Fuel cell rapid cold start method considering initial water content optimization

Abstract

The invention discloses a fuel cell rapid cold start method considering initial water content optimization, and belongs to the technical field of fuel cell automobile optimization control. According to the fuel cell rapid cold start method considering initial water content optimization, a closed-loop model prediction controller is firstly designed to accurately control the current of a galvanic pile, and then the initial film water content is optimized by using a particle swarm optimization algorithm, so that the rapidity of a cold start system is improved. The method can be structurally divided into the following parts: establishment of a cold start model, design of a parameter self-tuning model prediction controller and optimization of the optimal initial membrane water content. In the cold start process, the current output of the electric pile is adjusted in real time according to different environment temperatures and membrane water contents, controller parameters are adjusted in a self-adaptive mode, the freezing rate and the heating rate of water produced by the electric pile can be effectively balanced, and the performance of a cold start system is improved.

Description

Technical Field

[0001] The invention belongs to the technical field of fuel cell vehicle optimization control. Background Art

[0002] In the context of energy transformation, hydrogen energy, as a clean energy carrier with high energy density and zero emission characteristics, has gradually become a key technology to promote the development of a low-carbon economy, and has unique application prospects in transportation, industry, energy storage and other fields that are difficult to decarbonize. Proton exchange membrane fuel cells are widely promoted in the form of on-board batteries. They are an important application of hydrogen energy in the field of transportation and are also a research focus in the field of new energy vehicles.

[0003] The basic working principle of proton exchange membrane fuel cells is: hydrogen and air are supplied to the anode and cathode respectively, and the two react electrochemically in the membrane electrode catalyst layer under the action of the catalyst, converting chemical energy into electrical energy, while generating water and heat. During the normal operation of the fuel cell stack, part of the generated water remains inside the fuel cell stack to maintain the humidity of the proton exchange membrane and ensure the working efficiency of the fuel cell stack; the other part is discharged from the fuel cell stack by the water management system for recycling.

[0004] However, during the cold start process in a low-temperature environment, the temperature inside the stack is affected by the ambient temperature and is below 0°C. The water produced by the reaction cannot be discharged from the stack in time and will freeze on the surface of the proton exchange membrane, in the catalyst layer and in the porous structure of the gas diffusion layer. If the catalyst layer is completely covered by ice before the temperature of the stack rises to 0°C, the catalytic reaction will lose its reaction site, causing the reaction to stop, and then leading to a cold start failure. Therefore, it is necessary to precisely control the cold start process, balance the rate of water and heat production inside the stack, and ensure a successful cold start of the fuel cell.

[0005] Among the many parameters that affect the cold start efficiency of fuel cells, the initial membrane water content of the stack is one of the key parameters that determines the success of the cold start process. If the initial membrane water content is too high, the stack water content will rise too quickly in the early stage of the cold start, and a large amount of liquid water will precipitate and freeze until the cold start fails; if it is too low, the proton exchange membrane will have a low degree of hydration, and the early reaction speed will be too slow, which will slow down the overall cold start process. Therefore, it is particularly important to optimize the initial water content of the fuel cell cold start system so that the stack can improve the cold start speed under the premise of a successful cold start.

[0006] The main problems in designing the control strategy for the fuel cell cold start system are as follows: 1. The existing cold start strategy lacks optimization of the initial water content of the membrane. Using the same initial water content value at different ambient temperatures will result in a slow start-up speed in a sub-zero high temperature environment and difficulty in cold start in a low temperature environment.

[0007] 2. Existing fuel cell cold start strategies mostly use an open-loop method to set the stack operating point and maintain a constant output characteristic throughout the cold start process. However, this method is difficult to achieve a dynamic balance between the freezing rate and heating rate of the stack water, which not only affects the rapidity of the cold start, but may even cause cold start failure, significantly restricting the system's startup performance. Summary of the invention

[0008] The purpose of the present invention is to first design a closed-loop model predictive controller to accurately control the stack current, and then use a particle swarm algorithm to optimize the initial membrane moisture content, so as to achieve a fuel cell fast cold start method that considers the optimization of the initial water content to improve the rapidity of the cold start system.

[0009] The steps of the present invention are: S1. Establish a control-oriented model for the fuel cell cold start system: The state variable is selected as the stack ice volume fraction φ ice and stack temperature T cell , build the second-order control-oriented model as follows: Where: is the change rate of ice volume fraction in the stack, S f-i is the source term for the conversion of frozen water to solid water, S i-n is the source term for the conversion of solid water to unfrozen water, S l-i is the phase transition source term between solid water and liquid water, S v-i is the source term of the condensation process of gaseous water into solid water, ρ ice is the density of solid water ice, ε is the porosity of the catalyst layer, is the temperature change rate of the battery stack, Q rea The heat generated by the catalytic reaction, Q amb To dissipate heat in low temperature environment, C cell is the specific heat capacity of the battery stack, M cell is the mass of the battery stack, C ice is the specific heat of ice, M ice is the mass of ice inside the stack; S2. Based on the control-oriented model, the model predictive stack current controller with self-tuning parameters is designed: Select the state variables as ice volume fraction and stack temperature, select the controller control variable as stack current, and design the model to predict the controller cost function as follows: Among them I cell (·|k)=[I cell (k|k),I cell (k+1|k)…I cell (k+N p-1|k)] is the optimized control sequence; the cost function J is composed of three sub-functions J1, J2, and J3: J1 = μ1φ ice (k+i|k), J3=μ3ΔI cell (k+i|k); N p is the prediction time domain step; μ=[μ1,(μ 21 +μ 22 T amb ),μ3] is the weight coefficient of each penalty term in the cost function; By adding constraints on state quantities and control quantities, the performance function of the cold start system controller is obtained as follows: S3. Optimization of initial membrane water content of the stack: S31. Define the fitness function of the particle swarm algorithm The fitness function is defined as follows: In the formula A Gaussian function centered at is a fourth-order polynomial with vertices, α1, α2, σ1, σ2 are weight coefficients of each item in the fitness function; S32. Define the boundary conditions of the variables to be optimized Set the minimum initial membrane water content of the stack λ min and the stack saturated membrane water content λ sat , so the water content range of the stack membrane is λ ini ∈[λ min ,λ sat ], and set the number of particles N and the upper limit of the number of iterations T; S33, start loop optimization When the fitness function value of a particle reaches f i,k =α1+α2, it means that particle i has reached the optimal initial membrane moisture content under the current ambient temperature after iteration k, and the stopping condition of the particle swarm optimization algorithm is met. The freezing space is fully utilized to complete the fastest cold start, and the algorithm stops. If the stopping condition is not met, the particle corresponding to the extreme value of the group is the optimal solution when the number of iterations reaches the upper limit.

[0010] The beneficial effects of the present invention are: 1. The present invention optimizes the initial membrane moisture content during cold start, and optimizes the corresponding optimal initial membrane moisture content value for different ambient temperatures. Under the premise of ensuring a successful cold start, the fuel cell stack works at the optimal working point, thereby improving the rapidity of the cold start system.

[0011] 2. The present invention adopts a closed-loop control method of model predictive control. During the cold start process, the stack current output is adjusted in real time according to different ambient temperatures and membrane moisture content, and the controller parameters are adaptively adjusted. This can effectively balance the freezing rate and heating rate of the stack water production, and improve the performance of the cold start system. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a temperature control quantity curve under different environments; Figure 2 It is the temperature output curve of the battery stack at different ambient temperatures; Figure 3 It is the output curve of ice volume fraction at different ambient temperatures; Figure 4 It is the flow chart of particle swarm algorithm; Figure 5 It is the temperature curve of the battery stack with different initial water contents at an ambient temperature of -20℃; Figure 6 It is a curve of ice volume fraction with different initial water contents at -20℃ ambient temperature; Figure 7 It is a curve diagram of different initial water content control amount at -20℃ ambient temperature; Figure 8 It is the ambient temperature-optimal initial membrane moisture content curve. DETAILED DESCRIPTION

[0013] The present invention can be structurally divided into the following parts: cold start model establishment, parameter self-tuning model predictive controller design, and optimal initial membrane water content optimization.

[0014] The main function of the fuel cell cold start controlled object model is to simulate the structure of the external gas path and internal gas diffusion layer, catalyst layer, and proton exchange membrane of the real fuel cell, as well as the physical state change and water transfer process of solid, liquid, and gaseous three-phase water. The controlled object model can accurately describe the stack temperature and the transmission and phase change process of water inside the stack during the cold start of the fuel cell to verify the effectiveness of the designed controller and optimization algorithm; the control model is oriented to describe the trend of state quantity changes during the cold start process in the form of a reduced-dimensional second-order simple model, which is the basis of controller design. The model predictive controller uses a closed-loop control strategy to control the stack current, and adds the ambient temperature term to the penalty term coefficient of the cost function to achieve controller parameter self-tuning and improve the cold start speed. Finally, the cold start model and cold start controller are taken as a whole, as the optimization object of the particle swarm algorithm, and the fitness function is constructed with the cold start rapidity as the goal to achieve the optimization of the initial membrane water content.

[0015] The implementation method of the present invention includes the following parts: The controller design and initial membrane water content optimization of the fuel cell cold start system of the present invention are realized by a software system. The software system is composed of Matlab / Simulink and AVL / Cruise M commercial simulation software. AVL / Cruise M software is used to build a dynamic model of the controlled object of the fuel cell cold start system, and Matlab / Simulink software is used to build a controller and implement an optimization algorithm. The joint simulation of the two provides a simulation experiment environment.

[0016] Firstly, a fuel cell cold start controlled object model was built in AVL / Cruise M software. The actual fuel cell stack parameters of a large 250-cell stack were added to the fuel cell module. The fuel cell cathode and anode gas inlet and outlet modules and the simulated load module were set.

[0017] Then, a control-oriented model is established in Matlab / Simulink software, and a model predictive controller is designed based on the control-oriented model. The penalty term coefficient in the controller cost function is adjusted to the ambient temperature related coefficient, so that the controller can perform parameter self-tuning according to different ambient temperatures and strike a balance between cold start time and cold start safety.

[0018] Finally, the above cold start model and cold start controller are integrated into a cold start control system. The whole is taken as the object to be optimized. The particle swarm algorithm is used to optimize the initial membrane moisture content of the fuel cell stack at different ambient temperatures to obtain the optimal initial membrane moisture content.

[0019] The present invention is further described in detail below with reference to the accompanying drawings: The measurable interference quantity in the present invention is the ambient temperature T amb , used to simulate the environmental variables during the actual fuel cell cold start process. The fuel cell cold start controlled object model is built in AVL / Cruise M as the controlled object of the present invention. The optimization goal of the present invention is to amb , design a model predictive controller to control the stack current, and then optimize the cold start controlled object model and the controller overall control system to optimize the optimal initial membrane water content at the current temperature The cold start stack current output corresponding to the optimal initial membrane moisture content is selected as the optimal cold start current curve Enable the fuel cell stack to achieve a cold start and minimize the cold start time.

[0020] The present invention provides a device based on the above operating principle and operating process. The construction and operation process are as follows: 1. Cold start model establishment In order to ensure the authenticity and effectiveness of the subsequent controller effect verification, this patent builds a fuel cell cold start controlled object model in the commercial simulation software AVL / Cruise M to simulate the actual vehicle-mounted fuel cell cold start process. In the Cruise M software, the integrated proton exchange membrane fuel cell module is used to refer to the actual stack parameters: proton exchange membrane thickness, catalyst layer thickness, and gas flow channel length, width, depth, etc., to generate a fuel cell stack simulation module. Then add the fuel cell cathode anode gas inlet and outlet modules and the simulated stack load module to control the stack current output characteristics. The structural parameters used in the above modules have been verified in the vehicle bench and standard tests of individual components. Therefore, the accuracy of the model is close to the real vehicle, and the credibility can be guaranteed.

[0021] 2. Control-oriented model The control-oriented model was established in Matlab / Simulink. Because the cold start process is a process of balancing the freezing rate and heating rate of the stack water, if the stack temperature rises above zero before the catalyst layer is completely frozen, the cold start is successful, otherwise it fails. Therefore, the state quantity is selected as the stack ice volume fraction φ ice and stack temperature T cell , build the second-order control-oriented model as follows: The left side of the equation is the change rate of the ice volume fraction of the stack The right side of the equation is the source term S for the conversion of frozen water to solid water. f-i , solid water to unfrozen water conversion source term S i-n , solid water and liquid water phase transition source term S l-i , and the source term S of the condensation process of gaseous water into solid water v-i , ρ ice is the density of solid water ice, and ε is the porosity of the catalyst layer. The left side of the equation in the second formula is the temperature change rate of the battery stack. The right side of the equation is Q rea The heat generated by the catalytic reaction, Q amb To dissipate heat in low temperature environment, C cell is the specific heat capacity of the battery stack, M cell is the mass of the battery stack, C ice is the specific heat of ice, M ice is the mass of ice inside the stack.

[0022] 3. Design of parameter self-tuning model predictive controller Based on the above control-oriented model, the state variables are selected as ice volume fraction and stack temperature, and the controller control variable is selected as stack current. The design model predicts the controller cost function as follows: Among them I cell(·|k)=[I cell (k|k),I cell (k+1|k)...I cell (k+N p -1|k)] is the optimized control sequence, and the first item I will be selected later. cell (k|k) acts as the control quantity on the controlled object model. The cost function J consists of three sub-functions J1, J2, and J3: The first part J1 = μ1φ ice (k+i|k) is a constraint on the ice volume fraction of the catalyst layer. By constraining the growth rate of the ice volume fraction, the safety of the cold start process is guaranteed and cold start failure caused by the catalyst layer being completely covered by ice is prevented. Part II It is a constraint on the heating rate of the battery stack. Improving the heating rate of the battery stack can speed up the cold start speed. The third part J3 = μ3ΔI cell (k+i|k) is a constraint on the rate of change of the stack current. Since the stack current is an important reference standard for the stack gas supply system, it is necessary to constrain the stack current to ensure the stable operation of the gas supply system actuator. p To predict the time domain step length, the control quantity is selected so that 1 to N p The sum of the cost functions over time Minimum, thus achieving the best control effect.

[0023] In the above cost function, μ=[μ1,(μ 21 +μ 22 T amb ), μ3] are weight coefficients of each penalty item in the cost function, among which the second penalty system realizes self-tuning of the cost function weight under different ambient temperatures by introducing the ambient temperature: when the ambient temperature is high, the weight of the second penalty item increases, the overall controller effect tends to be rapid heating, the constraint on the ice volume fraction is weakened, and the rapidity of cold start can be improved; when the ambient temperature is low, the weight of the second penalty item decreases, the constraint on the ice volume fraction is strengthened, and the controller effect tends to be safe for the cold start system.

[0024] By adding constraints on the state quantity and the control quantity on the basis of the cost function, the performance function of the cold start system controller is obtained as follows:

[0025] To verify the effectiveness of the controller, the initial membrane water content of the stack is fixed. ini Under the premise of 5, simulation verification is carried out at three different ambient temperatures of -10℃, -15℃ and -20℃, and the control quantity curves at different ambient temperatures are obtained as follows Figure 1 The control effect results are shown as Figure 2 Stack temperature output curve and Figure 3The ice volume fraction output curve is shown. It can be seen from the figure that the controller can achieve a successful cold start within 50 seconds at different ambient temperatures, and the ice volume fraction does not exceed 0.6 during the startup process and remains at a safe level. In addition, affected by the weight of the self-tuning cost function, the controller adopts a more aggressive strategy at a temperature of -10℃, using a larger cold start stack current, which makes the stack warm up faster but the ice volume fraction increases faster; at a temperature of -20℃, the strategy is more conservative, and the expected goal is well achieved, proving that the self-tuning parameter model predicts the controller to be more effective.

[0026] 4. Optimization of the optimal initial membrane moisture content The initial membrane moisture content of the stack is a parameter that plays a decisive role in the cold start process. In the process of verifying the effect of the controller in the previous section, the cold start process was carried out under the premise of fixing the initial membrane moisture content to 5. The maximum ice volume was 0.6, and there was still sufficient ice space, that is, there was still a lot of room for improvement in the rapidity of the cold start system. Therefore, this section uses the particle swarm optimization algorithm to optimize the initial membrane moisture content. The particle swarm algorithm flow chart is shown in the figure. Figure 4 shown.

[0027] The following is a detailed description of the algorithm execution process: The first step is to define the fitness function of the particle swarm algorithm. The optimization goal of this patent is to minimize the cold start time and improve the rapidity of the cold start system under the premise of successful cold start. Fully utilizing the icable volume of the catalyst layer is an effective way to improve the rapidity of cold start. Therefore, the fitness function is defined as follows: This means that when the cold start is over, the closer the ice volume fraction of the catalyst layer is to 0.9, it means that the cold start process has fully utilized the ice volume of the catalyst layer, and the cold start speed has been maximized while ensuring the safety of the stack. The fitness function consists of two parts. The first part is a Gaussian function centered at 0.9 to ensure the overall trend of the fitness function curve; Part II It is a fourth-order polynomial with 0.9 as the vertex, which strengthens the decline speed of the fitness function when it is far away from the target value of 0.9 and accelerates the convergence speed of the particle swarm algorithm. Where α1, α2, σ1, σ2 are the weight coefficients of each item in the fitness function, and the weight distribution of the two items in the fitness function can be further adjusted.

[0028] The second step is to define the boundary conditions of the variables to be optimized. Considering that too low membrane water content will damage the proton exchange membrane of the stack, the minimum value of the initial membrane water content of the stack is set to λ min =2.7, and the water content of the stack saturated membrane is λ sat =15, so the range of water content in the stack membrane is λ ini ∈[λ min,λ sat ], and the subsequent optimization process is completed within this range. Since the high-order polynomial terms are added in the design of the fitness function, the particle convergence speed can be effectively accelerated. Therefore, the number of particles can be set to N = 50, and the upper limit of the number of iterations is T = 30. This setting can effectively reduce the time complexity of the algorithm and improve the speed of the optimization algorithm.

[0029] The third step is to start the cycle optimization process. When the fitness function value of a particle reaches f i,k =α1+α2, it means that particle i has reached the optimal initial membrane moisture content value at the current ambient temperature after iteration k, and the particle swarm optimization algorithm stops. The algorithm stops after fully utilizing the ice space. If the stop condition is not met, the particle corresponding to the extreme value of the group is the optimal solution when the number of iterations reaches the upper limit.

[0030] Based on the above particle swarm optimization algorithm, the initial membrane moisture content was optimized at an ambient temperature of -20°C. The optimal initial membrane moisture content was 5.9, and the corresponding shortest cold start time was 42.5s. Near the optimal initial membrane moisture content, the state quantity curves corresponding to different membrane moisture contents are as follows: Figure 5 Stack temperature curve and Figure 6 The ice volume fraction curve is shown in Figure 2, and the control volume curve is shown in Figure 2. Figure 7 As shown. It can be seen that at an ambient temperature of -20℃: when the initial water content is the minimum value λ ini,min =3, the stack current is always maintained at a low level, resulting in a stack cold start time of 51.7 seconds and a maximum ice volume fraction of only 0.59, indicating that the ice space is not fully utilized, resulting in a slow cold start. ini =7.1, the stack current is always large, but at 32 seconds after startup, the ice volume fraction increases to 0.97, resulting in cold start failure. Optimal initial water content Under the condition of , the cold start time of the stack was 42.5 seconds, and the startup was successfully completed. At this time, the maximum ice volume fraction was 0.905, which fully utilized the ice space, thereby improving the efficiency of the cold start process and proving the good performance of the optimization algorithm.

[0031] For different ambient temperatures between -15℃ and -30℃, the above method was used to optimize the initial water content, and some results were shown in Table 1.

[0032] Table 1 Optimization results of optimal initial moisture content

[0033] It can be seen that under different ambient temperatures, the maximum ice volume fraction during the cold start process can be controlled at around 0.9, which fully utilizes the iceable space of the catalyst layer, speeds up the cold start of the stack, and also ensures the safety of the cold start process, avoiding irreversible damage to the stack components caused by complete freezing. The initial membrane water content was optimized at intervals of 0.1°C between -15°C and -30°C, and the results were Figure 8 The ambient temperature-optimal initial membrane moisture content curve can be used as a reference standard for setting the initial cold start conditions of the fuel cell in a low-temperature environment.

Claims

1. A method for rapid cold start of a fuel cell taking into account the optimization of initial water content, characterized in that: The steps are: S1. Establish a control-oriented model for the fuel cell cold start system: The state variable is selected as the stack ice volume fraction φ ice and stack temperature T cell , build the second-order control-oriented model as follows: Where: is the change rate of ice volume fraction in the stack, S f-i is the source term for the conversion of frozen water to solid water, S i-n is the source term for the conversion of solid water to unfrozen water, S l-i is the phase transition source term between solid water and liquid water, S v-i is the source term of the condensation process of gaseous water into solid water, ρ ice is the density of solid water ice, ε is the porosity of the catalyst layer, is the temperature change rate of the battery stack, Q rea The heat generated by the catalytic reaction, Q amb To dissipate heat in low temperature environment, C cell is the specific heat capacity of the battery stack, M cell is the mass of the battery stack, C ice is the specific heat of ice, M ice is the mass of ice inside the stack; S2. Based on the control-oriented model, the model predictive stack current controller with self-tuning parameters is designed: Select the state variables as ice volume fraction and stack temperature, select the controller control variable as stack current, and design the model to predict the controller cost function as follows: Among them I cell (·|k)=[I cell (k|k),I cell (k+1|k)...I cell (k+N p -1|k)] is the optimized control sequence; the cost function J is composed of three sub-functions J1, J2, and J3: J1 = μ1φ ice (k+i|k), J3=μ3ΔI cell (k+i|k); N p is the prediction time domain step; μ=[μ1,(μ 21 +μ 22 T amb ),μ3] is the weight coefficient of each penalty term in the cost function; By adding constraints on state quantities and control quantities, the performance function of the cold start system controller is obtained as follows: S3. Optimization of initial membrane water content of the stack: S31. Define the fitness function of the particle swarm algorithm The fitness function is defined as follows: In the formula A Gaussian function centered at is a fourth-order polynomial of vertices, α1, α2, σ1, σ2 are weight coefficients of each item in the fitness function; S32. Define the boundary conditions of the variables to be optimized Set the minimum initial membrane water content of the stack λ min and the stack saturated membrane water content λ sat , so the water content range of the stack membrane is λ ini ∈[λ min ,λ sat ], and set the number of particles N and the upper limit of the number of iterations T; S33, start loop optimization When the fitness function value of a particle reaches f i,k =α1+α2, it means that particle i has reached the optimal initial membrane moisture content under the current ambient temperature after iteration k, and the stopping condition of the particle swarm optimization algorithm is met. The freezing space is fully utilized to complete the fastest cold start, and the algorithm stops. If the stopping condition is not met, the particle corresponding to the extreme value of the group is the optimal solution when the number of iterations reaches the upper limit.

Images