A method, device, equipment and medium for controlling temperature of air-cooled PEMFC

Abstract

The present invention discloses an air-cooled PEMFC temperature control method, device, equipment and medium, which relates to the technical field of fuel cell temperature control, and includes the following steps: inputting the actual temperature at the previous moment, the actual temperature at the current moment and the control amount into the RBF neural network, forward propagating the RBF neural network, and obtaining the identification temperature of the air-cooled PEMFC; and calculating the partial derivative of the ratio of the identification temperature to the control amount at the current moment to obtain Jacobi information; inputting the actual temperature at the current moment and the target temperature into the BP neural network, performing error back propagation training on the BP neural network based on the Jacobi information, and optimizing the control parameters of the PID controller through the trained BP neural network; optimizing the control amount of the PID controller through the optimized control parameters, and controlling the temperature of the air-cooled PEMFC through the optimized control amount. On the basis of maintaining the traditional PID controller, the present invention adopts the RBF neural network for forward propagation, identifies the temperature of the system, provides Jacobi information to the BP neural network, and the BP neural network is trained by the Jacobi information to realize the online setting of the PID parameters. This self-correcting PID improves the adaptive ability and anti-interference ability of the system, has high adjustment accuracy and fast response speed.

Description

Technical Field

[0001] The present invention relates to the technical field of fuel cell temperature control, and in particular to an air-cooled PEMFC temperature control method, device, equipment and medium. Background Art

[0002] Proton exchange membrane fuel cells (PEMFCs) generate electricity through hydrogen-oxygen combustion reactions with only water as a byproduct, so they are considered to be the most promising green energy. Since PEMFCs are not restricted by the Carnot cycle, their theoretical efficiency at room temperature and pressure can reach 83%. Compared with traditional heat engines, they have higher energy conversion efficiency. In addition, due to their advantages such as low operating temperature, low operating noise, and high power density, PEMFCs are currently being widely studied and vigorously developed. The operating temperature has a significant impact on the performance of PEMFCs. Too high a temperature will cause dehydration of the exchange membrane and some irreversible structural damage, and too low a temperature will reduce the chemical reaction rate. Therefore, it is very important to control PEMFCs to work at the appropriate temperature. PEMFCs generate a lot of heat during operation. Air-cooled PEMFCs dissipate heat through forced convection of the cooling fan to control the appropriate operating temperature and achieve safe and efficient operation. In this process, the air blown in by the cooling fan acts as both a reaction gas and a cooling gas, so the cathode flow channel provides both a reaction site and a cooling site.

[0003] The PEMFC temperature control system has the characteristics of strong nonlinearity, large time lag, large inertia, etc. However, the control parameters of the conventional PID controller cannot be adjusted online, so it cannot get rid of the influence of the system's nonlinearity, time-varying characteristics, etc., and lacks adaptability. In the PEMFC temperature control, problems such as low adjustment accuracy, slow response speed and poor anti-interference ability will occur. Summary of the invention

[0004] The present invention provides an air-cooled PEMFC temperature control method, device, equipment and medium, which solves the problems that the existing conventional PID controller will have low adjustment accuracy, slow response speed, large overshoot and poor anti-interference ability in PEMFC temperature control due to its own defects.

[0005] The present invention provides an air-cooled PEMFC temperature control method, comprising the following steps:

[0006] The actual temperature and target temperature of the air-cooled PEMFC at the previous moment are obtained, and the actual temperature and target temperature at the previous moment are input into the PID controller to obtain the control quantity at the current moment; the temperature of the air-cooled PEMFC is controlled by the control quantity at the current moment to obtain the actual temperature at the current moment;

[0007] The actual temperature at the previous moment, the actual temperature at the current moment and the control amount are input into the RBF neural network, and the RBF neural network is forward propagated to obtain the identification temperature of the air-cooled PEMFC; and the ratio of the identification temperature to the control amount at the current moment is partial derivatived to obtain the Jacobian information;

[0008] The actual temperature and target temperature at the current moment are input into the BP neural network, and the BP neural network is trained for error back propagation based on Jacobi information. The control parameters of the PID controller are optimized through the trained BP neural network.

[0009] The control quantity of the PID controller is optimized through the optimized control parameters, and the temperature of the air-cooled PEMFC is controlled through the optimized control quantity.

[0010] Preferably, the RBF neural network includes an input layer, a hidden layer and an output layer;

[0011] The input layer includes 3 nodes, and the input data includes the control value u(k-1) at time k-1, the actual temperature y(k) at time k, and the actual temperature y(k-1) at time k-1;

[0012] The hidden layer consists of 4 nodes, and its activation function is the Gaussian function basis function R j (x), j = 1, 2, 3, ..., m, m is the number of hidden layer nodes;

[0013] The output layer includes 1 node, and the output is the system temperature y m (k);

[0014] The input of the RBF neural network is:

[0015] x=[u(k-1),y(k),y(k-1)]

[0016] The output is:

[0017]

[0018] In the formula, w j (k-1) is the weight from the jth hidden layer node to the output layer node at time k-1.

[0019] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned air-cooled PEMFC temperature control method when executing the program.

[0020] Preferably, the method of taking the partial derivative of the ratio of the identified temperature to the control amount at the current moment to obtain Jacobian information comprises the following steps:

[0021] Subtract the actual temperature from the system temperature to get the error of the RBF neural network;

[0022] Through the δ learning algorithm, the error back propagation of the RBF neural network is carried out, the weights between the layers of the RBF neural network are adjusted, and the network parameter increment of the adjusted RBF neural network is obtained;

[0023] Calculate the network parameters of the RBF neural network based on the adjusted network parameter increments;

[0024] Calculate the Jacobian information of the system based on the network parameters of the RBF neural network;

[0025] The Jacobian information is shown as follows:

[0026]

[0027] In the formula, u(k) is the control quantity at time k, y m (k) is the identification temperature at time k, c j is the center point of the jth Gaussian function basis function, n y is the number of output nodes, b j is a freely selectable parameter that determines the width of the basis function around the center point.

[0028] Preferably, the BP neural network includes an input layer, a hidden layer and an output layer;

[0029] The input layer includes 4 nodes, and the input data includes y(k), e(k), r(k) and 1, where r(k) is the target temperature and e(k) is the control error; the output of the input layer is:

[0030]

[0031] The hidden layer consists of 5 nodes, and its output is:

[0032]

[0033] The output layer consists of 3 nodes, and its output is:

[0034]

[0035] In the formula, K P ,K I ,K D are proportional unit parameters, integral unit parameters, and differential unit parameters.

[0036] Preferably, error back propagation training is performed on the BP neural network based on Jacobi information, and the control parameters of the PID controller are optimized by the trained BP neural network, comprising the following steps:

[0037] Perform forward propagation on the BP neural network to obtain the actual temperature at time k;

[0038] Subtract the actual temperature at time k from the actual temperature at time k-1 to get the error of the BP neural network;

[0039] Calculate the loss function corresponding to the error;

[0040] Based on Jacobi information, the error back propagation of the loss function is carried out through the gradient descent method to train the network parameter increments of the BP neural network;

[0041] Calculate the network parameters of the BP neural network according to the trained network parameter increments to obtain the trained BP neural network;

[0042] The control parameters of the PID controller are optimized through the trained BP neural network.

[0043] Preferably, the PID controller is an incremental PID, and the algorithm expression of the incremental PID is:

[0044] u(k)=u(k-1)+K p (e(k)-e(k-1))+K i e(k)+K d (e(k)+e(k-2)-2e(k

[0045] -1))

[0046] In the formula, e(k) represents the error at the kth moment, e(k-1) represents the error at the k-1th moment, e(k-2) represents the error at the k-2th moment, u(k-1) represents the control amount at the k-1th moment, and u(k) represents the control amount at the kth moment.

[0047] Preferably, controlling the temperature of the air-cooled PEMFC by the optimized control amount comprises the following steps:

[0048] Constructing the air-cooled PEMFC temperature control system model including cooling fan model, thermal management model and PEMFC battery stack model;

[0049] The cooling fan model is used to calculate the surface heat transfer coefficient;

[0050] The thermal management model is used to calculate the temperature change rate of the PEMFC cell stack model;

[0051] The control quantity is the fan duty cycle;

[0052] The optimized fan duty cycle is substituted into the cooling fan model to change the air flow rate, realize forced convection heat dissipation, and control the temperature of the PEMFC battery stack model.

[0053] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the air-cooled PEMFC temperature control method is implemented.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] On the basis of maintaining the traditional PID controller, the present invention adopts RBF neural network for forward propagation, identifies the temperature of the system, and calculates the partial derivative of the ratio of the identified temperature to the control quantity at the current moment, and provides Jacobian information to the BP neural network. The BP neural network is trained with the Jacobian information to realize online adjustment of PID parameters. This self-correcting PID improves the adaptive ability and anti-interference ability of the system, has high adjustment accuracy and fast response speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0057] Figure 1 It is a structural diagram of the RBF-BP-PID controller of the present invention;

[0058] Figure 2 This is a model structure diagram of the air-cooled PEMFC temperature control system of the present invention;

[0059] Figure 3 Schematic diagram of comparison of step disturbance curves of two PIDs according to an embodiment of the present invention;

[0060] Figure 4 is the K of the RBF-BP-PID controller of the embodiment of the present invention p Parameter setting diagram;

[0061] Figure 5 is the K of the RBF-BP-PID controller of the embodiment of the present invention i Parameter setting diagram;

[0062] Figure 6 is the K of the RBF-BP-PID controller of the embodiment of the present invention d Parameter setting diagram;

[0063] Figure 7 A schematic diagram of an RBF neural network for online identification of the operating temperature of an air-cooled PEMFC according to an embodiment of the present invention;

[0064] Figure 8 The embodiment of the present invention is a schematic diagram of the error between the RBF neural network identified temperature and the actual operating temperature of the air-cooled PEMFC. DETAILED DESCRIPTION

[0065] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0066] The present invention proposes an air-cooled PEMFC temperature control method, specifically an air-cooled PEMFC temperature control method based on RBF-BP-PID, referring to Figure 1 and Figure 2 , a temperature control model of air-cooled PEMFC was established, and a self-correcting PID controller combining RBF neural network and BP neural network was used to overcome the problem that the traditional PID control parameters cannot be adjusted online, so that the air-cooled PEMFC temperature control system has a better control effect. Specifically, the following steps are included:

[0067] Step 1: Build an air-cooled PEMFC temperature control system model and control the air-cooled PEMFC temperature control system model through a PID controller.

[0068] An air-cooled PEMFC temperature control system model, namely a fuel cell thermal management system model, is established, which is mainly composed of a cooling fan model, a thermal management model and a PEMFC cell stack model, and a PID controller is established based on the air-cooled PEMFC temperature control model.

[0069] The cooling fan model, thermal management model and PEMFC battery stack model are as follows:

[0070] The PEMFC stack model uses a semi-empirical equation to calculate the output voltage. The semi-empirical PEMFC model is often used in PEMFC system-level simulation research because of its fast calculation speed and high accuracy:

[0071] V cell =E nerst -η act -η ohmic -η conc

[0072] in,

[0073]

[0074]

[0075] I=i+i n

[0076] η ohmic =IR int =I(R m +R c )

[0077]

[0078]

[0079]

[0080] Where V cell Represents the output voltage, E nerst represents the Nernst voltage, η act represents the activation loss, η ohmic represents the ohmic loss, η conc Represents concentration loss. ξ1, ξ2, ξ3, ξ4, i n , R C ,λ,b,J max There are 9 coefficients, which can be calibrated by experimental data.

[0081] The thermal management model uses the energy conservation equation to calculate the PEMFC temperature change rate:

[0082]

[0083]

[0084] P elec =V stack ×I

[0085]

[0086]

[0087] Where P tot Represents the total input energy of PEMFC, P elec Represents the PEMFC output power, Represents cooling air taking away heat, represents the heat loss due to radiation and natural convection, C t is the PEMFC heat capacity, ΔH is the hydrogen combustion enthalpy, h is the surface heat transfer coefficient, T amb is the ambient temperature, A is the total area of ​​the cathode flow channel, Rth is the equivalent thermal resistance.

[0088] The cooling fan model uses the Zeider-Tate correlation to calculate the surface heat transfer coefficient:

[0089]

[0090]

[0091]

[0092] In the formula, Nu f Re represents the Nusselt number of the cathode flow channel, Re f Reynolds number, Pr f represents the Prandtl number, l c represents the characteristic size, ρ represents the density of air, V max Represents the maximum air velocity, V min represents the minimum air flow rate, PWM represents the fan duty cycle, and V represents the actual air flow rate.

[0093] Step 2: Collect the current system status of the air-cooled PEMFC temperature control system model and the control quantity of the PID controller.

[0094] Step 3: Input the given temperature (control target), identify the identification temperature of the air-cooled PEMFC temperature control system model through the RBF neural network based on the current system state and control quantity, and obtain the Jacobian information of the system based on the identification temperature.

[0095] The RBF neural network is established to achieve the identification of the working temperature of the air-cooled PEMFC and obtain the Jacobian information of the system. The RBF neural network consists of an input layer, a hidden layer and an output layer.

[0096] Set the number of input layer nodes of the RBF neural network to 3, where the input data of the first input node is the control quantity u(k-1) at time k-1, the input data of the second input node is the actual temperature y(k), and the input data of the third input node is the actual temperature y(k-1) at time k-1.

[0097] The hidden layer of the RBF neural network includes 4 nodes, and its activation function is the Gaussian basis function. The Gaussian basis function is:

[0098]

[0099] In the formula, c j is the center point of the jth basis function, and c j =[c j1 ,c j2 ,…,cjn ] T , i∈[1,n], n is the number of input nodes, b j is a freely selectable parameter that determines the width of the basis function around the center point, and m is the number of hidden layer nodes;

[0100] The derivative of the Gaussian basis function is:

[0101]

[0102]

[0103] The output layer of the RBF neural network includes 1 node, and the output is the temperature y identified by the air-cooled PEMFC temperature control system. m (k).

[0104] The input of the RBF neural network is:

[0105] x=[u(k-1),y(k),y(k-1)] (4)

[0106] The output is:

[0107]

[0108] In the formula, w j (k-1) is the weight from the jth hidden layer node to the output layer node at time k-1.

[0109] The signal forward propagation process of the RBF neural network includes:

[0110] The output of the RBF neural network input layer node is:

[0111]

[0112] The output of the hidden layer node of the RBF neural network is:

[0113]

[0114] The output of the RBF neural network output layer node is:

[0115]

[0116] The error of the RBF neural network is:

[0117] e(k)=y(k)-y m (k) (9)

[0118] The performance index function of the RBF neural network is:

[0119]

[0120] The δ learning algorithm is used in the error back propagation process of the RBF neural network to adjust the weights between the layers of the RBF neural network. According to the gradient descent method, the learning algorithm of the weights is as follows:

[0121] The weight w from the hidden layer to the output layer j (k):

[0122] From equation (9) and equation (10), we can get:

[0123]

[0124] Substituting formula (5) into the above formula, we get:

[0125]

[0126] Then the weight w from the hidden layer to the output layer is j The learning algorithm for (k) is:

[0127]

[0128] w j (k) = w j (k-1)+Δw j (k)+α(w j (k-1)-w j (k-2)) (14)

[0129] Where α is the momentum factor and η is the learning rate.

[0130] Hidden layer Gaussian basis function parameter b j (k) and c ji (k):

[0131] According to formula (5) and formula (11), we can get:

[0132]

[0133] Substituting equation (2) and equation (3) into equation (15), we obtain:

[0134]

[0135]

[0136] Then b j (k) and c ji The learning algorithm for (k) is:

[0137]

[0138] b j (k) = bj (k-1)+Δb j (k)+α(b j (k-1)-b j (k-2)) (19)

[0139]

[0140] c ji (k) = c ji (k-1)+Δc ji (k)+α(c ji (k-1)-c ji (k-2)) (21)

[0141] When calculating PID parameters, you need to use (called Jacobian information), but because the model of the controlled object is unknown, the Jacobian information cannot be solved directly, and the following approximate processing can be done:

[0142]

[0143] Since the input of the RBF network is:

[0144] x=[x1(k),x2(k),…,x n (k)] T =[y(k-1),…,y(kn y );u(k),…,u(kn u )] T (twenty three)

[0145] Right now:

[0146]

[0147] The Jacobian information of the system is calculated by RBF neural network:

[0148]

[0149] The present invention further improves the general BP-PID controller, which uses a symbolic function To replace the Jacobian information of the system, the RBF-BP-PID controller can calculate the Jacobian information of the system through the RBF neural network.

[0150] The system parameter identification process of the RBF neural network model includes:

[0151] Input the initial data of the system and set the initial value of the RBF network parameter b j (0), c ji (0), w j(0), learning rate η, momentum factors α, β, etc.

[0152] S1: Use the actual system output y(k) to calculate the current network output y m (k).

[0153] S2: Calculate the network parameter increment Δw j (k), Δb j (k), Δc ji (k).

[0154] S3: Calculate network parameters w j (k), b j (k), c ji (k).

[0155] S4: Calculate the Jacobian information of the system and provide it to the BP neural network for the back propagation process.

[0156] S5: loop once at sampling time k, and enter the next loop at time k+1.

[0157] Step 4: Based on the current system state and Jacobian information, the control parameters of the PID controller are optimized through the BP neural network, and the control quantity of the PID controller is optimized through the optimized control parameters.

[0158] Establish a BP neural network, which accepts the Jacobian information provided by the RBF neural network and optimizes the parameter K of the PID controller online according to the current system state. P , K I and K D .

[0159] The BP neural network consists of an input layer, a hidden layer and an output layer. The input layer of the BP neural network consists of 4 nodes, and the input data includes y(k), e(k), r(k) and 1, where r(k) is the input setting value and e(k) is the control error;

[0160] The output of the input layer of the BP neural network is:

[0161]

[0162] The hidden layer of the BP neural network consists of 5 nodes, and its output is:

[0163]

[0164] The input of the jth node in the hidden layer of the BP neural network is:

[0165]

[0166] The output of the jth node in the hidden layer is:

[0167]

[0168] Its activation function is:

[0169]

[0170] The derivative is:

[0171]

[0172] The output layer of the BP neural network consists of 3 nodes, and its output is:

[0173]

[0174] The activation function of the output layer is:

[0175]

[0176] Its derivative is:

[0177]

[0178] The output of the kth node in the output layer is:

[0179]

[0180]

[0181] Input layer-hidden layer weight matrix W ji for:

[0182]

[0183] Hidden layer-output layer weight matrix V kj for:

[0184]

[0185] The back propagation process of the BP neural network is to determine the optimization direction of the current system parameters based on the system response results, so that the system response evolves in the expected direction:

[0186] Calculate the loss function of the BP neural network:

[0187]

[0188] In the formula, the smaller J is, the better the response performance of the system is.

[0189] The direction in which the loss function J decreases is the direction of optimization, which is achieved using the gradient descent method:

[0190] The gradient descent formula between the hidden layer and the output layer is:

[0191]

[0192] Among them, η represents the learning rate or the descent step size, and the Jacobian information Provided by the RBF neural network, the decomposition expressions of the partial differentials are as follows:

[0193]

[0194]

[0195]

[0196]

[0197] make

[0198]

[0199] The gradient descent formula between the hidden layer and the input layer is:

[0200]

[0201]

[0202] make

[0203]

[0204]

[0205]

[0206] The entire neural network has a total of 35 weight coefficients. The partial derivative of J for each weight coefficient is performed separately. According to the change value obtained by the gradient descent method, the optimization iteration of the entire network parameters is performed. The final iterative expression of the weight coefficient is:

[0207] w ji (k) = w ji (k-1)+Δw ji (k)+αΔw ji (k-1) (51)

[0208] w kj (k) = w kj (k-1)+Δw kj (k)+αΔw kj (k-1) (52)

[0209] Among them, α is the inertia coefficient and αΔw is the inertia term.

[0210] The system parameter identification process of BP neural network includes:

[0211] S1: Input the initial data of the system, set the initial values ​​of BP network parameters y(0), e(0), r(0) and other parameters, such as the number of hidden layer nodes m, learning rate η, inertia coefficient α, etc.;

[0212] S2: Calculate the network parameter increment w ji (k)+αΔw ji (k-1), Δw kj (k)+αΔw kj (k-1);

[0213] S3: Calculate network parameters w kj (k), w ji (k);

[0214] S4: Loop once at sampling time k, enter the next loop at time k+1, and terminate the output when the maximum number of loops is reached.

[0215] Step 5: Control the air-cooled PEMFC temperature control model through the optimized control quantity to achieve the control of the air-cooled PEMFC temperature.

[0216] The PID self-correcting controller accepts the control parameter K provided by the BP neural network P , K I and K D And control the action u(K) through incremental PID output;

[0217] The expression of the incremental PID algorithm is:

[0218] u(k)=u(k-1)+K p (e(k)-e(k-1))+K i e(k)+K d (e(k)+e(k-2)-2e(k-1)) (53)

[0219] In the formula, e(k) represents the error at the kth moment, e(k-1) represents the error at the k-1th moment, e(k-2) represents the error at the k-2th moment, u(k-1) represents the control amount at the k-1th moment, and u(k) represents the control amount at the kth moment.

[0220] The air-cooled PEMFC temperature control model is controlled by the optimized control quantity, and the cooling fan duty cycle PWM is adjusted to change the air flow, thereby changing the surface heat transfer coefficient, and finally achieving the temperature control target of the air-cooled PEMFC. The thermal management system is affected by the change in air flow, and the actual temperature calculated by the temperature control model. The actual temperature of the system is transmitted to the RBF-BP-PID controller in real time for the next round of control.

[0221] Based on the simulation of the present invention and the comparison with the existing self-correcting controller, the experimental results of the present invention and the results of other controllers are as follows:

[0222] from Figure 3 It can be seen that the RBF-BP-PID proposed in the present invention has better control performance than the traditional PID, which is specifically manifested in smaller overshoot and shorter adjustment time.

[0223] Table 1 shows the results of RBF-BP-PID compared with the existing self-correcting controller for the same current disturbance curve.

[0224] Table 1: Results of self-tuning controllers for different controllers

[0225] Controller PID BP-PID RBF-PID RBF-BP-PID IAE 269.3 55.34 35.66 29.89

[0226] IAE is the absolute error integral criterion. The smaller the IAE, the better the control effect. Its definition is as follows:

[0227] IAE=∫|e(t)|dt

[0228] It can be found that the control performance of RBF-BP-PID is improved to a certain extent compared with RBF-PID and BP-PID.

[0229] from Figure 4-Figure 6 It can be seen that the RBF-BP-PID proposed in this invention can achieve K P , K I and K D According to the adaptive changes of system status, the system's anti-interference ability and control performance are improved.

[0230] from Figure 7 and Figure 8 It can be seen that the RBF neural network in the present invention can track the actual operating temperature of the system very well, and even in the process of temperature fluctuation, the absolute error does not exceed 0.04°C.

[0231] Based on the same concept, the present invention also provides an air-cooled PEMFC temperature control device, including an acquisition module, an identification module, an optimization module and a control module.

[0232] The acquisition module is used to obtain the actual temperature and target temperature of the air-cooled PEMFC at the previous moment, and input the actual temperature and target temperature at the previous moment into the PID controller to obtain the control quantity at the current moment; the air-cooled PEMFC is temperature controlled by the control quantity at the current moment to obtain the actual temperature at the current moment.

[0233] The identification module is used to input the actual temperature of the previous moment, the actual temperature of the current moment and the control quantity into the RBF neural network, perform forward propagation on the RBF neural network, and obtain the identification temperature of the air-cooled PEMFC; and calculate the partial derivative of the ratio of the identification temperature to the control quantity at the current moment to obtain the Jacobian information.

[0234] The optimization module is used to input the actual temperature and target temperature at the current moment into the BP neural network, perform error back propagation training on the BP neural network based on Jacobi information, and optimize the control parameters of the PID controller through the trained BP neural network.

[0235] The control module is used to optimize the control quantity of the PID controller through the optimized control parameters, and to control the temperature of the air-cooled PEMFC through the optimized control quantity.

[0236] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the above-mentioned air-cooled PEMFC temperature control method is implemented when the processor executes the program.

[0237] The present invention also provides a computer-readable storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned air-cooled PEMFC temperature control method is implemented.

[0238] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0239] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. An air-cooled PEMFC temperature control method, characterized in that: The following steps are involved: Obtain the actual temperature and target temperature of the air-cooled PEMFC at the previous moment, and input the actual temperature and target temperature at the previous moment into the PID controller to obtain the control quantity at the current moment; The temperature of the air-cooled PEMFC is controlled by the control amount at the current moment to obtain the actual temperature at the current moment; The actual temperature at the previous moment, the actual temperature at the current moment and the control quantity are input into the RBF neural network, and the RBF neural network is forward propagated to obtain the identification temperature of the air-cooled PEMFC; And the ratio of the identified temperature to the control quantity at the current moment is partial derivatived to obtain the Jacobian information; The actual temperature and target temperature at the current moment are input into the BP neural network, and the BP neural network is trained for error back propagation based on Jacobi information. The control parameters of the PID controller are optimized through the trained BP neural network. The control amount of the PID controller is optimized by the optimized control parameters, and the temperature of the air-cooled PEMFC is controlled by the optimized control amount; The RBF neural network includes an input layer, a hidden layer and an output layer; The input layer includes 3 nodes, and the input data includes the control value u(k-1) at time k-1, the actual temperature y(k) at time k, and the actual temperature y(k-1) at time k-1; The hidden layer consists of 4 nodes, and its activation function is the Gaussian function basis function R j (x), j = 1, 2, 3, ..., m, m is the number of hidden layer nodes; The output layer includes 1 node, and the output is the system temperature y m (k); The input of the RBF neural network is: x=[u(k-1),y(k),y(k-1)] The output is: In the formula, w j (k-1) is the weight from the jth hidden layer node to the output layer node at time k-1; The method of obtaining the Jacobian information by taking the partial derivative of the ratio of the identified temperature to the control amount at the current moment comprises the following steps: Subtract the actual temperature from the system temperature to get the error of the RBF neural network; Through the δ learning algorithm, the error back propagation of the RBF neural network is carried out, the weights between the layers of the RBF neural network are adjusted, and the network parameter increment of the adjusted RBF neural network is obtained; Calculate the network parameters of the RBF neural network based on the adjusted network parameter increments; Calculate the Jacobian information of the system based on the network parameters of the RBF neural network; The Jacobian information is shown as follows: In the formula, u(k) is the control quantity at time k, y m (k) is the identification temperature at time k, c j is the center point of the jth Gaussian function basis function, n y is the number of output nodes, b j is a freely selectable parameter that determines the width of the basis function around the center point; The BP neural network includes an input layer, a hidden layer and an output layer; The input layer includes 4 nodes, and the input data includes y(k), e(k), r(k) and 1, where r(k) is the target temperature and e(k) is the control error; the output of the input layer is: The hidden layer consists of 5 nodes, and its output is: The output layer consists of 3 nodes, and its output is: In the formula, K P ,K I ,K D are proportional unit parameters, integral unit parameters and differential unit parameters; The BP neural network is trained with error back propagation based on Jacobi information, and the control parameters of the PID controller are optimized by the trained BP neural network, including the following steps: Perform forward propagation on the BP neural network to obtain the actual temperature at time k; Subtract the actual temperature at time k from the actual temperature at time k-1 to get the error of the BP neural network; Calculate the loss function corresponding to the error; Based on Jacobian information, the error back propagation of the loss function is carried out through the gradient descent method to train the network parameter increments of the BP neural network; Calculate the network parameters of the BP neural network according to the trained network parameter increments to obtain the trained BP neural network; The control parameters of the PID controller are optimized through the trained BP neural network.

2. The air-cooled PEMFC temperature control method according to claim 1, characterized in that: The PID controller is an incremental PID, and the expression of the algorithm of the incremental PID is: u(k)=u(k-1)+K p (e(k)-e(k-1))+K i e(k)+K d (e(k)+e(k-2)-2e(k-1)) In the formula, e(k) represents the error at the kth moment, e(k-1) represents the error at the k-1th moment, e(k-2) represents the error at the k-2th moment, u(k-1) represents the control amount at the k-1th moment, and u(k) represents the control amount at the kth moment.

3. The air-cooled PEMFC temperature control method according to claim 1, characterized in that: The temperature of the air-cooled PEMFC is controlled by the optimized control amount, including the following steps: Constructing the air-cooled PEMFC temperature control system model including cooling fan model, thermal management model and PEMFC battery stack model; The cooling fan model is used to calculate the surface heat transfer coefficient; The thermal management model is used to calculate the temperature change rate of the PEMFC cell stack model; The control quantity is the fan duty cycle; The optimized fan duty cycle is substituted into the cooling fan model to change the air flow rate, realize forced convection heat dissipation, and control the temperature of the PEMFC battery stack model.

4. An air-cooled PEMFC temperature control device, characterized in that: include: The acquisition module is used to obtain the actual temperature and target temperature of the air-cooled PEMFC at the previous moment, and input the actual temperature and target temperature at the previous moment into the PID controller to obtain the control quantity at the current moment; The temperature of the air-cooled PEMFC is controlled by the control amount at the current moment to obtain the actual temperature at the current moment; The identification module is used to input the actual temperature at the previous moment, the actual temperature at the current moment and the control amount into the RBF neural network, and forward propagate the RBF neural network to obtain the identification temperature of the air-cooled PEMFC; And the ratio of the identified temperature to the control quantity at the current moment is partial derivatived to obtain the Jacobian information; The optimization module is used to input the actual temperature and target temperature at the current moment into the BP neural network, perform error back propagation training on the BP neural network based on Jacobi information, and optimize the control parameters of the PID controller through the trained BP neural network; A control module, used for optimizing a control amount of a PID controller through optimized control parameters, and controlling the temperature of an air-cooled PEMFC through the optimized control amount; The RBF neural network includes an input layer, a hidden layer and an output layer; The input layer includes 3 nodes, and the input data includes the control value u(k-1) at time k-1, the actual temperature y(k) at time k, and the actual temperature y(k-1) at time k-1; The hidden layer consists of 4 nodes, and its activation function is the Gaussian function basis function R j (x), j = 1, 2, 3, ..., m, m is the number of hidden layer nodes; The output layer includes 1 node, and the output is the system temperature y m (k); The input of the RBF neural network is: x=[u(k-1),y(k),y(k-1)] The output is: In the formula, w j (k-1) is the weight from the jth hidden layer node to the output layer node at time k-1; The method of obtaining the Jacobian information by taking the partial derivative of the ratio of the identified temperature to the control amount at the current moment comprises the following steps: Subtract the actual temperature from the system temperature to get the error of the RBF neural network; Through the δ learning algorithm, the error back propagation of the RBF neural network is carried out, the weights between the layers of the RBF neural network are adjusted, and the network parameter increment of the adjusted RBF neural network is obtained; Calculate the network parameters of the RBF neural network based on the adjusted network parameter increments; Calculate the Jacobian information of the system based on the network parameters of the RBF neural network; The Jacobian information is shown as follows: In the formula, u(k) is the control quantity at time k, y m (k) is the identification temperature at time k, c j is the center point of the jth Gaussian function basis function, n y is the number of output nodes, b j is a freely selectable parameter that determines the width of the basis function around the center point; The BP neural network includes an input layer, a hidden layer and an output layer; The input layer includes 4 nodes, and the input data includes y(k), e(k), r(k) and 1, where r(k) is the target temperature and e(k) is the control error; the output of the input layer is: The hidden layer consists of 5 nodes, and its output is: The output layer consists of 3 nodes, and its output is: In the formula, K P ,K I ,K D are proportional unit parameters, integral unit parameters and differential unit parameters; The BP neural network is trained with error back propagation based on Jacobi information, and the control parameters of the PID controller are optimized by the trained BP neural network, including the following steps: Perform forward propagation on the BP neural network to obtain the actual temperature at time k; Subtract the actual temperature at time k from the actual temperature at time k-1 to get the error of the BP neural network; Calculate the loss function corresponding to the error; Based on Jacobian information, the error back propagation of the loss function is carried out through the gradient descent method to train the network parameter increments of the BP neural network; Calculate the network parameters of the BP neural network according to the trained network parameter increments to obtain the trained BP neural network; The control parameters of the PID controller are optimized through the trained BP neural network.

5. A computer device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the air-cooled PEMFC temperature control method as described in any one of claims 1 to 3 is implemented.

6. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the air-cooled PEMFC temperature control method according to any one of claims 1 to 3 is implemented.

Images