Online fault diagnosis method and system suitable for AC / DC charger

Abstract

The invention discloses an on-line fault diagnosis method and system suitable for an AC / DC charger, and belongs to the technical field of electrical engineering and automation, and the method carries out the on-line fault diagnosis of the charging operation process of the AC / DC charger, and judges the charging operation real-time state and fault type of the AC / DC charger. According to the method, related data parameters of the real-time charging state of the AC / DC charger are obtained, and the charging state characterization data set of the AC / DC charger is obtained through preprocessing. And establishing an adaptive dynamic diagnosis state network, and optimizing the adaptive dynamic diagnosis state network through an adaptive elastic particle algorithm to obtain an optimal dynamic diagnosis state network. And performing online fault diagnosis on the AC / DC charger through the well trained state network, and outputting a fault diagnosis result. According to the invention, online fault diagnosis of the AC / DC charger can be realized, and large-scale application, normal operation and the like of the AC / DC charger are guaranteed.

Description

Technical Field

[0001] The present invention belongs to the field of electrical engineering and automation technology, and more specifically, relates to an online fault diagnosis method and system suitable for an AC / DC charger. Background Art

[0002] With the development of science and technology, the wide application of AC and DC chargers is a future development trend. In the application of chargers, online fault diagnosis of AC and DC chargers is an important part of the motor charging process. In the prior art, it is a relatively difficult problem to accurately perform online fault diagnosis during the charging process of AC and DC chargers. Therefore, in order to solve the above problems, it is of great significance to provide an online fault diagnosis method and system suitable for AC and DC chargers. Summary of the invention

[0003] In view of the above defects or improvement needs of the prior art, the present invention discloses an online fault diagnosis method and system suitable for AC / DC chargers, thereby solving the technical problem that it is difficult to accurately and in real time diagnose the operating status of AC / DC chargers.

[0004] To achieve the above object, according to one aspect of the present invention, there is provided an online fault diagnosis method applicable to an AC / DC charger, comprising:

[0005] Obtain relevant data parameters of the real-time charging status of the AC / DC charger, and obtain a data set representing the charging status of the AC / DC charger through preprocessing;

[0006] Build an adaptive dynamic diagnosis state network, initialize relevant parameters of the network, and construct an online fault diagnosis model suitable for AC and DC chargers;

[0007] The adaptive dynamic diagnosis state network is optimized by adaptive elastic particle algorithm to obtain the optimal dynamic diagnosis state network;

[0008] Through the well-trained state network, the AC and DC chargers are diagnosed online and the fault diagnosis results are output.

[0009] Preferably, the AC / DC charger refers to an AC charger and a DC charger.

[0010] Preferably, the data parameters related to the real-time operation status of the AC / DC charger are obtained, including: the input voltage V r , output voltage V c 、Input current I r With the output current I c , AC / DC charger operating temperature T r ; AC / DC charger charging start time T0 and charging time T, AC / DC charger insulation resistance R, rechargeable battery rated voltage Ue With rated current I e .

[0011] Preferably, the preprocessing of obtaining the AC / DC charger charging state characterization data set includes normalization, missing value and noise reduction.

[0012] Preferably, the process of building an adaptive dynamic diagnosis state network for charger fault diagnosis, initializing relevant parameters of the network, and constructing an online fault diagnosis model suitable for AC and DC chargers is:

[0013] Step 1: Initialize the number of iterations g = 1, predefine integers N1, N g1 With N g2 , predefined maximum number of iterations g max , initialize the subnetwork Ψ1 with N1 neurons, design the diagonal matrix S1 and two orthogonal matrices U1 and V1 as:

[0014]

[0015]

[0016] Where i,j=1,2,…,N1,0<σ i <1,-1≤u ij ,v ij ≤1; Set the internal connection weight matrix of subnetwork Ψ1 to W1=U1S1V1;

[0017] Step 2: Randomly generate input weight matrix W1 in , the dynamic diagnostic state x1 is obtained by driving the sub-network through the input signal, and the internal state matrix H1 is obtained through formula (5):

[0018] H1=[X1(1),X1(2),…,X1(L)] T (5)

[0019] Where X1(k)=[x1(k) T ,u(k) T ] T , x(k) is the state vector at time k, u(k) is the input vector at time k, k = 1, 2, ..., L, and the following loss function is solved by the adaptive elastic particle algorithm to estimate the output weight matrix W1 out , λ and C are random numbers in the range of 0-1;

[0020]

[0021] Step 3: Initialize the temporary network generation times p=1 during the iteration process and compare pN g1 With N g-1 The size of Ng-1 is the number of sub-network neurons generated in the previous iteration; if pN g1 <N g-1 , then randomly generate a gp The diagnostic status network Ψ gp , where N gp =pN g1 , set the network sub-network weight matrix W according to calculation formula (7), formula (8), formula (9) and formula (10) gp :

[0022] W gp =U gp S gp V gp (7)

[0023]

[0024]

[0025] Among them, S gp is a diagonal matrix, U gp and V gp is an orthogonal matrix, i,j=1,2,…,N1,0<σ i <1,-1≤u ij ,v ij ≤1;

[0026] Step 4: Randomly generate input weight matrix Drive temporary diagnostic status network gp , the internal state matrix H is obtained by formula (5) gp , calculate the output weight matrix And the residual error E(Ψ gp ), compare E(Ψ gp ) and E(Ψ g-1 ), E(Ψ g-1 ) is the residual error of the previous sub-network; if E(Ψ gp )≤E(Ψ g-1 ), set the current diagnosis status network to a temporary network ψ g , record the weight matrix inside the sub-network Internal State Input weight matrix And the internal state matrix Otherwise, set p=p+1 and return to step 3;

[0027] Step 5: Generate a g2 The sub-network Φ of neurons g , calculate the sub-network internal weight matrix according to the method of steps 1 and 2 And the input weight matrix The driver network collects the internal state matrix And calculate the output weight matrix

[0028] Step 6: Add subnetwork Φ g To the temporary network ψ g Get the subnetwork Ψ g , update Ψ according to formula (11), formula (12), formula (13) and formula (14) g The output weight matrix

[0029]

[0030] Update Ψ according to formula (15), formula (16) and formula (17) g The weight matrix W g and the internal state matrix H g :

[0031]

[0032] Step 7: Let g = g + 1, if g < g max , return to step 3, otherwise, the training ends.

[0033] Preferably, the output weight matrix W1 is estimated by an adaptive elastic particle algorithm out The calculation process is:

[0034] Step 1: Initialize the number of elastic particles Initialize the space vector of each elastic particle:

[0035] Step 2: Initialize the maximum number of iterations T max , initialize the current iteration number label t = 1;

[0036] Step 3: Calculate the fitness of each elastic particle, select a part of elastic particles with relatively good fitness values ​​as the search particle group, and the remaining elastic particles are the adaptive particle group. The space vector update method of the search particle group is as shown in formula (18):

[0037]

[0038] in, m=1,2,…,d, is the space vector of the nth elastic particle at the tth iteration, z is a random number, z∈(0,1], P is the warning value, P∈[0,1], ST is the safety threshold, ST∈[0.5,1], Q is a random value that obeys the normal distribution; L is a 1×d matrix, and all internal elements are 1;

[0039] Step 4: The initial space vector update method adapted to the particle swarm is shown in formula (19):

[0040]

[0041] in, is the worst space vector at the tth iteration, X P To search for the best space vector of elastic particles, A is a 1×d matrix, and each element in the matrix is ​​randomly assigned 1 or -1. + =A T (AA T ) -1 ,when When , the fitness of the nth elastic particle is poor, and its space vector is updated;

[0042] Step 5: Update the elastic particle swarm using adaptive Cauchy distribution according to formula (20):

[0043]

[0044] in, is the elastic particle space vector after k mutation updates, X n is the space vector of the nth elastic particle, k(t) is the Cauchy distribution with the degree of freedom being the number of iterations t;

[0045] Step 6: Calculate the space vector of the nth elastic particle at the t+1th iteration using formula (21);

[0046]

[0047] in, is the optimal space vector at the tth iteration, and

[0048] is a random number in the range of 0-1, J n is the fitness value of the nth elastic particle, J worst is the worst fitness value, J best is the best fitness value, is the step length coefficient;

[0049] Step 7: Determine whether the current number of iterations t is equal to the maximum number of iterations T max , if not, set t=t+1 and return to step 3;

[0050] Step 8: Output the optimal space vector, that is, the estimated output weight matrix W1 out .

[0051] Preferably, the online fault diagnosis of the AC / DC charger is performed by training a well-trained state network, and the calculation process of outputting the fault diagnosis result is:

[0052] Step 1: Initialize the well-trained adaptive dynamic diagnosis state network and various parameters, including: number of iterations g, integers N1, N g1 With N g2 , the maximum number of iterations g max , a subnetwork Ψ1 with N1 neurons;

[0053] Step 2: Initialize the adaptive elastic particle algorithm and various parameters, including: the number of elastic particles The space vector of each elastic particle, the maximum number of iterations of the algorithm T max , current iteration number label t;

[0054] Step 3: Collect the relevant data parameters of the real-time operating status of the AC / DC charger and input them into the network, including: the input voltage V r , output voltage V c 、Input current I r With the output current I c , AC / DC charger operating temperature T r ; AC / DC charger charging start time T0 and charging time T, AC / DC charger insulation resistance R, rechargeable battery rated voltage U e With rated current I e .

[0055] Step 4: Estimate the output weight matrix W1 through the adaptive elastic particle algorithm out , the output y(k) of the adaptive dynamic diagnosis state network is calculated by formula (22):

[0056] y(k)=(W1 out ) T X(k) (22)

[0057] where X(k) = [x(k) T ,u(k) T ] T , x(k) is the state vector of the sub-network at time k, and u(k) is the input vector at time k.

[0058] Step 5: Determine the current operating status or fault type of the charger based on the network output result y(k).

[0059] According to another aspect of the present invention, there is provided an online fault diagnosis system for an AC / DC charger, comprising:

[0060] Charger data acquisition controller, used to obtain online relevant data parameters of the real-time operating status of AC and DC chargers;

[0061] A charger data processing controller, used for pre-processing the AC and DC charger operating parameter data;

[0062] A charger fault diagnosis controller is used to perform online fault diagnosis on AC and DC chargers using an adaptive dynamic diagnosis status network.

[0063] Preferably, the charger data acquisition controller collects data including:

[0064] The input voltage V of the AC / DC charger is obtained through the Hall sensor r , output voltage V c 、Input current I r With the output current I c ;

[0065] The operating temperature T of the AC / DC charger is obtained through a thermocouple sensor r ;

[0066] Obtain the charging start time T0 and charging time T of the AC / DC charger through the internal timer;

[0067] Obtain the insulation resistance value R of the AC / DC charger through an insulation resistance meter;

[0068] Obtain the rated voltage U of the rechargeable battery through the communication interface module e With rated current I e .

[0069] Preferably, the charger fault diagnosis controller comprises:

[0070] A parameter initialization module is used to initialize the relevant parameters of the adaptive dynamic diagnosis state network and build an online fault diagnosis model suitable for AC and DC chargers;

[0071] Adaptive dynamic diagnosis state network training module, used to train the state network using the normal operation data and historical fault data of the AC and DC chargers to obtain a well-trained state learning network;

[0072] The adaptive dynamic diagnosis state network prediction module performs fault diagnosis on the AC and DC chargers by training a well-trained adaptive dynamic diagnosis state network to determine whether the charger is currently in normal operation or has some kind of fault.

[0073] Preferably, the charger fault diagnosis controller and the adaptive dynamic diagnosis state network training module mainly complete the following training process, including:

[0074] Step 1: Initialize the number of iterations g = 1, predefine integers N1, N g1 With N g2 , predefined maximum number of iterations g max , initialize the subnetwork Ψ1 with N1 neurons, design the diagonal matrix S1 and two orthogonal matrices U1 and V1 as:

[0075]

[0076] Where i,j=1,2,…,N1,0<σ i <1,-1≤u ij ,v ij ≤1; Set the internal connection weight matrix of subnetwork Ψ1 to W1=U1S1V1;

[0077] Step 2: Randomly generate input weight matrix W1 in , the dynamic diagnostic state x1 is obtained by driving the sub-network through the input signal, and the internal state matrix H1 is obtained through formula (5):

[0078] H1=[X1(1),X1(2),…,X1(L)] T (5)

[0079] Where X1(k)=[x1(k) T ,u(k) T ] T , x(k) is the state vector at time k, u(k) is the input vector at time k, k = 1, 2, ..., L, and the following loss function is solved by the adaptive elastic particle algorithm to estimate the output weight matrix W1 out , λ and C are random numbers in the range of 0-1;

[0080]

[0081] Step 3: Initialize the temporary network generation times p=1 during the iteration process and compare pN g1 With N g-1 The size of N g-1 is the number of sub-network neurons generated in the previous iteration; if pN g1 <N g-1 , then randomly generate a gp The diagnostic status network Ψ gp , where N gp =pN g1 , set the network sub-network weight matrix W according to calculation formula (7), formula (8), formula (9) and formula (10) gp :

[0082] W gp =Ugp S gp V gp (7)

[0083]

[0084] Among them, S gp is a diagonal matrix, U gp and V gp is an orthogonal matrix, i,j=1,2,…,N1,0<σ i <1,-1≤u ij ,v ij ≤1;

[0085] Step 4: Randomly generate input weight matrix Drive temporary diagnostic status network gp , the internal state matrix H is obtained by formula (5) gp , calculate the output weight matrix And the residual error E(Ψ gp ), compare E(Ψ gp ) and E(Ψ g-1 ), E(Ψ g-1 ) is the residual error of the previous sub-network; if E(Ψ gp )≤E(Ψ g-1 ), set the current diagnosis status network to a temporary network ψ g , record the weight matrix inside the sub-network Internal State Input weight matrix And the internal state matrix Otherwise, set p=p+1 and return to step 3;

[0086] Step 5: Generate a g2 The sub-network Φ of neurons g , calculate the sub-network internal weight matrix according to the method of steps 1 and 2 And the input weight matrix The driver network collects the internal state matrix And calculate the output weight matrix

[0087] Step 6: Add subnetwork Φ g To the temporary network ψ g Get the subnetwork Ψ g , update Ψ according to formula (11), formula (12), formula (13) and formula (14) g The output weight matrix

[0088]

[0089] Update Ψ according to formula (15), formula (16) and formula (17) g The weight matrix W g and the internal state matrix H g :

[0090]

[0091] Step 7: Let g = g + 1, if g < g max , return to step 3, otherwise, the training ends.

[0092] Preferably, the charger fault diagnosis controller, the adaptive dynamic diagnosis state network training process, estimates the output weight matrix W1 through the adaptive elastic particle algorithm out Complete the following calculation process, including:

[0093] Step 1: Initialize the number of elastic particles Initialize the space vector of each elastic particle:

[0094] Step 2: Initialize the maximum number of iterations T max , initialize the current iteration number label t = 1;

[0095] Step 3: Calculate the fitness of each elastic particle, select a part of elastic particles with relatively good fitness values ​​as the search particle group, and the remaining elastic particles are the adaptive particle group. The space vector update method of the search particle group is as shown in formula (18):

[0096]

[0097] in, m=1,2,…,d, is the space vector of the nth elastic particle at the tth iteration, z is a random number, z∈(0,1], P is the warning value, P∈[0,1], ST is the safety threshold, ST∈[0.5,1], Q is a random value that obeys the normal distribution; L is a 1×d matrix, and all internal elements are 1;

[0098] Step 4: The initial space vector update method adapted to the particle swarm is shown in formula (19):

[0099]

[0100] in, is the worst space vector at the tth iteration, X P To search for the best space vector of elastic particles, A is a 1×d matrix, and each element in the matrix is ​​randomly assigned 1 or -1. + =A T (AA T) -1 ,when When , the fitness of the nth elastic particle is poor, and its space vector is updated;

[0101] Step 5: Update the elastic particle swarm using adaptive Cauchy distribution according to formula (20):

[0102]

[0103] in, is the elastic particle space vector after k mutation updates, X n is the space vector of the nth elastic particle, k(t) is the Cauchy distribution with the degree of freedom being the number of iterations t;

[0104] Step 6: Calculate the space vector of the nth elastic particle at the t+1th iteration using formula (21);

[0105]

[0106] in, is the optimal space vector at the tth iteration, and is a random number in the range of 0-1, J n is the fitness value of the nth elastic particle, J worst is the worst fitness value, J best is the best fitness value, is the step length coefficient;

[0107] Step 7: Determine whether the current number of iterations t is equal to the maximum number of iterations T max , if not, set t=t+1 and return to step 3;

[0108] Step 8: Output the optimal space vector, that is, the estimated output weight matrix W1 out .

[0109] Preferably, the adaptive dynamic diagnosis state network prediction module performs online fault diagnosis on the AC / DC charger by training a well-trained state network and outputs the fault diagnosis result to complete the following calculation process, including:

[0110] Step 1: Initialize the well-trained adaptive dynamic diagnosis state network and various parameters, including: number of iterations g, integers N1, N g1 With N g2 , the maximum number of iterations g max , a subnetwork Ψ1 with N1 neurons;

[0111] Step 2: Initialize the adaptive elastic particle algorithm and various parameters, including: the number of elastic particles The space vector of each elastic particle, the maximum number of iterations of the algorithm T max , current iteration number label t;

[0112] Step 3: Collect the relevant data parameters of the real-time operating status of the AC / DC charger and input them into the network, including: the input voltage V r , output voltage V c 、Input current I r With the output current I c , AC / DC charger operating temperature T r ; AC / DC charger charging start time T0 and charging time T, AC / DC charger insulation resistance R, rechargeable battery rated voltage U e With rated current I e .

[0113] Step 4: Estimate the output weight matrix W1 through the adaptive elastic particle algorithm out , the output y(k) of the adaptive dynamic diagnosis state network is calculated by formula (22):

[0114] y(k)=(W1 out ) T X(k) (22)

[0115] where X(k) = [x(k) T ,u(k) T ] T , x(k) is the state vector of the sub-network at time k, and u(k) is the input vector at time k.

[0116] Step 5: Determine the current operating status or fault type of the charger based on the network output result y(k).

[0117] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art:

[0118] Obtain relevant data parameters of the real-time charging status of the AC / DC charger, and obtain a data set representing the charging status of the AC / DC charger through preprocessing;

[0119] Build an adaptive dynamic diagnosis state network, initialize relevant parameters of the network, and construct an online fault diagnosis model suitable for AC and DC chargers;

[0120] The adaptive dynamic diagnosis state network is optimized by adaptive elastic particle algorithm to obtain the optimal dynamic diagnosis state network;

[0121] Through the well-trained state network, the AC and DC chargers are diagnosed online and the fault diagnosis results are output.

[0122] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, an online fault diagnosis method applicable to an AC / DC charger is implemented.

[0123] A computer-readable storage medium stores computer instructions, which, when executed by a processor, implement an online fault diagnosis method applicable to an AC / DC charger.

[0124] The online fault diagnosis method for AC and DC chargers of the present invention can use the adaptive elastic particle algorithm to optimize the adaptive dynamic diagnosis state network to perform online fault diagnosis on AC and DC chargers, making up for the defect that the existing technology in charger technology is not easy to accurately perform online fault diagnosis on AC and DC chargers. It lays the foundation for future charger technologies, such as electric vehicle charging and automatic diagnosis of charger faults. During the use of the present invention, the operating status of AC and DC chargers can be monitored in real time, remote diagnosis functions can be realized, on-site services for AC and DC charger fault repairs can be reduced, utilization efficiency can be improved, operation and maintenance costs can be reduced, and it is suitable for unmanned charger work occasions. BRIEF DESCRIPTION OF THE DRAWINGS

[0125] Figure 1 It is a schematic diagram of an AC / DC charger online fault diagnosis system provided by an embodiment of the present invention;

[0126] Figure 2 It is a flow chart of an adaptive dynamic diagnosis state network training process provided by an embodiment of the present invention;

[0127] Figure 3 The present invention provides a flowchart of estimating an output weight matrix through an adaptive elastic particle algorithm. DETAILED DESCRIPTION

[0128] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0129] Example: Figure 1 The figure shows a schematic diagram of an AC / DC charger online fault diagnosis system provided by an embodiment of the present invention. The system includes:

[0130] Charger data acquisition controller, used to obtain online relevant data parameters of the real-time operating status of AC and DC chargers;

[0131] A charger data processing controller, used for pre-processing the AC and DC charger operating parameter data;

[0132] A charger fault diagnosis controller is used to perform online fault diagnosis on AC and DC chargers using an adaptive dynamic diagnosis status network.

[0133] The charger data acquisition controller collects data including:

[0134] The input voltage V of the AC / DC charger is obtained through the Hall sensor r , output voltage V c 、Input current I r With the output current I c ;

[0135] The operating temperature T of the AC / DC charger is obtained through a thermocouple sensor r ;

[0136] Obtain the charging start time T0 and charging time T of the AC / DC charger through the internal timer;

[0137] Obtain the insulation resistance value R of the AC / DC charger through an insulation resistance meter;

[0138] Obtain the rated voltage U of the rechargeable battery through the communication interface module e With rated current I e .

[0139] The charger fault diagnosis controller includes:

[0140] A parameter initialization module is used to initialize the relevant parameters of the adaptive dynamic diagnosis state network and build an online fault diagnosis model suitable for AC and DC chargers;

[0141] Adaptive dynamic diagnosis state network training module, used to train the state network using the normal operation data and historical fault data of the AC and DC chargers to obtain a well-trained state learning network;

[0142] The adaptive dynamic diagnosis state network prediction module performs fault diagnosis on the AC and DC chargers by training a well-trained adaptive dynamic diagnosis state network to determine whether the charger is currently in normal operation or has some kind of fault.

[0143] like Figure 2 As shown, an adaptive dynamic diagnosis state network training process flow chart provided by an embodiment of the present invention includes:

[0144] Step 1: Initialize the number of iterations g = 1, predefine integers N1, N g1 With N g2 , predefined maximum number of iterations g max, initialize the subnetwork Ψ1 with N1 neurons, design the diagonal matrix S1 and two orthogonal matrices U1 and V1 as:

[0145]

[0146] Where i,j=1,2,…,N1,0<σ i <1,-1≤u ij ,v ij ≤1; Set the internal connection weight matrix of subnetwork Ψ1 to W1=U1S1V1;

[0147] Step 2: Randomly generate input weight matrix W1 in , the dynamic diagnostic state x1 is obtained by driving the sub-network through the input signal, and the internal state matrix H1 is obtained through formula (5):

[0148] H1=[X1(1),X1(2),…,X1(L)] T (5)

[0149] Where X1(k)=[x1(k) T ,u(k) T ] T , k = 1, 2, ..., L, solve the following loss function through the adaptive elastic particle algorithm and estimate the output weight matrix W1 out ;

[0150]

[0151] Step 3: Initialize the temporary network generation times p=1 during the iteration process and compare pN g1 With N g-1 The size of pN g1 <N g-1 , then randomly generate a gp The diagnostic status network Ψ gp , where N gp =pN g1 , set the network sub-network weight matrix W according to calculation formula (7), formula (8), formula (9) and formula (10) gp :

[0152] W gp =U gp S gp V gp (7)

[0153]

[0154]

[0155] Among them, Sgp is a diagonal matrix, U gp and V gp is an orthogonal matrix, i,j=1,2,…,N1,0<σ i <1,-1≤u ij ,v ij ≤1;

[0156] Step 4: Randomly generate input weight matrix Drive temporary diagnostic status network gp , the internal state matrix H is obtained by formula (5) gp , calculate the output weight matrix And the residual error E(Ψ gp ), compare E(Ψ gp ) and E(Ψ g-1 ); if E(Ψ gp )≤E(Ψ g-1 ), set the current diagnosis status network to a temporary network ψ g , record the weight matrix inside the sub-network Internal State Input weight matrix And the internal state matrix Otherwise, set p=p+1 and return to step 3;

[0157] Step 5: Generate a g2 The sub-network Φ of neurons g , calculate the sub-network internal weight matrix according to the method of steps 1 and 2 And the input weight matrix The driver network collects the internal state matrix And calculate the output weight matrix

[0158] Step 6: Add subnetwork Φ g To the temporary network ψ g Get the subnetwork Ψ g , update Ψ according to formula (11), formula (12), formula (13) and formula (14) g The output weight matrix

[0159]

[0160] Update Ψ according to formula (15), formula (16) and formula (17) g The weight matrix and internal state matrix of:

[0161]

[0162] Step 7: Let g = g + 1, if g < gmax , return to step 3, otherwise, the training ends.

[0163] like Figure 3 As shown, a flowchart of estimating an output weight matrix by an adaptive elastic particle algorithm provided by an embodiment of the present invention includes:

[0164] Step 2-1: Initialize the number of elastic particles Initialize the space vector of each elastic particle:

[0165] Step 2-2: Initialize the maximum number of iterations T max , initialize the current iteration number label t = 1;

[0166] Step 2-3: Calculate the fitness of each elastic particle, select a part of elastic particles with relatively good fitness values ​​as the search particle group, and the remaining elastic particles are the adaptive particle group. The space vector of the search particle group is updated as shown in formula (18):

[0167]

[0168] in, m=1,2,…,d, is the position of the nth elastic particle in the m-space at the tth iteration, z is a random number, z∈(0,1], P is the warning value, P∈[0,1], ST is the safety threshold, ST∈[0.5,1], Q is a random value that obeys the normal distribution; L is a 1×d matrix, and all internal elements are 1;

[0169] Step 2-4: The initial space vector update method adapted to the particle swarm is shown in formula (19):

[0170]

[0171] Among them, X worst (t) is the worst space vector at the tth iteration, X P To search for the best space vector of elastic particles, A is a 1×d matrix, and each element in the matrix is ​​randomly assigned 1 or -1. + =A T (AA T ) -1 ,when When , the fitness of the nth elastic particle is poor, and its space vector is updated;

[0172] Step 2-5: Update the elastic particle swarm using adaptive Cauchy distribution according to formula (20):

[0173]

[0174] in, is the elastic particle space vector after k mutation updates, x n is the space vector of the nth elastic particle, k(t) is the Cauchy distribution with the degree of freedom being the number of iterations t;

[0175] Step 2-6: Calculate the space vector of the nth elastic particle at the t+1th iteration using formula (21);

[0176]

[0177] Among them, x worst (t) is the worst space vector at the tth iteration, x best (t) is the optimal space vector at the tth iteration, and

[0178] is a random number in the range of 0-1, J n is the fitness value of the nth elastic particle, J worst is the worst fitness value, is the step length coefficient;

[0179] Step 2-7: Determine whether the current number of iterations t is equal to the maximum number of iterations T max , if not, set t=t+1 and return to step 2-3;

[0180] Step 2-8: Output the optimal space vector, that is, the estimated output weight matrix W1 out .

[0181] The specific implementation of each module can refer to the description of the above method embodiment, and the embodiment of the present invention will not be repeated.

[0182] It should be pointed out that, according to the needs of implementation, the various steps / components described in this application can be split into more steps / components, and two or more steps / components or partial operations of steps / components can be combined into new steps / components to achieve the purpose of the present invention.

[0183] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An online fault diagnosis method for an AC / DC charger, characterized in that: The method comprises the following steps: Obtain relevant data parameters of the real-time charging status of the AC / DC charger, and obtain a charging status characterization data set of the AC / DC charger through preprocessing; Build an adaptive dynamic diagnosis state network, initialize relevant parameters of the network, and construct an online fault diagnosis model suitable for AC and DC chargers; The adaptive dynamic diagnosis state network is optimized by adaptive elastic particle algorithm to obtain the optimal dynamic diagnosis state network; Through the well-trained state network, the AC and DC chargers are diagnosed online and the fault diagnosis results are output.

2. The online fault diagnosis method for an AC / DC charger according to claim 1, characterized in that: AC and DC chargers refer to AC chargers and DC chargers.

3. The online fault diagnosis method for an AC / DC charger according to claim 1, characterized in that: Get the relevant data parameters of the real-time operation status of the AC / DC charger, including: the input voltage V r , output voltage V c 、Input current I r With the output current I c , AC / DC charger operating temperature T r ; AC / DC charger charging start time T0 and charging time T, AC / DC charger insulation resistance R, rechargeable battery rated voltage U e With rated current I e .

4. The online fault diagnosis method for an AC / DC charger according to claim 1, characterized in that: The preprocessing of the AC / DC charger state of charge characterization dataset includes normalization, missing value and noise reduction.

5. The online fault diagnosis method for AC / DC charger according to claim 1, characterized in that: The process of building an adaptive dynamic diagnosis state network for charger fault diagnosis, initializing relevant parameters of the network, and constructing an online fault diagnosis model suitable for AC and DC chargers is as follows: Step 1: Initialize the number of iterations g = 1, predefine integers N1, N g1 With N g2 , predefined maximum number of iterations g max , initialize the subnetwork Ψ1 with N1 neurons, design the diagonal matrix S1 and two orthogonal matrices U1 and V1 as: Where i,j=1,2,…,N1,0<σ i <1,-1≤u ij ,v ij ≤1; Set the internal connection weight matrix of subnetwork Ψ1 to W1=U1S1V1; Step 2: Randomly generate input weight matrix W1 in , the dynamic diagnostic state x1 is obtained by driving the sub-network through the input signal, and the internal state matrix H1 is obtained through formula (5): H1=[X1(1),X1(2),…,X1(L)] T (5) Where X1(k)=[x1(k) T ,u(k) T ] T , x(k) is the state vector at time k, u(k) is the input vector at time k, k = 1, 2, ..., L, and the following loss function is solved by the adaptive elastic particle algorithm to estimate the output weight matrix W1 out , λ and C are random numbers in the range of 0-1; Step 3: Initialize the temporary network generation times p=1 during the iteration process and compare pN g1 With N g-1 The size of N g-1 is the number of sub-network neurons generated in the previous iteration; if pN g1 <N g-1 , then randomly generate a gp The diagnostic status network Ψ gp , where N gp =pN g1 , set the network sub-network weight matrix W according to calculation formula (7), formula (8), formula (9) and formula (10) gp : W gp =U gp S gp V gp (7) S gp =diag{σ1,σ2,…,σ Ngp } (8) Among them, S gp is a diagonal matrix, U gp and V gp is an orthogonal matrix, i,j=1,2,…,N1,0<σ i <1,-1≤u ij ,v ij ≤1; Step 4: Randomly generate input weight matrix Drive temporary diagnostic status network gp , the internal state matrix H is obtained by formula (5) gp , calculate the output weight matrix And the residual error E(Ψ gp ), compare E(Ψ gp ) and E(Ψ g-1 ), E(Ψ g-1 ) is the residual error of the previous sub-network; if E(Ψ gp )≤E(Ψ g-1 ), set the current diagnosis status network to a temporary network ψ g , record the weight matrix inside the sub-network Internal State Input weight matrix And the internal state matrix Otherwise, set p=p+1 and return to step 3; Step 5: Generate a g2 A subnetwork of neurons Φ g , calculate the sub-network internal weight matrix according to the method of steps 1 and 2 And the input weight matrix The driver network collects the internal state matrix And calculate the output weight matrix Step 6: Add subnetwork Φ g To the temporary network ψ g Get the subnetwork Ψ g , update Ψ according to formula (11), formula (12), formula (13) and formula (14) g The output weight matrix Update Ψ according to formula (15), formula (16) and formula (17) g The weight matrix W g and the internal state matrix H g : Step 7: Let g = g + 1, if g < g max , return to step 3, otherwise, the training ends.

6. The online fault diagnosis method for an AC / DC charger according to claim 5, characterized in that: An online fault diagnosis model for AC and DC chargers is constructed, and the output weight matrix W1 is estimated by adaptive elastic particle algorithm. out The calculation process is: Step 1: Initialize the number of elastic particles N and the space vector of each elastic particle: Step 2: Initialize the maximum number of iterations T max , initialize the current iteration number label t = 1; Step 3: Calculate the fitness of each elastic particle, select a part of elastic particles with relatively good fitness values ​​as the search particle group, and the remaining elastic particles are the adaptive particle group. The space vector update method of the search particle group is as shown in formula (18): in, m=1,2,…,d, is the space vector of the nth elastic particle at the tth iteration, z is a random number, z∈(0,1], P is the warning value, P∈[0,1], ST is the safety threshold, ST∈[0.5,1], Q is a random value that obeys the normal distribution; L is a 1×d matrix, and all internal elements are 1; Step 4: The initial space vector update method adapted to the particle swarm is shown in formula (19): in, is the worst space vector at the tth iteration, X P To search for the best space vector of elastic particles, A is a 1×d matrix, and each element in the matrix is ​​randomly assigned 1 or -1. + =A T (AA T ) -1 ,when When , the fitness of the nth elastic particle is poor, and its space vector is updated; Step 5: Update the elastic particle swarm using adaptive Cauchy distribution according to formula (20): in, is the elastic particle space vector after k mutation updates, X n is the space vector of the nth elastic particle, k(t) is the Cauchy distribution with the degree of freedom being the number of iterations t; Step 6: Calculate the space vector of the nth elastic particle at the t+1th iteration using formula (21); in, is the optimal space vector at the tth iteration, and is a random number in the range of 0-1, J n is the fitness value of the nth elastic particle, J worst is the worst fitness value, J best is the best fitness value, is the step length coefficient; Step 7: Determine whether the current number of iterations t is equal to the maximum number of iterations T max , if not, set t=t+1 and return to step 3; Step 8: Output the optimal space vector, that is, the estimated output weight matrix W1 out .

7. The online fault diagnosis method for AC / DC charger according to claim 1, characterized in that: By training a well-trained state network, the online fault diagnosis of the AC / DC charger is performed, and the calculation process of outputting the fault diagnosis result is as follows: Step 1: Initialize the well-trained adaptive dynamic diagnosis state network and various parameters, including: number of iterations g, integers N1, N g1 With N g2 , the maximum number of iterations g max , a subnetwork Ψ1 with N1 neurons; Step 2: Initialize the adaptive elastic particle algorithm and various parameters, including: the number of elastic particles The space vector of each elastic particle, the maximum number of iterations of the algorithm T max , current iteration number label t; Step 3: Collect the relevant data parameters of the real-time operating status of the AC / DC charger and input them into the network, including: the input voltage V r , output voltage V c 、Input current I r With the output current I c , AC / DC charger operating temperature T r ; AC / DC charger charging start time T0 and charging time T, AC / DC charger insulation resistance R, rechargeable battery rated voltage U e With rated current I e , Step 4: Estimate the output weight matrix W1 through the adaptive elastic particle algorithm out , the output y(k) of the adaptive dynamic diagnosis state network is calculated by formula (22): y(k)=(W1 out ) T X(k) (22) where X(k) = [x(k) T ,u(k) T ] T , x(k) is the state vector of the sub-network at time k, u(k) is the input vector at time k, Step 5: Determine the current operating status or fault type of the charger based on the network output result y(k).

8. An online fault diagnosis system suitable for AC and DC chargers, characterized in that: Used to implement an online fault diagnosis method for an AC / DC charger as described in any one of claims 1 to 7, The system comprises: Charger data acquisition controller, used to obtain online relevant data parameters of the real-time operating status of AC and DC chargers; A charger data processing controller, used for pre-processing the AC and DC charger operating parameter data; A charger fault diagnosis controller is used to perform online fault diagnosis on AC and DC chargers using an adaptive dynamic diagnosis status network; Among them, the charger data acquisition controller collects data including: The input voltage V of the AC / DC charger is obtained through the Hall sensor r , output voltage V c 、Input current I r With the output current I c ; The operating temperature T of the AC / DC charger is obtained through a thermocouple sensor r ; Obtain the charging start time T0 and charging time T of the AC / DC charger through the internal timer; Obtain the insulation resistance value R of the AC / DC charger through an insulation resistance meter; Obtain the rated voltage U of the rechargeable battery through the communication interface module e With rated current I e ; The charger fault diagnosis controller includes: A parameter initialization module is used to initialize the relevant parameters of the adaptive dynamic diagnosis state network and build an online fault diagnosis model suitable for AC and DC chargers; Adaptive dynamic diagnosis state network training module, used to train the state network using the normal operation data and historical fault data of the AC and DC chargers to obtain a well-trained state learning network; The adaptive dynamic diagnosis state network prediction module performs fault diagnosis on the AC and DC chargers by training a well-trained adaptive dynamic diagnosis state network to determine whether the charger is currently in normal operation or has some kind of fault.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, an online fault diagnosis method applicable to an AC / DC charger as described in any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the computer instruction is executed by the processor, an online fault diagnosis method suitable for an AC / DC charger is implemented as described in any one of claims 1-8.

Images