNet_2blocks, and the Arc_EffNet_4blocks have the largest parameter quantity which is 3.4 times that of the Arc_EffNet_3 blocks. From the aspect of recognition accuracy, the accuracy of the training set and the verification set of arc_EffNet_3blocks is highest and is 97.214% and 96.585%, respectively. In addition, the training time of the network increases with the number of arc_affnet modules, and basically has a linear relationship. The arc_EffNet_2blocks have small parameters, but the accuracy is low, and the Arc fault identification requirement cannot be met, so that the arc_EffNet_3blocks with moderate parameter and highest accuracy are selected as an Arc fault identification model.
Further, the Batchsize is the number of samples that the neural network chooses per training. When the Batchsize is larger, more samples are selected each time, the more accurate the gradient is, the smaller the training oscillation is, but the operation amount is large, the memory is insufficient, and meanwhile, the network parameters are slowly updated and the accuracy is reduced due to the reduction of the iteration times. When the Batchsize is smaller, the time for completing one training is increased, the network training time is prolonged, and the training oscillation is increased. In order to balance the training time against the fault arc identification rate, an optimal lot size needs to be sought. Six kinds of the Batchsize of 32, 64, 128, 256, 512 and 1024 are selected for training respectively, and the accuracy curves are shown in FIG. 8. It can be seen that when the batch size is 32, because the batch size is too small, the gradient falling direction is different from the overall gradient falling direction of the data set, so that the accuracy of the network is not obviously improved in the previous training iteration process, the accuracy starts to be rapidly improved after 9 iterations, and the final accuracy reaches 92.180% after 150 generations of training. When the Batchsize is 1024, the accuracy curve is similar to that when the Batchsize is 32, the accuracy of the network is not obviously improved before 12 th generation, the trend of the accuracy curve is similar to that when the Batchsize is 32 after 12 th generation, and the final accuracy is 92.128%. The difference of the accuracy curves is not obvious when the Batchsize is 128, 256 and 512, the accuracy rising rates of the three are almost consistent in the earlier stage of training, and the curves are nearly parallel. The three accuracy curves overlap in the 25 th to 40 th generation training, and the accuracy curves are similar after that, and the final accuracy of the three curves is 93.916%, 94.756% and 94.568% respectively. When the Batchsize is 64, the network performance is optimal, and the accuracy rate rising speed and the final accuracy rate value are the highest. After 150 generations of training, the final accuracy is 96.875%.
FIG. 9 shows the final accuracy and training time at different Batchsize. It can be seen that the training time decreases with increasing battsize, but the difference is smaller, with the longest training time of 150 generations being only 106 seconds more than the shortest time. The accuracy and training time of different BatchSize are combined, and the BatchSize size is selected to be 64.
Specifically, the characteristic data is input into an arc fault recognition model determined based on the scheme to obtain an arc fault recognition result of the current detection period, wherein the arc fault recognition result comprises an arc or no arc.
Therefore, in this embodiment, the affnet network is improved based on the current data characteristics, so that the affnet network can be suitable for arc data, and then whether arc faults exist in the current data of the electrical equipment or not is detected through an arc fault identification model obtained by training the affnet network based on the arc fault current data under various working conditions, so that the arc faults of the electrical equipment under various working conditions can be effectively identified, and meanwhile, the affnet network is a lightweight neural network, has fewer parameter amounts, does not need a large amount of storage space and calculation resources, and is convenient to apply to an embedded platform.
Based on the above embodiments:
as an alternative embodiment, the process of training the affnet network based on arc fault current data under various working conditions to obtain the arc fault recognition model includes:
arc fault current data under various working conditions are obtained;
preprocessing all arc fault current data to obtain a data set meeting training requirements;
dividing the data set into a training set, a verification set and a test set;
training an EffNet network through a training set to obtain an initial recognition model;
inputting the verification set into an initial recognition model, and judging whether the initial recognition model meets arc fault recognition requirements or not based on an arc fault recognition result output by the initial recognition model;
if not, adjusting network super parameters of the EffNet network, repeating the step of training the EffNet network through a training set to obtain an initial recognition model until an arc fault recognition model meeting the arc fault recognition requirement is obtained.
As an alternative embodiment, the process of preprocessing all arc fault current data to obtain a dataset meeting training requirements includes:
cutting all arc fault current data according to a preset time window to obtain a plurality of first data samples with preset lengths;
performing FFT processing on each first data sample to obtain energy spectrum data, and selecting the energy spectrum data of a target frequency band as a second data sample;
carrying out normalization processing on each second data sample to obtain a third data sample;
removing abnormal data in each third data sample to obtain a fourth data sample;
adding a label to each fourth data sample to obtain a fifth data sample;
a data set is derived based on all fifth data samples.
As an alternative embodiment, the target frequency band is 20kHz-100 kHz.
Specifically, considering that the training affnet network has a need for input data, after arc fault current data under various working conditions are obtained, the arc fault current data is preprocessed to obtain a data set meeting the training need, and the preprocessing of the data includes four steps of data segmentation, data normalization, data cleaning and label making, which are described below.
Firstly, carrying out FFT processing on each first data sample to obtain energy spectrum data, and selecting the energy spectrum data of a target frequency band as a second data sample;
specifically, training of the affnet network requires a large amount of arc fault current data, and in order to increase the number of samples, the collected arc fault current data needs to be segmented, and in addition, training of the affnet network has strict requirements on the size of input data, and the size of each data sample in the data set must be kept consistent. According to the method, the data is segmented by using a time window with 10ms as a basic unit, so that a plurality of first data samples are obtained, and the number of arc fault current samples is increased to the greatest extent while the characteristics such as the waveform symmetry of the samples are reserved. For example, assuming a data sampling rate of 250kHz, each arc fault current sample length is 2500, the data length of the energy spectrum data obtained by FFT processing the first data sample is halved to 1250, and since the low frequency signal has little effect on arc identification, the energy spectrum data of 20kHz-100kHz is selected as the second data sample, and is input into the affnet network, i.e., the input sample length of the affnet network is 800.
Then, carrying out normalization processing on each second data sample to obtain a third data sample;
specifically, considering that the result of the collected arc fault current data after the FFT is 0-1420, if the data is directly used for training the neural network, the difference of the numerical values of input data samples is large, the data with small frequency spectrum amplitude can be submerged by the data with large frequency spectrum amplitude, the network can not extract the characteristics from the data samples with small frequency spectrum amplitude, and the accuracy is reduced. Therefore, the input data, namely the second data sample, is normalized to obtain a third data sample, so that each current data in the third data sample is in the [0,1] interval.
Wherein the normalized calculation formula isWherein X is max And X min Maximum and minimum values, respectively, of 800 data points in one data sample.
Then, eliminating abnormal data in each third data sample to obtain a fourth data sample;
specifically, considering that the collection of fault arc current data can be performed at different times, abnormal data such as a blank data value, data waveform distortion and the like can appear in the collection process due to different test conditions, test devices and the like, and EffNet is a data-driven identification model, and the data can cause the reduction of model accuracy. In order to improve the data quality, the collected arc fault current data needs to be cleaned. Abnormal data is deleted through manual screening, so that a fourth data sample is obtained, and 34224 arc fault current samples can be obtained specifically, wherein the arc fault current samples comprise 16026 arc samples and 18198 arc-free samples.
Finally, adding a label to each fourth data sample to obtain a fifth data sample;
specifically, after the data cleaning is completed, a data tag is also required to be manufactured, and the load type and the arc state/no-arc state of the arc fault current data are marked. And integrating the arc fault current data with the corresponding tag to obtain a fifth data sample, and obtaining an arc fault current data set based on all the fifth data samples.
The data set is divided into a training set, a verification set and a test set, and the data set can be divided into the three parts according to the proportion of 75%, 10% and 15%.
The training set is used for training the EffNet network to obtain an initial recognition model, the training set contains most of data in the data set, and in order to ensure the accuracy of arc faults of various loads recognized by the EffNet, the arc data and the arc-free data of the various loads in the training set are as equal as possible; the verification set is used for verifying the training degree of the network in the training process, if over fitting or under fitting occurs, the verification set is also used for adjusting network super parameters such as learning rate and the like, and therefore an arc fault recognition model meeting the arc fault recognition requirement is obtained.
Specifically, a training curve of the arc fault recognition model is shown in fig. 10, and a change curve of the learning rate in the training process of the arc fault recognition model is shown in fig. 11. After 10 times of training, the loss value of the arc fault identification model is reduced from the initial value of 1.918 to 0.199, and the accuracy reaches 93.433%. Within 10-35 generations, the loss value of the training set steadily decreases, and the loss value of the verification set has larger fluctuation but still has a decreasing trend. When training is carried out to 66 th generation, the condition that the loss value drop of 5 continuous periods is smaller than the learning rate drop judgment standard already occurs, which indicates that the neural network is about to converge, and the neural network oscillates at the lowest point of the gradient of the loss function due to the overlarge learning rate. At the moment, the learning rate is reduced for the first time, the gradient descent step length is reduced, the network loss value of the training set is continuously reduced, and the accuracy is further improved. And at the 80 th generation, the learning rate is reduced to the minimum learning rate of 0.00001 again, the loss value is not reduced after 10 th generation training, and the training is stopped automatically to prevent overfitting. After training, the arc fault recognition model has a recognition accuracy rate of 99.904% in the training set and a verification set accuracy rate of 98.574%. In the training process of the arc fault recognition model, the accuracy rate curve trend of the verification set is always consistent with that of the training set, and the conditions of over fitting, under fitting and the like are avoided, so that the effectiveness of the network learning rate updating strategy and the automatic stopping strategy is proved.
The test set is used for verifying the performance of the arc fault recognition model, and assuming that the number of arc samples correctly recognized by the arc fault recognition model is True (TP), the number of correctly recognized arc-free samples is True Negative (TN), the number of samples that are recognized as being arc-free is noted as False Positive (FP), and the number of samples that are recognized as being arc-free is noted as False Negative (FN). According to the first calculation typeCalculating the accuracy of the arc fault recognition model according to the second calculation formula +.>Calculating recall of arc fault recognition model according to third meterArithmetic->And calculating the accuracy of the arc fault identification model.
Specifically, the calculation results of the accuracy rate, recall rate and accuracy rate of the arc fault identification model provided by the application are shown in table 2. It can be seen that the accuracy, recall rate and accuracy of the EffNet network are all above 99.5%, which meets the requirements of arc fault identification.
Table 2 table of accuracy, recall and accuracy of arc fault recognition models
In summary, the application improves the network structure of the EffNet network, deletes the 1×3 convolution layer in the EffNet module, replaces the standard convolution with the one-dimensional convolution, provides the arc_EffNet module structure suitable for one-dimensional Arc current data, compares the influence of the number of analysis modules and the network training super parameters on the Arc fault recognition accuracy, selects the number of modules as 3, builds a lightweight Arc fault recognition model with the batch size as 64, verifies the arc_EffNet model by using the data of the test set, and shows that the model can accurately recognize series Arc faults, and the accuracy, the precision and the recall rate are 99.750%, 99.584% and 99.688%, respectively.
Referring to fig. 12, fig. 12 is a schematic structural diagram of an arc fault detection device provided in the present application, where the arc fault detection device includes:
an acquisition module 1, configured to acquire current data of an electrical device in a current detection period;
an extracting module 2, configured to extract feature data in the current data;
the detection module 3 is used for inputting the characteristic data into the arc fault recognition model to obtain an arc fault recognition result of the current detection period;
the arc fault recognition model is a recognition model obtained by training an EffNet network based on arc fault current data under various working conditions, the EffNet network comprises n EffNet modules, each EffNet module comprises an input layer, an intermediate layer and an output layer, the input layer is a one-dimensional convolution layer with the width of 1, the intermediate layer comprises a space separable convolution layer with the width of 1 and a maximum pooling layer, the output layer is a one-dimensional convolution layer with the width of 1, and n is a positive integer.
Therefore, in this embodiment, the affnet network is improved based on the current data characteristics, so that the affnet network can be suitable for arc data, and then whether arc faults exist in the current data of the electrical equipment or not is detected through an arc fault identification model obtained by training the affnet network based on the arc fault current data under various working conditions, so that the arc faults of the electrical equipment under various working conditions can be effectively identified, and meanwhile, the affnet network is a lightweight neural network, has fewer parameter amounts, does not need a large amount of storage space and calculation resources, and is convenient to apply to an embedded platform.
As an alternative embodiment, the process of training the affnet network based on arc fault current data under various working conditions to obtain the arc fault recognition model includes:
arc fault current data under various working conditions are obtained;
preprocessing all arc fault current data to obtain a data set meeting training requirements;
dividing the data set into a training set, a verification set and a test set;
training an EffNet network through a training set to obtain an initial recognition model;
inputting the verification set into an initial recognition model, and judging whether the initial recognition model meets arc fault recognition requirements or not based on an arc fault recognition result output by the initial recognition model;
if not, adjusting network super parameters of the EffNet network, repeating the step of training the EffNet network through a training set to obtain an initial recognition model until an arc fault recognition model meeting the arc fault recognition requirement is obtained.
As an alternative embodiment, the process of preprocessing all arc fault current data to obtain a dataset meeting training requirements includes:
cutting all arc fault current data according to a preset time window to obtain a plurality of first data samples with preset lengths;
performing FFT processing on each first data sample to obtain energy spectrum data, and selecting the energy spectrum data of a target frequency band as a second data sample;
carrying out normalization processing on each second data sample to obtain a third data sample;
removing abnormal data in each third data sample to obtain a fourth data sample;
adding a label to each fourth data sample to obtain a fifth data sample;
a data set is derived based on all fifth data samples.
As an alternative embodiment, the target frequency band is 20kHz-100 kHz.
As an alternative embodiment, the process of extracting the characteristic data in the current data includes:
performing FFT conversion on the current data so as to convert the current data in the time domain into energy spectrum data in the frequency domain;
and carrying out normalization processing on the energy spectrum data to obtain characteristic data.
As an alternative embodiment, the input layer is a 1×1 one-dimensional convolution layer, the intermediate layer comprises a 3×1 spatially separable convolution layer and a step-size 2 max-pooling layer, and the output layer is a 2×1 one-dimensional convolution layer.
As an alternative embodiment, the affnet network is composed of a stack of 3 affnet modules.
As an alternative embodiment, the batch size of the affnet network is 64.
In another aspect, the present application also provides a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of the arc fault detection method as described in any one of the embodiments above.
For an introduction to a computer readable storage medium provided in the present application, reference is made to the above embodiments, and the description thereof is omitted herein.
The computer readable storage medium provided by the application has the same beneficial effects as the arc fault detection method.
It should also be noted that in this specification, relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (10)
1. An arc fault detection method, comprising:
acquiring current data of electrical equipment in a current detection period;
extracting characteristic data in the current data;
inputting the characteristic data into an arc fault identification model to obtain an arc fault identification result of the current detection period;
the arc fault recognition model is a recognition model obtained by training an EffNet network based on arc fault current data under various working conditions, the EffNet network comprises n EffNet modules, each EffNet module comprises an input layer, an intermediate layer and an output layer, the input layer is a one-dimensional convolution layer with the width of 1, the intermediate layer comprises a space separable convolution layer with the width of 1 and a maximum pooling layer, the output layer is a one-dimensional convolution layer with the width of 1, and n is a positive integer.
2. The arc fault detection method of claim 1, wherein training the affnet network based on arc fault current data under various conditions to obtain an arc fault identification model comprises:
arc fault current data under various working conditions are obtained;
preprocessing all the arc fault current data to obtain a data set meeting training requirements;
dividing the data set into a training set, a verification set and a test set;
training the EffNet network through the training set to obtain an initial recognition model;
inputting the verification set into the initial recognition model, and judging whether the initial recognition model meets arc fault recognition requirements or not based on an arc fault recognition result output by the initial recognition model;
if not, adjusting network super parameters of the EffNet network, and repeating the step of training the EffNet network through the training set to obtain an initial recognition model until an arc fault recognition model meeting the arc fault recognition requirement is obtained.
3. The arc fault detection method according to claim 2, wherein the preprocessing of all the arc fault current data to obtain a data set satisfying training requirements comprises:
cutting all the arc fault current data according to a preset time window to obtain a plurality of first data samples with preset lengths;
performing FFT processing on each first data sample to obtain energy spectrum data, and selecting the energy spectrum data of a target frequency band as a second data sample;
carrying out normalization processing on each second data sample to obtain a third data sample;
removing abnormal data in each third data sample to obtain a fourth data sample;
adding a label to each fourth data sample to obtain a fifth data sample;
a data set is obtained based on all of the fifth data samples.
4. The arc fault detection method according to claim 3, wherein the target frequency band is 20kHz to 100kHz.
5. The arc fault detection method according to claim 1, wherein the process of extracting characteristic data in the current data includes:
performing FFT (fast Fourier transform) on the current data so as to convert the current data in the time domain into energy spectrum data in the frequency domain;
and carrying out normalization processing on the energy spectrum data to obtain characteristic data.
6. The arc fault detection method of claim 1, wherein the input layer is a 1 x 1 one-dimensional convolution layer, the intermediate layer comprises a 3 x 1 spatially separable convolution layer and a step size 2 max pooling layer, and the output layer is a 2 x 1 one-dimensional convolution layer.
7. The arc fault detection method according to any one of claims 1 to 6, wherein the EffNet network is composed of a stack of 3 EffNet modules.
8. The arc fault detection method of claim 7, wherein the batch size of the EffNet network is 64.
9. An arc fault detection apparatus, comprising:
the acquisition module is used for acquiring current data of the electrical equipment in the current detection period;
the extraction module is used for extracting characteristic data in the current data;
the detection module is used for inputting the characteristic data into an arc fault recognition model to obtain an arc fault recognition result of the current detection period;
the arc fault recognition model is a recognition model obtained by training an EffNet network based on arc fault current data under various working conditions, the EffNet network comprises n EffNet modules, each EffNet module comprises an input layer, an intermediate layer and an output layer, the input layer is a one-dimensional convolution layer with the width of 1, the intermediate layer comprises a space separable convolution layer with the width of 1 and a maximum pooling layer, the output layer is a one-dimensional convolution layer with the width of 1, and n is a positive integer.
10. A computer-readable storage medium, characterized in that it has stored thereon a computer program which, when executed by a processor, implements the steps of the arc fault detection method according to any one of claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210931972.7A CN117572161A (en) | 2022-08-04 | 2022-08-04 | Arc fault detection method and device and related components |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210931972.7A CN117572161A (en) | 2022-08-04 | 2022-08-04 | Arc fault detection method and device and related components |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117572161A true CN117572161A (en) | 2024-02-20 |
Family
ID=89883126
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210931972.7A Pending CN117572161A (en) | 2022-08-04 | 2022-08-04 | Arc fault detection method and device and related components |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117572161A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117454166A (en) * | 2023-10-11 | 2024-01-26 | 国网四川省电力公司电力科学研究院 | Method for identifying arc faults of ignition based on EffNet lightweight model |
-
2022
- 2022-08-04 CN CN202210931972.7A patent/CN117572161A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117454166A (en) * | 2023-10-11 | 2024-01-26 | 国网四川省电力公司电力科学研究院 | Method for identifying arc faults of ignition based on EffNet lightweight model |
| CN117454166B (en) * | 2023-10-11 | 2024-05-10 | 国网四川省电力公司电力科学研究院 | Method for identifying arc faults of induced thermal power based on EffNet lightweight model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP3832553A1 (en) | Method for identifying energy of micro-energy device on basis of bp neural network | |
| CN110108992B (en) | Cable partial discharge fault identification method and system based on improved random forest algorithm | |
| CN111650453B (en) | Power equipment diagnosis method and system based on windowing characteristic Hilbert imaging | |
| CN104167208B (en) | A kind of method for distinguishing speek person and device | |
| CN110044623B (en) | Intelligent rolling bearing fault identification method based on empirical mode decomposition residual signal characteristics | |
| CN108594161B (en) | Noise reduction method and system for foreign matter sound signals in electric energy meter | |
| CN110705456A (en) | Micro motor abnormity detection method based on transfer learning | |
| CN113488060B (en) | Voiceprint recognition method and system based on variation information bottleneck | |
| CN112426160A (en) | Electrocardiosignal type identification method and device | |
| CN116167010B (en) | Rapid identification method for abnormal events of power system with intelligent transfer learning capability | |
| CN113866684B (en) | Mixed sampling and cost sensitivity-based distribution transformer fault diagnosis method | |
| CN109493873A (en) | Livestock method for recognizing sound-groove, device, terminal device and computer storage medium | |
| CN112151067B (en) | Digital audio tampering passive detection method based on convolutional neural network | |
| CN110726898A (en) | Power distribution network fault type identification method | |
| CN110909302A (en) | Method and system for learning local disturbance characteristics of operating state parameters of alternating-current and direct-current power grid | |
| CN117572161A (en) | Arc fault detection method and device and related components | |
| CN114386452A (en) | Method for detecting faults of sun wheel of nuclear power circulating water pump | |
| CN103728135A (en) | Bearing fault feature extraction and diagnosis method of non-negative matrix factorization | |
| CN114881077A (en) | Voltage sag source classification method and system based on time sequence trajectory characteristics | |
| CN113378673B (en) | Semi-supervised electroencephalogram signal classification method based on consistency regularization | |
| CN114330430A (en) | Elevator fault judgment method and system based on big data characteristic analysis | |
| CN117768022A (en) | Training method of optical fiber detection model, optical fiber detection method and related equipment | |
| CN103505189A (en) | Pulse signal classification method based on wavelet packet conversion and hidden markov models | |
| CN111929489B (en) | Fault arc current detection method and system | |
| CN109730672B (en) | Characteristic extraction method for multi-lead electrocardiosignal and corresponding monitoring system |
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 |