CN114814391A - Charging pile fault identification method and storage medium - Google Patents
Charging pile fault identification method and storage medium Download PDFInfo
- Publication number
- CN114814391A CN114814391A CN202110063079.2A CN202110063079A CN114814391A CN 114814391 A CN114814391 A CN 114814391A CN 202110063079 A CN202110063079 A CN 202110063079A CN 114814391 A CN114814391 A CN 114814391A
- Authority
- CN
- China
- Prior art keywords
- fault
- layer
- representing
- charging pile
- charging module
- 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
Images
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
Abstract
The invention provides a charging pile fault identification method and a storage medium, comprising the steps of sampling three-phase input current of a charging module; carrying out wavelet decomposition on the three-phase input current by using a wavelet packet energy spectrum method to obtain the fault characteristics of the charging module; and based on the obtained fault characteristics of the charging module, carrying out fault identification by using the trained neural network fault identification model to realize fault positioning. The electric vehicle charging pile fault location method and device have the advantages that the electric vehicle charging pile fault location is realized, the daily maintenance efficiency of the charging pile is improved, the safe and stable work of the charging pile is ensured, the charging experience of electric vehicle users is improved, and the development of the electric vehicle industry is promoted.
Description
Technical Field
The invention relates to a novel charging pile fault identification method, in particular to a novel charging pile fault identification method, and belongs to the technical field of electric automobile driving safety.
Background
The development and progress of the electric automobile industry make people increasingly comfortable and convenient, but the problem with the development and progress is that the energy crisis and the environmental protection crisis which are serious are increased. The traditional fuel oil automobile has the problems of low energy utilization efficiency and pollution. In contrast, electric vehicles have great advantages in both aspects, and in terms of energy efficiency, the full-cycle energy efficiency of electric vehicles is greatly improved along with the improvement of the technical level of power generation equipment and the application of various renewable energy sources such as hydroenergy, wind energy, solar energy and the like in novel power generation. And secondly, in the aspect of emission reduction, the electric automobile does not release any harmful substances when running, and is far superior to a fuel automobile in the aspects of environmental cycle self-cleaning and management supervision. Therefore, the development of the electric automobile is imperative from the two aspects of energy conservation and emission reduction.
The most important of the electric automobiles are batteries and charging facilities, and the charging facilities of the electric automobiles comprise an alternating current charging pile, a vehicle-mounted charger and a direct current charging pile at present. The charging module is not arranged in the alternating-current charging pile, and the alternating-current charging pile needs to be matched with a vehicle-mounted charger for use, and the vehicle-mounted charger is low-power and slow in charging rate and long in charging time. The direct current fills electric pile and fills soon for high-power, embeds the module of charging for satisfy domestic car and passenger train etc. quick charge's demand, be the direction of electric automobile charging facility future development. At the beginning of development and starting of the direct-current charging pile, the quantity of the direct-current charging pile can not meet the requirement of an electric automobile far away, and the charging pile lacks perfect and unified design standards, so that the compatibility problem of a vehicle pile system is caused. At present, along with the development of charging pile technology, the problem of pile supply and demand and the problem of design standard have been solved, but direct current charging pile software and hardware trouble frequently occurs in the operation and maintenance process, and the fault rate is high, and the problem of "bad pile" is obvious. The fault types of the direct current charging pile include communication faults, man-machine interaction device faults, charging pile main circuit faults, charging module faults, charging gun faults, fan faults and the like. Among the above components that may break down, the charging module is a component with the highest probability of failure in the dc charging pile due to the fact that the charging module has a higher working voltage, contains more power electronic power devices that are prone to aging and failure, and has a decreased electrolytic capacitance with time performance, and is a safety weak link of the charging pile.
Daily operation and maintenance of the charging pile mainly relate to four items of content of charging pile body inspection, charging gun inspection, communication inspection and charging function inspection. The first three inspections are mainly inspections by operation and inspection personnel, are easy to find, can be repaired on site mostly, and the fourth inspection is that the owner of the electric automobile finds that the electric automobile cannot be charged for repair. One important reason for the failure of the charging of the dc charging pile is the failure of the charging module, and among various failure reasons of the charging module, the open circuit of the device in the charging module will not cause the breakdown of the charging module in a short time, but if the charging module is operated in a failure state for a long time, the charging module will be damaged irreversibly, and finally the charging pile will be unable to charge, so the operation state of the charging module needs to be paid special attention to in the daily operation and maintenance of the dc charging pile. However, the fault characteristics of the charging module device are not obvious, operation and maintenance personnel cannot carry out on-site maintenance and overhaul, and if the charging module device fails, the charging module device can only be returned to a factory for maintenance, which wastes time and labor. Therefore, the topological structure and the working principle of the charging module of the direct-current charging pile are researched and analyzed, the effective fault characteristics of the charging module are extracted, and the corresponding fault device positioning research is carried out, so that the method has very important significance for the operation and maintenance of the charging module in the direct-current charging pile of the electric automobile.
Disclosure of Invention
The invention aims to provide a novel charging pile fault identification method, so that fault location of a charging pile of an electric automobile is realized, the daily maintenance efficiency of the charging pile is improved, the safe and stable work of the charging pile is ensured, the charging experience of an electric automobile user is improved, and the development of the electric automobile industry is promoted.
The invention adopts the following technical scheme. The invention provides a charging pile fault identification method which comprises the following steps:
sampling three-phase input current of a charging module;
carrying out wavelet decomposition on the three-phase input current by using a wavelet packet energy spectrum method to obtain the fault characteristics of the charging module;
and based on the obtained fault characteristics of the charging module, fault recognition is carried out by using the trained neural network fault recognition model to realize fault positioning.
Further, the method for obtaining the fault characteristics of the charging module by performing wavelet decomposition on the three-phase input current by using a wavelet packet energy spectrum method comprises the following steps:
step 1: carrying out three-layer wavelet packet decomposition on the sampling data, and extracting the decomposition coefficients of 8 frequency bands from low frequency to high frequency of the third-layer wavelet packet decomposition;
step 2: performing single-branch reconstruction on the wavelet packet decomposition coefficient, and extracting a reconstruction signal of each frequency band range of a third layer;
and step 3: and (4) solving the energy value of each frequency band range, and constructing a charging module energy spectrum fault characteristic vector.
Still further, step 2 specifically includes the following steps: representing the original signal by S, S 3,k Representing the layer 3 kth reconstructed signal, the wavelet packet decomposition can be expressed as:
carrying out three-layer wavelet packet decomposition reconstruction on the three-phase input current of the charging module;
the energy value of each band represented by each reconstructed signal is obtained and expressed as follows:
wherein E 3，k Reconstructing a signal s for a layer 3 kth node wavelet packet 3,k E represents the total energy value of the original three-phase input current,
representing the reconstructed signal s 3,k K is 0,1, 7; n is 1,2, and N is the number of sampling points.
The energy spectrum fault feature vector T is constructed in the form of:
T＝[E 3,0 E 3,1 E 3,2 E 3,3 E 3,4 E 3,5 E 3,6 E 3,7 ]。
furthermore, the energy spectrum fault feature vector is normalized, and the finally determined fault feature vector of the charging module is as follows:
where I is the magnitude of the three-phase input current.
Further, the neural network adopts a neural network optimized based on the improved wolf colony algorithm.
Further, the neural network includes: an input layer, a hidden layer and an output layer; the number of the neurons in the input layer is the same as the number of the fault feature vector components; the number of the neurons in the output layer is the same as the number of the charging pile fault identification categories.
Further, a preset membership function is adopted in the hidden layer to perform fuzzification processing on eigenvalues in the fault eigenvector, and fault identification is performed on the processed data; the membership function is shown below:
wherein, i ═ 1,2, …, n 1 ，n 1 The number of the wolfs is the number of the individual wolfs; j ═ 1,2, …, n 2 n 2 Dividing the number of classes, x, for fuzzification i′ Representing the position of the individual wolf i' in the feasible domain space, d i′j′ Representing mean value of membership function, σ i′j′ And represents the standard deviation of the membership function.
Further, the excitation function of the input layer is a hyperbolic tangent function; the hidden layer input is (x) 1 ，x 2 ，...，x j ) Each element in the hidden layer output vector is represented as:
where J is the number of elements of the hidden layer input vector, J represents the element number of the hidden layer input vector, σ () represents the response function of the neuron, b i Threshold, ω, representing hidden layer neurons ij Representing the connection weight value between each layer in the hidden layer, m representing the number of elements of the output vector of the hidden layer, and i representing the number of elements of the output vector of the hidden layer;
the output expression of the output layer is as follows:
in the formula (I), the compound is shown in the specification,
a threshold value representing a neuron of the output layer,
representing the connection weight between layers of the output layer, n representing the number of elements of the output vector of the output layerNumber j 1 Element number, J, representing the hidden layer output vector 1 Representing the number of elements of the hidden layer output vector; k is the element number of the output layer output vector, y k Is the expression of the kth element, S i Representing each element in the hidden layer output vector.
Still further, the determination method of the number l of hidden layer neurons is as follows:
in the formula n 0 Is the number of neurons in the input layer, m 0 Is the number of neurons in the output layer, and a is [1, 10 ]]Constant in between.
The present invention further provides a computer-readable storage medium, which stores a computer program, and the computer program, when executed by a processor, implements the steps of the charging pile fault identification method provided in any one of the possible embodiments provided in the above technical solutions.
Compared with the prior art, the technical scheme adopted by the invention has the following technical effects:
aiming at the two conditions of normal work and open circuit of a filter capacitor of a post-stage phase-shifted full-bridge circuit of a charging module, namely three-phase input current energy spectrums are very similar and fault states are not easy to distinguish, fault characteristics of energy value distribution of each frequency band hidden in a three-phase input current waveform of the charging module are extracted by using a wavelet packet energy spectrum method; increasing the amplitude of the input current of the charging module as a fault characteristic vector element, and establishing a fault characteristic vector formed by the input current and the energy value of each frequency band; the method comprises the steps of extracting a wavelet packet energy spectrum charging module fault sample based on the neural network charging module fault state identification optimized by the improved wolf colony algorithm, inputting the neural network optimized by the improved wolf colony algorithm for training and testing, searching the nonlinear mapping relation between the fault state and the fault characteristic vector, completing the positioning of a charging module fault device, proving the effectiveness of the extracted charging module fault characteristic vector and completing the charging module fault state identification by the neural network optimized by the improved wolf colony algorithm with higher accuracy, and realizing the positioning of the fault device.
The electric vehicle charging pile fault location method and device have the advantages that the electric vehicle charging pile fault location is realized, the daily maintenance efficiency of the charging pile is improved, the safe and stable work of the charging pile is ensured, the charging experience of electric vehicle users is improved, and the development of the electric vehicle industry is promoted.
Drawings
Fig. 1 is a flowchart of a novel charging pile fault identification method.
Fig. 2 is a three-phase input current energy spectrum in normal operation and fault conditions.
Fig. 3 is a flow chart of a neural network algorithm for improving optimization of the wolf pack algorithm in an embodiment of the present invention.
Fig. 4 is a BP network structure diagram.
Fig. 5 is a charging module main circuit.
FIG. 6 is a test failure set diagnostic result in an exemplary embodiment.
Detailed Description
The technical scheme of the invention is further explained in detail by combining the attached drawings:
firstly, the input current of a charging module is determined to be used as a fault characteristic parameter by analyzing the fault of a power tube and the fault of an electrolytic capacitor and combining the VIENNA rectifier working mode and the phase-shifted full-bridge circuit working mode. The method specifically comprises the following steps: (1) power tube fault analysis
1) VIENNA rectifier single power tube open-circuit fault characteristic analysis
When the VIENNA rectifier has a power tube open-circuit fault, the alternating current inductance loses the energy storage loop and only the follow current loop is left, so that the fault phase current is seriously distorted, and the amplitude is greatly reduced. The input current contains abundant information of open-circuit fault characteristics of single power devices of the VIENNA rectifier.
2) Phase-shifted full-bridge converter single power tube open-circuit fault characteristic analysis
The working mode of the phase-shifted full-bridge circuit is known as Q 1 Or Q 4 ，(Q 2 Or Q 3 ) When open circuit fault occursThe phase-shifted full bridge circuit will lose the positive (negative) half cycle power output path. And the loss of the positive (negative) half-cycle power transmission path can reduce the power originally transmitted to the rear-stage phase-shifted full bridge by half, thereby reducing the amplitude of the three-phase input current and having corresponding small distortion. Therefore, the three-phase input current of the charging module also contains fault information of open circuit of the power tube of the phase-shifted full-bridge converter.
(2) Electrolytic capacitor fault analysis
1) Fault feature analysis of filter capacitor on direct current side of VIENNA rectifier
The VIENNA rectifier direct-current side filter capacitor has the functions of buffering energy exchange of the direct-current side and the alternating-current side, restraining harmonic voltage of direct-current side output voltage caused by high-frequency action of a switch, and stabilizing the direct-current side output voltage value of a system within a period of time after a load changes. Therefore, when the capacitor C on the dc side of the VIENNA rectifier or the capacitor C2 is open, a large amount of harmonics are introduced into the input current, which causes distortion of the input current, so that the input current of the charging module also contains fault characteristic information of the open circuit of the filter capacitor of the VIENNA rectifier.
2) Phase-shifted full-bridge circuit filter capacitor fault characteristic analysis
The filter capacitor of the phase-shifted full-bridge circuit has three functions, firstly, the filter capacitor filters the output voltage; then, the voltage stabilizing task is undertaken, and when the load or the input power fluctuates in a small amplitude, the output voltage is stabilized within a certain time; and finally, after the primary side current is in a slowly-changing mode, the primary side current and the filter inductor supply power to a load together. Therefore, when the filter capacitor of the phase-shifted full-bridge converter has an open-circuit fault, a large amount of higher harmonics can be introduced into the output voltage of the charging module, and a capacitor energy storage follow current path is lost for a load. When the phase-shifted full-bridge converter normally works, when the secondary side voltage is in a negative half cycle and the power supply power cannot be transmitted to a load in the primary side current slow changing mode of the phase-shifted full-bridge converter, the filter capacitor and the filter inductor supply power to the load, but if the filter capacitor has an open circuit fault, only the filter inductor supplies power to the load, and the output voltage cannot be maintained to start to drop. When the output voltage of the charging module begins to drop, the charging module begins to increase the input power, i.e., the input current amplitude increases, in order to maintain the output voltage as much as possible. Therefore, the input current of the charging module also contains fault characteristic information of the open circuit of the filter capacitor of the phase-shifted full-bridge converter.
Example (b): the charging pile fault identification method comprises the following steps:
1. sampling three-phase input current of a charging module;
2. the method comprises the steps of extracting energy spectrums of three-phase input circuits in all fault states and constructing energy spectrum fault feature vectors of the charging modules
Through the fault characteristic analysis, it can be known that the three-phase input current of the charging module contains fault information of ten diagnosed faults. The frequency characteristics of the three-phase input current are hidden in time domain waveforms and cannot be directly observed, the three-phase input current of the charging module is subjected to wavelet packet transformation through a wavelet packet analysis method, wavelet packet reconstruction coefficients of each frequency band are extracted, then the energy spectrum of the three-phase input current is calculated, and the unobvious frequency characteristics can be expressed in the form of obvious energy value distribution of each frequency band. The wavelet packet energy spectrum is a spectrum in which the result of decomposition of a wavelet packet is represented as an energy spectrum of each frequency band.
Define the square of the Euclid norm of the signal S | | | S 0 || 2 Is the energy of the original signal and has according to the Parseval energy integral identity:
||S 0 || 2 ＝||S 1,0 || 2 +||S 1,1 || 2 (1)
by analogy, the total energy of the j-th layer after wavelet decomposition is obtained as follows:
after the original signal is decomposed by the wavelet packet, the sum of the energy values of each frequency band of the jth layer is equal to the energy of the original signal, and the total energy of the signal is conserved before and after the decomposition of the wavelet packet. The energy distribution of each frequency band of the signal represents the damage condition of some devices in the circuit, which is consistent with the analysis of the energy spectrum of the three-phase input current of the charging module which can represent the fault type of the module, so that the adoption of the wavelet packet to extract the fault characteristics of the three-phase input current of the charging module so as to realize the fault diagnosis of the charging module is reliable. The specific steps of extracting the charging module fault characteristics by the wavelet packet energy spectrum method are as follows:
(1) sampling the three-phase input current of the charging module, performing three-layer wavelet packet decomposition on the sampled data, and extracting the decomposition coefficients of 8 frequency bands from low frequency to high frequency of the third-layer wavelet packet decomposition.
According to the shannon sampling theorem, the original signal can be completely restored only if the sampling frequency is 2 times of the maximum frequency component of the original signal. When a fault that a single power tube is disconnected occurs, Fourier analysis is carried out on input three-phase current, the maximum frequency component is 2000Hz, the sampling frequency is 4000Hz, and the analysis frequency is half of the sampling frequency, namely 2000 HZ.
(2) And performing single-branch reconstruction on the wavelet packet decomposition coefficient, and extracting a reconstruction signal of each frequency band range of the third layer. Representing the original signal by S, S 3,k Representing the layer 3 kth reconstructed signal, the wavelet packet decomposition can be expressed as:
carrying out three-layer wavelet packet decomposition reconstruction on the three-phase input current of the charging module, and then extracting s 3,k The frequency ranges represented by (k ═ 0,1, 2.., 7)8 reconstructed signals are shown in table 1.
TABLE 1 frequency ranges represented by the respective reconstructed signals
(3) The energy values for the various frequency band ranges in the table above are found. Let E 3，k Reconstructing a signal s for a layer 3, kth node wavelet packet 3,k E represents the total energy value of the original three-phase input current, then:
wherein the content of the first and second substances,
( k 0, 1.. 7; N1, 2.. N; N is the number of sample points) represents the reconstructed signal s 3,k The discrete point amplitude of (a).
(4) And constructing a charging module energy spectrum fault feature vector. The frequency band energy value distribution of the three-phase input current in the normal working state of the charging module is different from the frequency band energy value distribution in the fault state, so that the energy value of the frequency band can be used as an element to form a fault feature vector of the charging module. Taking single-phase current as an example, the energy spectrum fault feature vector T is constructed in the form of:
T＝[E 3,0 E 3,1 E 3,2 E 3,3 E 3,4 E 3,5 E 3,6 E 3,7 ] (5)
because the energy value is usually large, in order to avoid gradient explosion in the neural network learning process, and for convenience of analysis, the energy value of each frequency band is divided by the total energy of the signal, and the fault feature vector is processed in a normalization mode.
The total energy of the frequency band decomposed by the three-phase input current and the third-layer wavelet packet of the charging module is as follows:
when extracting the energy spectrum of the three-phase input circuit in each fault state, it is found that the waveform and the energy spectrum of the three-phase input circuit in the normal working condition and the fault condition of the filter capacitor open circuit of the phase-shifted full-bridge converter are very similar, as shown in fig. 2, based on the situation, if the input current energy spectrum is guessed as the only fault characteristic vector by the guessing neural network, the two states cannot be well distinguished. The accuracy rate of the extracted energy spectrum of the charging module after being input into the neural network for training and testing is only 70%, and the guess is verified. Therefore, other features are needed to assist in training the neural network and improve the accuracy of fault diagnosis. Through the analysis of fault waveforms, the amplitude of three-phase input current is supplemented to the fault characteristic vector elements and is input into the neural network together with the energy values of each frequency band.
Finally, determining the fault feature vector of the charging module as follows:
where I is the magnitude of the three-phase input current.
Taking the open circuit of the phase a power tube of the VIENNA circuit as an example, the extracted fault eigenvector from the eigenvector in this fault state is shown in table 2.
TABLE 2 open-circuit fault eigenvector of A-phase power tube of VIENNA circuit
After the charging module fault feature vector is determined, the charging module fault feature vector is used as an input vector and is input into an input vector of a fault type classifier such as a neural network and a support vector machine, and then fault diagnosis can be carried out.
3. The charging module fault state recognition of the neural network based on the improved wolf colony algorithm optimization is as follows:
(1) neural network design for improving optimization of wolf colony algorithm
The charging module fault state identification researched by the invention selects a neural network optimized by a three-layer improved wolf colony algorithm as a fault classifier, the flow of the neural network optimized by the improved wolf colony algorithm is shown in figure 3, and the charging module fault state identification selects the neural network optimized by the three-layer improved wolf colony algorithm of an input layer, a hidden layer and an output layer as the fault classifier.
The neuron and excitation function design for each layer is as follows:
the method comprises the steps that collected original data are preprocessed through analysis of structural characteristics, functions, common faults and fault mechanisms of a charging pile, obtained key characteristic values are used as input of a fault diagnosis model TWPA-FN, and simulation training is conducted; and the IWPA-FNN model finally outputs a fault diagnosis result with certain credibility after a series of operations are carried out on the input data representing the key characteristic values of the fault.
1) Input layer
And taking the extracted key characteristic value as the input of the model, wherein the number n of the input layer neurons is the same as the number of the fault characteristic vector components, and n is 27.
2) Hidden layer
Fuzzification processing is carried out on the key characteristic value of the fault, for example, the key characteristic value of the fault is divided by using low L, normal N and high H, data with fault attributes are obtained and used for operation of a model, and a membership function of the data is shown as a formula 9.
Wherein i 'is 1,2, j' is 1,2,3, x i′ Representing the position of the individual wolf i' in the feasible domain space, d i′j′ Representing mean value of membership function, σ i′j′ And represents the standard deviation of the membership function.
And after the key characteristic value input by the input layer is processed by the fuzzification layer, establishing a fault diagnosis rule by using the processed data for diagnosis. The excitation function of the layer (i.e. the input layer) is a hyperbolic tangent function, namely, delta (x) is equal to (1-e) x )/(1+e x )。
Let the hidden layer input be (x) 1 ，x 2 ，...，x j ) Implicit inOutput vector S of a layer i As shown in equation 10.
Where 27 is the number of elements of the hidden layer input vector, j represents the element number of the hidden layer input vector, σ () represents the response function of the neuron, b i Threshold, ω, representing hidden layer neurons ij Representing the connection weight value between each layer in the hidden layer, m representing the number of elements of the output vector of the hidden layer, and i representing the number of elements of the output vector of the hidden layer;
the number of neurons in the hidden layer has no definite standard, although theoretically, the network precision can be improved by increasing the number of neurons in the hidden layer, the calculated amount can be increased, and even an overfitting phenomenon can occur after a certain limit is exceeded. Determining the number l of hidden layer neurons according to empirical formula 9 as:
in the formula n 0 Is the number of neurons in the input layer, m 0 Is the number of neurons in the output layer, and a is [1, 10 ]]Constant in between.
The excitation function of the hidden layer adopts a tangent S-shaped excitation function tangsi, and the training method adopts a traindx. The traingdx function in MATLAB is a neural network training algorithm that employs a gradient descent with impulse and an adaptive learning rate. Compared with the traditional BP neural network training algorithm (the BP network structure is shown in figure 4), the algorithm improves the problems that the traditional learning algorithm is easy to fall into the local minimum value problem and the convergence time is long.
3) Output layer
Each element S in the hidden layer output vector i The product of elements corresponding to a connection weight matrix between the hidden layer and the output layer is used as the input quantity of the excitation function of the output layer of the neural network;
inputting the fault key characteristic value of the input layerAnd outputting a fault diagnosis result after the fuzzy transportation. The quantization function (i.e. the excitation function) of the output layer is an s-type function, where f (x) is 1/(1+ e) -x ) After the input key characteristic parameters are operated, the output (y) of the neuron in the layer is obtained 1 ，y 2 ，...，y n ) As shown in equation 12.
In the formula (I), the compound is shown in the specification,
a threshold value representing a neuron of the output layer,
representing the connection weight between each layer of the output layer, n representing the number of elements of the output vector of the output layer, j 1 The element sequence number of the hidden layer output vector is represented, and 18 represents the element number of the hidden layer output vector in the embodiment; k is the element number of the output layer output vector, y k Is the expression of the kth element, S i Representing each element in the hidden layer output vector.
The number of the output layer neurons is the number of the charging module fault states, so that the number of the output layer neurons is 11 (including normal working states), and a linear transfer function purelin is selected as an excitation function of the output layer. The output result is expressed by adopting a '1 in n' expression, namely that the charging module 00000000000 works normally, and the A-phase power tube of the 01000000000 electric module VIENNA rectifier circuit is open, namely F 1 By analogy, the 11 fault states of the charging module are represented by 11-bit numbers in this form.
(2) Neural network sample set construction for improving optimization of wolf colony algorithm
The neural network optimized by the improved wolf colony algorithm can complete the nonlinear mapping of the fault characteristic vector and the fault state, has good generalization capability and depends on a complete and sound sample set to a great extent, so that the construction of a reasonable and comprehensive sample set is very important. It should be noted that, the method for optimizing the neural network by using the improved wolf colony algorithm is the prior art, and is not described in detail in this application.
When designing a sample, the number of the samples and the quality of the samples are considered, and the sample design is generally carried out from three aspects of the number of the samples, the composition of the samples and the form of the samples:
the positions of ten fault devices needing positioning in a charging module are shown in fig. 5, three load conditions of rated load, 1.2-time overload and 0.8-time underload are considered for the eleven fault states, the random fluctuation of electrolytic capacitance within a 10% range is considered under each load, finally 50 working conditions are determined, 11 fault states are respectively set under the 50 working conditions to carry out 550 fault simulations, 550 sets of fault data are collected, and 50 sets of collected data exist in each fault state.
For the 550 collected three-phase input currents, 550 groups of fault characteristic vectors are extracted by a MATLAB wavelet packet energy spectrum program to form a neural network sample set optimized by the improved wolf colony algorithm. The fault feature vector table under the rated working condition is shown as an appendix A, wherein 550 sets of fault samples are divided into two parts, 429 sets are used for training the neural network optimized by the improved wolf colony algorithm, and 121 sets are used for testing the fault state identification effect of the neural network optimized by the improved wolf colony algorithm. Table 3 inputs training data for the improved wolf pack algorithm optimized neural network for the charging module, including 429 sets of fault feature vectors and corresponding 429 sets of expected outputs. The training data of the charging module is input into the neural network optimized by the improved wolf colony algorithm for neural network training, a plurality of groups of test samples of the charging module are input into the trained neural network optimized by the improved wolf colony algorithm for testing, the plurality of groups of fault characteristic vectors are diagnosed, fault state recognition is completed, and a fault device is positioned.
TABLE 3 charging Module training data
(3) Charging module fault state recognition result
Writing a neural network charging module fault state recognition program optimized by an improved wolf pack algorithm in MATLAB, inputting the table 3 into the neural network optimized by the improved wolf pack algorithm for neural network training, adjusting the connection weight of the neural network optimized by the improved wolf pack algorithm according to expected output, and finally mastering the mapping relation between the charging module fault state and the fault characteristic vector. Then, the test samples of the set 121 of the charging module are input into a trained neural network optimized by the improved wolf colony algorithm for testing, and the neural network does not refer to the expected output any more and diagnoses the set 121 of fault feature vectors alone. If the diagnosis fault state obtained by the certain group of fault characteristic vectors through the neural network optimized by the improved wolf colony algorithm is the same as the expected fault state, the fault characteristic vectors represent that the group of fault characteristic vectors complete fault state identification through the neural network, and a fault device is positioned.
The failure state recognition results of the charging module 121 sets of test samples (11 sets of data for each failure state) are shown in a scatter diagram 6. In fig. 6, 11 points on the abscissa represent the serial numbers of the test samples, 11 points on the ordinate represent 11 fault states, a scatter "·" represents the expected outputs (expected fault states) of 121 sets of test samples (fault feature vectors), and a scatter "□" represents the outputs (diagnostic fault states) obtained by inputting the 121 sets of test samples (fault feature vectors) into the trained neural network optimized by the improved wolf pack algorithm. Drawing an expected output at each of 121 test samples, and drawing a scatter point □ at the expected output of a certain test sample of the neural network if the diagnostic output is the same as the expected output, wherein □ is coincident with
I.e. indicating that the fault status of the group of samples is correctly identified; if the diagnostic output and the expected output of a certain group of test samples of the neural network are different, a scatter point □ is not drawn at the expected output "·" of the test sample, and only "·" exists at the moment, namely the fault state of the group of test samples is identified incorrectly. It can be seen from the figure that, in the 121 test samples, the sample number 10 in the F5 fault state is identified incorrectly, and the F6 fault state is identified incorrectlySample No. 9 below identifies an error, and sample No. 6 in the F10 fault state identifies an error.
And (3) counting the fault state identification results of the test samples of the neural network optimized by the improved wolf colony algorithm, wherein the statistical results are shown in a table 4.
TABLE 4 charging Module test sample Fault status identification results
In the current charging module fault state identification, only three groups of fault samples are tested, wherein the accuracy is 97.52%, the diagnosis time is 26 seconds, and the iteration times are 15591, so that the neural network model optimized by the improved wolf colony algorithm is correct and has excellent performance, the identification of 11 fault states of the charging module can be completed at a high speed, and the positioning of a fault device is realized.
The invention discloses a novel charging pile fault identification method, which aims at the problems that a charging module of a direct-current charging pile of an electric automobile has high fault frequency and fault features are not obvious and fault devices are difficult to find and identify, deeply analyzes the fault features of the charging module through the working mode and the fault waveform of the charging module, extracts the fault features of three-phase input current of the charging module by adopting a wavelet packet energy spectrum method, constructs a charging module fault feature vector formed by energy values of all frequency bands of the extracted three-phase input current and amplitude values of the three-phase input current together, and finally identifies the fault state of the charging module by using a neural network method optimized by an improved wolf pack algorithm, realizes the positioning of the fault devices, and verifies the effectiveness of the extracted fault features and the correctness of the fault state identification method. The method provides reference for fault identification of the electric vehicle charging pile.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing a flow process in the flowchart
Or steps of a function specified in one or more blocks of the flowchart and/or block diagram.
While the present invention has been described with reference to the embodiments shown in the drawings, the present invention is not limited to the embodiments, which are illustrative and not restrictive, and it will be apparent to those skilled in the art that the invention can be practiced without departing from the spirit of the invention and the scope of the appended claims
While the invention has been described in connection with what is presently considered to be the most practical and preferred embodiment, it is to be understood that the invention is not to be limited to the disclosed embodiment.
Claims (10)
1. Charging pile fault identification method is characterized by comprising the following steps:
acquiring three-phase input current of a charging module;
carrying out wavelet decomposition on the three-phase input current by using a wavelet packet energy spectrum method to obtain the fault characteristics of the charging module;
and based on the obtained fault characteristics of the charging module, carrying out fault identification by using the trained neural network fault identification model to realize fault positioning.
2. The charging pile fault identification method according to claim 1, wherein the method for obtaining the charging module fault characteristics by performing wavelet decomposition on three-phase input current by using a wavelet packet energy spectrum method comprises the following steps:
step 1: carrying out three-layer wavelet packet decomposition on the sampling data, and extracting the decomposition coefficients of 8 frequency bands from low frequency to high frequency of the third-layer wavelet packet decomposition;
step 2: performing single-branch reconstruction on the wavelet packet decomposition coefficient, and extracting a reconstruction signal of each frequency band range of a third layer;
and step 3: and (4) solving the energy value of each frequency band range, and constructing a charging module energy spectrum fault characteristic vector.
3. The charging pile fault identification method according to claim 2, wherein the step 2 specifically comprises the following steps: representing the original signal by S, S 3,k Representing the layer 3 kth reconstructed signal, the wavelet packet decomposition can be expressed as:
carrying out three-layer wavelet packet decomposition reconstruction on the three-phase input current of the charging module;
the energy value of each band represented by each reconstructed signal is obtained and expressed as follows:
wherein E 3，k Reconstructing a signal s for a layer 3 kth node wavelet packet 3,k E represents the total energy value of the original three-phase input current,
representing the reconstructed signal s 3,k K is 0,1, 7; n is 1,2, and N is the number of sampling points;
the energy spectrum fault feature vector T is constructed in the form of:
T＝[E 3,0 E 3,1 E 3,2 E 3,3 E 3,4 E 3,5 E 3,6 E 3,7 ]。
4. the charging pile fault identification method according to claim 3,
normalizing the energy spectrum fault characteristic vector, wherein the finally determined fault characteristic vector of the charging module is as follows:
where I is the magnitude of the three-phase input current.
5. The charging pile fault identification method according to claim 1, wherein the neural network comprises: an input layer, a hidden layer and an output layer; the number of the neurons in the input layer is the same as the number of the fault characteristic vectors, and the number of the neurons in the output layer is the same as the number of the charging pile fault identification categories.
6. The charging pile fault identification method according to claim 5, wherein the neural network is optimized based on a modified wolf pack algorithm.
7. The charging pile fault identification method according to claim 5, wherein a preset membership function is adopted in the hidden layer to fuzzify eigenvalues in fault eigenvectors, and fault identification is performed on processed data; the membership function is shown below:
wherein, i ═ 1,2, …, n 1 ，n 1 The number of wolfs is individual; j ═ 1,2, …, n 2 n 2 Dividing the number of classes, x, for fuzzification i′ Representing the position of the individual wolf i' in the feasible domain space, d i′j′ Representing mean value of membership function, σ i′j′ And represents the standard deviation of the membership function.
8. The charging pile fault identification method according to claim 5, wherein an excitation function of the input layer is a hyperbolic tangent function; the hidden layer input is (x) 1 ，x 2 ，...，x j ) Each element in the hidden layer output vector is represented as:
where J is the number of elements of the hidden layer input vector, J represents the element number of the hidden layer input vector, σ () represents the response function of the neuron, b i Threshold, ω, representing hidden layer neurons ij Representing the connection weight between layers in the hidden layer, m representing the number of elements of the output vector of the hidden layer, and i representing the number of elements of the output vector of the hidden layer;
the output expression of the output layer is as follows:
in the formula (I), the compound is shown in the specification,
a threshold value representing a neuron of the output layer,
representing the connection weight between each layer of the output layer, n representing the number of elements of the output vector of the output layer, j 1 Number of elements representing hidden layer output vector, J 1 Representing the number of elements of the hidden layer output vector; k is the element number of the output layer output vector, y k Is the expression of the kth element, S i Representing each element in the hidden layer output vector.
9. The charging pile fault identification method according to claim 5,
the determination method of the number l of the hidden layer neurons is as follows:
in the formula n 0 Is the number of neurons in the input layer, m 0 Is the number of neurons in the output layer, and a is [1, 10 ]]Constant in between.
10. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110063079.2A CN114814391A (en) | 2021-01-18 | 2021-01-18 | Charging pile fault identification method and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110063079.2A CN114814391A (en) | 2021-01-18 | 2021-01-18 | Charging pile fault identification method and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114814391A true CN114814391A (en) | 2022-07-29 |
Family
ID=82523870
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110063079.2A Pending CN114814391A (en) | 2021-01-18 | 2021-01-18 | Charging pile fault identification method and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114814391A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116384980A (en) * | 2023-05-25 | 2023-07-04 | 杭州青橄榄网络技术有限公司 | Repair reporting method and system |
-
2021
- 2021-01-18 CN CN202110063079.2A patent/CN114814391A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116384980A (en) * | 2023-05-25 | 2023-07-04 | 杭州青橄榄网络技术有限公司 | Repair reporting method and system |
| CN116384980B (en) * | 2023-05-25 | 2023-08-25 | 杭州青橄榄网络技术有限公司 | Repair reporting method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102841251B (en) | Electric car charging station harmonic wave detection method in microgrid | |
| CN112069727B (en) | Intelligent transient stability evaluation system and method with high reliability for power system | |
| CN113093058A (en) | NPC three-level inverter open-circuit fault diagnosis method | |
| CN110619321A (en) | Charging pile technical condition diagnosis method and system based on deep neural network | |
| CN113702767B (en) | Island direct-current microgrid fault diagnosis method based on wavelet sliding window energy | |
| Damala et al. | A simple decision tree-based disturbance monitoring system for VSC-based HVDC transmission link integrating a DFIG wind farm | |
| CN114814391A (en) | Charging pile fault identification method and storage medium | |
| Sun et al. | Data augmentation strategy for power inverter fault diagnosis based on wasserstein distance and auxiliary classification generative adversarial network | |
| CN105550450B (en) | Electric energy quality interference source characteristic harmonic modeling method | |
| CN116882838A (en) | Short-term voltage stability assessment method based on Stacking integration strategy | |
| CN113092934B (en) | Single-phase earth fault judgment method and system based on clustering and LSTM | |
| Behzadi et al. | Identification of combined power quality disturbances in the presence of distributed generations using variational mode decomposition and K-nearest neighbors classifier | |
| CN110542835B (en) | Detection method, detection system and test system for vehicle arc fault | |
| CN113419172A (en) | New energy automobile three-phase asynchronous motor fault identification based on GCN and VMD-ED | |
| Du et al. | Research on fault diagnosis method of DC charging pile based on deep learning | |
| CN112993977A (en) | HHT and RVM based active power distribution network electric energy quality disturbance identification method | |
| Panigrahi et al. | Fault detection and classification using wavelet transform and neuro fuzzy system | |
| Geng et al. | IGBT Open Circuit Fault Diagnosis Based on Improved Support Vector Machine | |
| Zhan et al. | Fault diagnosis of DAB converter based on wavelet packet-RBF neural network | |
| Malik et al. | Investigation of intelligent deep convolution neural network for DC-DC converters faults detection in electric vehicles applications | |
| Chu et al. | A relaxed support vector data description algorithm based fault detection in distribution systems | |
| Wang et al. | Fault diagnosis method for hybrid HVDC transmission system based on stacked sparse auto-encoder | |
| CN114019310B (en) | Open-circuit fault pre-judging method, equipment and medium for energy storage converter | |
| CN214041604U (en) | On-line monitoring and evaluating device for electric energy quality of ship power supply system | |
| Yang et al. | Study on artificial immune mechanism for distribution network fault diagnosis with large scale renewable energy integration |
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 |