Fuel cell system variable scale performance prediction method combined with neural network

Abstract

The invention discloses a fuel cell system variable-scale performance prediction method combined with a neural network, and the method comprises the steps: constructing a proton exchange membrane fuel cell model combined with a neural network algorithm, and achieving the variable-scale performance prediction of a proton exchange membrane fuel cell system. The proton exchange membrane fuel cell model comprises an air supply module, a hydrogen supply module, a hydrothermal management module, a cell stack module and a neural network fitting module; calculating the actual output power of the fuel cell system; after all the modules are combined and connected, a proton exchange membrane fuel cell system variable-scale numerical simulation model combined with a neural network algorithm is formed. According to the method, the advantages of the neural network and mathematical modeling are combined, limited data resources, computer resources and time cost are used, high-precision simulation modeling of the proton exchange membrane fuel cell is completed, and therefore guidance is provided for application of the proton exchange membrane fuel cell in the aviation field.

Description

Technical Field

[0001] The invention belongs to the field of aviation and relates to a proton exchange membrane fuel cell, in particular to a method for predicting the variable-scale performance of a proton exchange membrane fuel cell system combined with a neural network algorithm. Background Art

[0002] Since the beginning of this century, in order to reduce carbon emissions worldwide, the aviation industry has begun to increasingly pursue clean energy, and electrification and pure electric aircraft have become one of the current mainstream trends. Under the existing technical framework, proton exchange membrane fuel cells have become one of the most likely aircraft electrification technologies to be realized in the short term. However, in the face of the strict safety requirements of flight vehicles, proton exchange membrane fuel cells must undergo a complex and rigorous safety assessment process before they can be used as part of airborne equipment. To achieve this goal, it is particularly critical to establish an accurate proton exchange membrane fuel cell model.

[0003] The core purpose of modeling is to accurately and quickly simulate the performance response of fuel cells under environmental factors. Although finite element modeling can provide accurate and comprehensive result output, it often requires high computer resources and a lot of time cost, and is usually not considered the main way to model the system; on the other hand, since the overall research on the aviation application of proton exchange membrane fuel cells is still in its early stages, it is difficult for researchers to obtain a large amount of data to support the construction of a purely data-driven model; at the same time, a single mathematical model often cannot accurately describe all the external conditions that proton exchange membrane fuel cells may encounter during the flight phase. Therefore, it is particularly important to use limited resources to build an efficient, accurate and comprehensive proton exchange membrane fuel cell simulation model, so as to provide guidance for the application of proton exchange membrane fuel cells in the aviation field. Summary of the invention

[0004] The purpose of the present invention is to provide a method for predicting the variable-scale performance of a fuel cell system in combination with a neural network, combining the advantages of neural networks and mathematical modeling, using a limited amount of data resources, computer resources and time costs, to complete high-precision simulation modeling of proton exchange membrane fuel cells.

[0005] The purpose of the present invention is achieved through the following technical solutions:

[0006] A method for predicting the variable-scale performance of a fuel cell system combined with a neural network is characterized in that a proton exchange membrane fuel cell model combined with a neural network algorithm is constructed to realize the variable-scale performance prediction of the proton exchange membrane fuel cell system. The proton exchange membrane fuel cell model includes an air supply module, a hydrogen supply module, a water and heat management module, a battery stack module and a neural network fitting module. The specific steps are as follows:

[0007] Step 1: Air supply module, including air compressor, flow control system and humidifier.

[0008] The air compressor module is defined as a gas property control module, which is used to set a certain pressure and flow signal for the air in the pipeline. The flow rate at a certain target pressure is determined by the flow control system:

[0009] rpm=k1·OER·IM act (1)

[0010] M set =k2·M des (2)

[0011] M des =f(rpm,π,p,T) (3)

[0012] Where rpm is the speed, OER is the air excess coefficient, I is the stack current, M act is the current flow rate, M set is the set flow rate, k1 and k2 are the speed correction coefficient and flow correction coefficient, respectively, which are related to the number of single batteries and the current temperature and pressure. des It is determined by the properties of the compressor used and is a function of rpm, compression ratio π, temperature T, and pressure p.

[0013] The humidifier module is used to change the proportion of water in the air pipeline. The method is: compare the relative humidity of the air before and after the module, add the mass flow of water, and simulate the humidification process of the air.

[0014] M water_in =k3·(RH out -RH in ) (4)

[0015] Where k3 is the humidification correction factor and RH is the relative humidity.

[0016] Step 2: Hydrogen supply module, including hydrogen tank, pressure reducing valve, and hydrogen recirculation device.

[0017] A hydrogen tank is defined as an insulated gas cavity that satisfies the laws of conservation of energy and mass. It has a material and energy exchange interface with the outside world and provides a hydrogen signal to the hydrogen pipeline.

[0018] Mass conservation within the mesh:

[0019]

[0020] Conservation of Energy:

[0021]

[0022] Exchange of substances with the outside world:

[0023]

[0024] Energy exchange with the outside world:

[0025] Q Ports =Φ A (8)

[0026] Where V0 is the volume of the cavity, ρ is the density, p is the pressure, t is the time, T is the temperature, x i is the mass fraction, M is the mass fraction, h is the enthalpy, c p is the isobaric specific heat capacity, Q is the heat, Φ A is the internal heat source, the subscript Ports is the interface, and Cond is the interior of the cavity.

[0027] The pressure reducing valve is defined as a valve that can limit the outlet area. The hydrogen pressure signal at the outlet is controlled by setting the throttling area, that is:

[0028] A min =f(p act ,p set ) (9)

[0029] Among them, p act is the actual pressure, p set is to set the pressure.

[0030] The hydrogen recirculation device extracts a portion of the gas flow from the exhaust gas according to the stack current, and enters the humidifier module together with the hydrogen provided by the hydrogen tank.

[0031] M H2_rec =f(I) (10)

[0032] The hydrogen humidifier is consistent with the oxygen humidifier, that is,

[0033] M water_in =k3·(RH out -RH in ) (11)

[0034] Step 3: Water thermal management module, which consists of a cooling box, a radiator, a heat exchanger, and a water pump.

[0035] The cooling box is used to transmit the property signal of the coolant into the coolant pipeline;

[0036] The radiator is set as a heat exchange device. The coolant and the external medium are located on both sides of the radiator respectively, and the heat flow signal is transferred between the coolant and the other medium through the radiator. The heat exchange process includes heat convection and heat conduction.

[0037] Heat convection:

[0038]

[0039] Heat conduction:

[0040]

[0041] Among them, Re is the Reynolds number, area is the heat transfer cross-sectional area, and D h is the hydraulic diameter, Pr is the Planck number, k is the thermal conductivity, and ΔT is the temperature difference between the coolant and the radiator wall.

[0042] Step 4: Battery stack module, including flow channel assembly and membrane electrode assembly.

[0043] The flow channel assembly is a cavity for a gas mixture, which satisfies the energy conservation and mass conservation as well as the condensation change of the gas. There are four interfaces connected to the outside world, which transmit the mass fraction signal of the internal material, the mass fraction signal of the outflowing material, the temperature signal of the outflowing material to the membrane electrode assembly, and transmit the heat flow signal to the cooling system.

[0044] The energy conservation and mass conservation equations are similar to those of hydrogen tanks, and the condensation behavior control equation is:

[0045]

[0046] x H2O is the mass fraction of water, x sat is the mass fraction of water at saturation, and tau_c is the condensation time constant of water.

[0047] The membrane electrode assembly is used to realize the electrochemical reaction process of hydrogen and oxygen, including voltage output, electrochemical heat and water transfer.

[0048] The overall electrochemical reaction can be expressed as:

[0049]

[0050] The stack output voltage is equal to the theoretical voltage minus the actual voltage loss, which includes activation loss, ohmic loss and concentration loss. The actual output voltage can be calculated by this formula:

[0051] V stack =V nernst -V act -V ohmic -V conc (16)

[0052] In formula (16), the theoretical voltage V nernst for:

[0053]

[0054] Where Tst is the stack temperature, is the partial pressure of the corresponding substance;

[0055] In formula (16), the activation loss V act for:

[0056]

[0057] α is the charge transfer coefficient, i cell is the stack current, i0 is the exchange current density;

[0058] In formula (16), the ohmic loss V ohmic for:

[0059] V ohmic =i cell ·Rohm(19)

[0060] Where Rohm is the resistance, and its value is T mem / σ, σ is the conductivity of the membrane, which is related to the stack temperature T st , relative humidity RH, and the water content λ of the membrane;

[0061] In formula (16), the concentration loss V conc for:

[0062]

[0063] where i L is the limiting current density.

[0064] The energy represented by electrochemical heat can be expressed by the difference between the theoretical voltage and the actual voltage, that is,

[0065] Heat=(V theorv -V stack )·i cell ·area_cell (21)

[0066] Where area_cell is the single cell reaction area, V stack Calculated by formula (16), V theorv The calculation of takes into account the high calorific value of hydrogen and the heat of evaporation of water, that is,

[0067]

[0068] Where N cell is the number of single batteries, is the higher calorific value of hydrogen, 286 kJ / mol, h w0 is the heat of vaporization of water, which is determined by temperature and pressure. is the molar mass of water, which is 18.015 kilograms per mole (kg / mol).

[0069] Water transfer Water Including water diffusion drag and electroosmotic drag of water drag ,Right now

[0070] n Water =n drag -n diff (twenty three)

[0071] The specific calculation formula is:

[0072]

[0073] Where D H2O is the diffusion coefficient of water, C ccl , C acl are the concentrations of water at the cathode and anode, respectively, which can be calculated by the following formulas:

[0074]

[0075] where ρ mem is the density of the membrane, M mem is the molar mass of the membrane and λ is the water content of the membrane.

[0076] The electroosmotic drag of water can be calculated as follows:

[0077] n drag =0.0029·λ mem 2 +0.05·λ mem (28)

[0078] Step 5: Neural Network Fitting Module

[0079] The neural network fitting system is defined as a data-driven model. It is used to solve the problem that the mathematical model cannot quickly simulate individual complex working conditions. The current current signal and voltage signal are obtained from the fuel cell module, and then transmitted back to the fuel cell module after being corrected by the neural network fitting result. The method is: Use the neural network fitting to establish a function U der =f(I,Φ1,Φ2,...), where Φ represents external influencing factors, such as (tilt, vibration, impact, etc.).

[0080] Based on experimental data, a function of the degree of influence of external influencing factors on voltage at different current densities that are difficult to simulate is established, thereby achieving the ability of rapid response.

[0081] Set I, Φ1, Φ2, ... as the "response" in the neural network fitting, set U as the "predictor variable", determine the percentage of training data Tr and the number of training layers N. After the training is completed, the fitting function U = f (I, Φ1, Φ2, ...) can be obtained. Its quotient with U0 = f (I, 0, 0, ...) is U der :

[0082]

[0083] Step 6: Calculate the actual output power of the fuel cell system

[0084] The main power-generating component in the fuel cell system is the stack, and the power-consuming components are the compressor and the water pump. Therefore, the actual output power of the system is equal to the stack power minus the component power plus the effect on the tilt angle. It can be calculated by the following formula:

[0085] P act =U der ·U theo ·IW com -W pum (30)

[0086] Among them U theo is the theoretical stack voltage without considering external factors, W com is the compressor power consumption, W pum The water pump consumes power.

[0087] Step 7: After all modules are combined and connected, a variable-scale numerical simulation model of a proton exchange membrane fuel cell system combined with a neural network algorithm can be constructed.

[0088] The beneficial effects of the present invention are:

[0089] The present invention combines the advantages of neural networks and mathematical modeling, uses limited data resources, computer resources and time costs, and completes high-precision simulation modeling of proton exchange membrane fuel cells, thereby providing guidance for the application of proton exchange membrane fuel cells in the aviation field. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] Figure 1 It is a schematic diagram of the variable scale modeling principle of a proton exchange membrane fuel cell system combined with a neural network algorithm. DETAILED DESCRIPTION

[0091] A method for predicting the variable-scale performance of a proton exchange membrane fuel cell system combined with a neural network algorithm is proposed. A proton exchange membrane fuel cell model combined with a neural network algorithm is constructed to realize the variable-scale performance prediction of a proton exchange membrane fuel cell system. The proton exchange membrane fuel cell model includes an air supply module, a hydrogen supply module, a water and heat management module, a battery stack module, and a neural network fitting module. The effects of the tilt angle, ambient temperature, humidity, and pressure on the system performance are considered. Figure 1 This is a schematic diagram of the variable scale modeling principle of a proton exchange membrane fuel cell system combined with a neural network algorithm. The method of building a fuel cell system model for coupling neural network fitting is as follows:

[0092] Step 1: Air Supply Module.

[0093] The air compressor module is defined as a gas property control module, which is used to set a certain pressure and flow signal for the air in the pipeline. The flow rate at a certain target pressure is determined by the flow control system, which controls the flow rate by adjusting the speed of the compressor.

[0094] According to Faraday's law, the theoretical air flow required for the current can be calculated:

[0095]

[0096] Where OER is the excess air factor, N cell is the number of single batteries, M set is the set flow rate, and F is the Faraday constant.

[0097] The difference between the theoretical air flow and the actual air flow is controlled by PID with a ratio of 5 and an integral of 0.5, and then multiplied by the maximum speed of the compressor, 3600, to obtain the current required speed:

[0098]

[0099] Among them, rpm is the speed, M act is the current flow.

[0100] In the air compressor characteristic diagram, the compressor output flow can be obtained according to the compression ratio, speed, temperature and pressure:

[0101] M des =f(rpm,π,p,T) (33)

[0102] A humidifier module is built to change the proportion of water in the air pipeline. The method is: compare the relative humidity of the air before and after the module, add the mass flow of water, and simulate the humidification process of the air.

[0103]

[0104] Among them, M water_in is the mass flow rate of water added and RH is the relative humidity of the water.

[0105] Step 2: Hydrogen supply module.

[0106] The definition of a hydrogen tank is established as an insulated gas cavity, which satisfies the conservation of energy and mass inside the cavity, has a material and energy exchange interface with the outside world, and provides hydrogen signals to the hydrogen pipeline.

[0107] Conservation of mass in the cavity:

[0108]

[0109] Conservation of Energy:

[0110]

[0111] Exchange of substances with the outside world:.

[0112]

[0113] Energy exchange with the outside world:

[0114] Q Ports =Φ A (38)

[0115] Where V0 is the volume of the cavity, ρ is the density, p is the pressure, t is the time, T is the temperature, x i is the mass fraction, M is the mass fraction, h is the enthalpy, c p is the isobaric specific heat capacity, Q is the heat, Φ A is the internal heat source, the subscript Ports is the interface, and the subscript Cond is the interior of the cavity.

[0116] The pressure reducing valve is defined as a valve that can limit the outlet area. The hydrogen pressure signal at the outlet is controlled by setting the throttling area, that is,

[0117]

[0118] Among them, p act is the actual pressure, p set is the set pressure, r min and r max are the minimum and maximum throttling areas, p range is the maximum pressure change percentage, and D is the average channel diameter.

[0119] At this time, the pressure reducing valve outlet pressure can be calculated as:

[0120]

[0121] Where M is the mass flow rate, rat area is the ratio of the throttle area to the port area, rat ρ is the ratio of the density at the throttling point to the density at the port, ρ A is the density at the throttling point.

[0122] A hydrogen recirculation device is established to recover hydrogen from the anode exhaust according to the stack current and enter the humidifier module together with the hydrogen provided by the hydrogen tank.

[0123]

[0124] Among them, I is the stack current, and area_cell is the single cell reaction area.

[0125] The hydrogen humidifier is consistent with the oxygen humidifier, that is,

[0126]

[0127] Step 3: Water Heat Management Module

[0128] A cooling box is established to transmit the property signal of the coolant into the coolant pipeline;

[0129] The radiator is set as a heat exchange device. The coolant and the external medium are located on both sides of the radiator, and the heat flow signal is transferred between the coolant and the other medium through the radiator. The heat exchange process includes heat convection and heat conduction.

[0130] Heat convection:

[0131]

[0132] Where Re is the Reynolds number and Pr is the Planck number, which can be calculated by the following formula:

[0133]

[0134] Pr=μ·c p ·k (45)

[0135] Where, area is the flow cross-sectional area, D h is the hydraulic diameter, k is the thermal conductivity, μ is the dynamic viscosity, c p is the isobaric specific heat capacity, and ΔT is the temperature difference between the coolant and the radiator wall. ΔT can be calculated by the following formula:

[0136] ΔT=(T wall -T l )×(1-e -NTU ) (46)

[0137] Where T wall is the wall temperature, T lis the fluid temperature, and NTU is the number of heat transfer units, which is calculated by the following formula:

[0138]

[0139] Where surface is the heat transfer area and Nu is the Nusselt number, which can be calculated by the following formula:

[0140]

[0141] Where f is the friction coefficient, which is determined by the properties of the fluid and the wall.

[0142] Heat conduction:

[0143]

[0144] Step 4: Battery stack module.

[0145] The flow channel component is established as a cavity of a gas mixture, and the energy conservation and mass conservation as well as the condensation change of the gas are satisfied in the cavity. There are four interfaces connected to the outside world, which transmit the mass fraction signal of the internal material, the mass fraction signal of the outflowing material, the temperature signal of the outflowing material to the membrane electrode assembly, and transmit the heat flow signal to the cooling system.

[0146] The energy conservation and mass conservation equations are similar to those of hydrogen tanks, and the condensation behavior control equation is:

[0147]

[0148] Among them, x H2O is the mass fraction of water, x sat is the mass fraction of water at saturation, which is determined by the temperature and pressure at that time. In the model, it is obtained by looking up the saturated water vapor temperature and pressure table. tau_c is the condensation time constant of water, which is 1 in this model.

[0149] The membrane electrode assembly is used to realize the electrochemical reaction process of hydrogen and oxygen, including voltage output, electrochemical heat and water transfer.

[0150] The overall electrochemical reaction can be expressed as:

[0151]

[0152] Its output voltage is equal to the theoretical voltage minus the actual voltage loss, which includes activation loss, ohmic loss and concentration loss. The actual stack output voltage can be calculated by this formula:

[0153] V stack =V nernst -V act -V ohmic -V conc (52)

[0154] The theoretical voltage V in formula (52) nernst for:

[0155]

[0156] Where T st is the stack temperature, is the partial pressure of the corresponding substance;

[0157] The activation loss V in formula (52) act for:

[0158]

[0159] Where α is the charge transfer coefficient, i cell is the stack current, i0 is the exchange current density;

[0160] The ohmic loss V in formula (52) ohmic for:,

[0161] V ohmic =i cell ·Rohm (55)

[0162] Where Rohm is the resistance, and its value is T mem / σ, σ is the conductivity of the membrane, which is related to the stack temperature T st , relative humidity RH, and the water content λ of the membrane;

[0163] The concentration loss V in formula (52) conc for:

[0164]

[0165] where i L is the limiting current density.

[0166] The energy represented by electrochemical heat can be calculated from the difference between the theoretical voltage and the actual voltage, that is,

[0167] Heat=(V theorv -V stack )·I·area_cell (57)

[0168] Where area_cell is the single cell reaction area, V stack According to the above formula, V theorv The calculation of takes into account the high calorific value of hydrogen and the heat of evaporation of water, that is,

[0169]

[0170] Where N cell is the number of single batteries, is the higher calorific value of hydrogen, which is 286 kJ / mol, h w0 is the heat of vaporization of water, which is determined by temperature and pressure. is the molar mass of water, which is 18.015 kilograms per mole (kg / mol).

[0171] Water transfer Water Including water diffusion drag and electroosmotic drag of water drag ,Right now

[0172] n Water =n drag -n diff (59)

[0173] The diffusion of water can be calculated by the following formula:

[0174]

[0175] Where D H2O is the diffusion coefficient of water, C ccl , C acl The calculation methods for the concentration of water at the cathode and anode are:

[0176]

[0177]

[0178] where ρ mem is the density of the membrane, M mem is the molar mass of the membrane and λ is the water content.

[0179] The electroosmotic drag of water can be calculated as follows:

[0180] n drag =0.0029·λ mem 2 +0.05·λ mem (64)

[0181] The calculation method of water content is:

[0182]

[0183] Step 5: Neural Network Fitting Module

[0184] Define the neural network fitting system as a data-driven model. Use neural network fitting to establish a function U der =f(I, Til1, Til2, ...), where Til represents different tilt angles.

[0185] Based on the experimental data, the voltage data of the fuel cell system at different tilt angles and different currents are obtained.

[0186] Set I, Til1, Til2, ... as the "response" in the neural network fitting, set U as the "predictor variable", determine the percentage of training data Tr = 70% and the number of training layers N = 20.

[0187] After the training is completed, the neural fitting function U = f(I, Til1, Til2, ...) can be obtained. Its quotient with U0 = f(I, 0, 0, ...) is U der ,Right now:

[0188]

[0189] The model loading mode is current loading, so after considering the influence of the tilt angle, the actual stack output voltage can be calculated as:

[0190] U act =U der ·U theo (67)

[0191] Among them U act is the actual output voltage of the battery stack, U thoe is the theoretical output voltage of the battery stack without considering the tilt angle.

[0192] Step 6: Calculate the actual output power of the fuel cell system

[0193] The main power-generating components in the fuel cell system are the stack, and the power-consuming components are the compressor and the water pump. The power of the compressor can be calculated by the following formula:

[0194]

[0195] Among them, W com W is the power consumption of the compressor, which is determined by the pressures p1 and p2 before and after, the initial pressure T1 and the mass flow rate M. pum is the power consumption of the water pump, which is determined by the mass flow rate M, the inlet and outlet pressure p out 、p in and the average density ρ ave Decide.

[0196] Therefore, the actual output power of the system is equal to the stack power minus the component power plus the effect on the tilt angle. It can be calculated by the following formula:

[0197] P act =U der ·U theo ·IW com -W pum (70)

[0198] Step 7: After all components are built, they can be assembled into a variable-scale simulation model of a proton exchange membrane fuel cell system combined with a neural network algorithm.

[0199] The present invention utilizes limited resources to build an efficient, accurate and comprehensive proton exchange membrane fuel cell simulation model, thereby providing guidance for the application of proton exchange membrane fuel cells in the aviation field.

Claims

1. A method for predicting variable-scale performance of a fuel cell system in combination with a neural network, characterized in that: A proton exchange membrane fuel cell model combined with a neural network algorithm is constructed to realize the variable-scale performance prediction of the proton exchange membrane fuel cell system. The proton exchange membrane fuel cell model includes an air supply module, a hydrogen supply module, a water and heat management module, a battery stack module, and a neural network fitting module. The specific steps are as follows: Step 1: The air supply module includes an air compressor, a flow control system and a humidifier; The air compressor is used to set a certain pressure and flow signal for the air in the pipeline. The flow rate at a certain target pressure is determined by the flow control system; The humidifier is used to change the proportion of water in the air pipeline by comparing the relative humidity of the air before and after the humidifier, adding the mass flow of water to simulate the humidification process of the air; Step 2: The hydrogen supply module includes a hydrogen tank, a pressure reducing valve, and a hydrogen recirculation device; The hydrogen tank is an insulated gas cavity that satisfies the conservation of energy and mass. It has a material and energy exchange interface with the outside world and provides hydrogen signals to the hydrogen pipeline. The pressure reducing valve is a valve that can limit the outlet area. It controls the hydrogen pressure signal at the outlet by setting the throttling area. The hydrogen recirculation device extracts a portion of the gas flow from the exhaust gas according to the stack current, and enters the humidifier together with the hydrogen provided by the hydrogen tank. The hydrogen humidifier is consistent with the oxygen humidifier; Step 3: The water thermal management module includes a cooling box, a radiator, and a heat exchanger; The cooling box is used to transmit the property signal of the coolant into the coolant pipeline; The radiator is a heat exchange device. The coolant and the external medium are located on both sides of the radiator. The heat flow signal is transmitted between the coolant and other media through the radiator. The heat exchanger completes the heat exchange process, including heat convection and heat conduction. Step 4: The battery stack module includes a flow channel assembly and a membrane electrode assembly; The flow channel component is a cavity for a gas mixture, which satisfies the energy conservation and mass conservation as well as the condensation change of the gas. It has four interfaces connected to the outside world, which transmit the mass fraction signal of the internal material, the mass fraction signal of the outflowing material, the temperature signal of the outflowing material to the membrane electrode assembly, and transmit the heat flow signal to the cooling system. The membrane electrode assembly is used to realize the electrochemical reaction process of hydrogen and oxygen, including voltage output, electrochemical heat and water transfer; The stack output voltage is equal to the theoretical voltage minus the actual voltage loss. The voltage loss includes activation loss, ohmic loss and concentration loss. The energy represented by electrochemical heat is represented by the difference between the theoretical voltage and the actual voltage. Water Including water diffusion drag and electroosmotic drag of water drag ; Step 5: The neural network fitting module is a data-driven model used to solve the problem that the mathematical model cannot quickly simulate certain complex working conditions. It obtains the current current signal and voltage signal from the fuel cell module and transmits them back to the fuel cell module after being corrected by the neural network fitting result. Step 6: Calculate the actual output power of the fuel cell system; The power-generating component in the fuel cell system is the stack, and the power-consuming components are the compressor and the water pump. The actual output power of the system is equal to the stack power minus the component power, plus the effect on the tilt angle, calculated by the following formula: Q act =U der ·U theo ·IW com -W pum Among them U theo is the theoretical stack voltage without considering external factors, W com is the compressor power consumption, W pum U is the power consumption of the water pump; der It is to use neural network fitting to establish a function; Step 7: After all modules are combined and connected, a variable-scale numerical simulation model of a proton exchange membrane fuel cell system combined with a neural network algorithm is formed.

2. The method for predicting variable-scale performance of a fuel cell system in combination with a neural network according to claim 1, characterized in that: In step 1, the flow rate at a certain target pressure is determined by the flow control system as follows: rpm=k1·OER·I-M act (1) M set =k2·M des (2) M des =f(rpm,π,p,T) (3) Where rpm is the speed, OER is the air excess coefficient, I is the stack current, M act is the current flow rate, M set is the set flow rate, k1 and k2 are the speed correction coefficient and flow correction coefficient, respectively, which are related to the number of single batteries and the current temperature and pressure, M des It is determined by the nature of the compressor used and is a function of rpm, compression ratio π, temperature T, and pressure p; The humidifier simulates the humidification process of the air as follows: I water_in =k3·(RH out -RH in ) (4) Where k3 is the humidification correction factor, (RH out is the outlet relative humidity, RH in ) is the inlet relative humidity.

3. The method for predicting variable-scale performance of a fuel cell system in combination with a neural network according to claim 1, characterized in that: In step 2, Conservation of mass: Conservation of Energy: Exchange of substances with the outside world: Energy exchange with the outside world: Q Ports =Φ A (8) Where V0 is the volume of the cavity, ρ is the density, p is the pressure, t is the time, T is the temperature, x i is the mass fraction, M is the mass fraction, h is the enthalpy, c p is the isobaric specific heat capacity, Q is the heat, Φ A is the internal heat source, the subscript Ports is the interface, and Cond is the interior of the cavity; The pressure reducing valve controls the hydrogen pressure signal at the outlet by setting the throttling area, that is: A min =f(p act ,p set ) (9) Among them, p act is the actual pressure, p set is to set the pressure; The hydrogen recirculation device extracts a portion of the gas flow from the exhaust gas according to the stack current, and enters the humidifier together with the hydrogen provided by the hydrogen tank. The hydrogen humidifier is consistent with the oxygen humidifier, that is, I water_in =k3·(RH out -RH in )。 (11) 4. The method for predicting variable-scale performance of a fuel cell system in combination with a neural network according to claim 1, characterized in that: In step 3, Heat convection: Heat conduction: Among them, Re is the Reynolds number, area is the heat transfer cross-sectional area, and D h is the hydraulic diameter, Pr is the Planck number, k is the thermal conductivity, and ΔT is the temperature difference between the coolant and the radiator wall.

5. The method for predicting variable-scale performance of a fuel cell system in combination with a neural network according to claim 1, characterized in that: In step 4, The energy conservation and mass conservation equations are similar to those of hydrogen tanks, and the condensation behavior control equation is: x H2O is the mass fraction of water, x sat is the mass fraction of water at saturation, tau_c is the condensation time constant of water; The overall electrochemical reaction is expressed as: The actual output voltage of the battery stack is calculated by this formula: V stack =V nernst -V act -V ohmic -V conc (16) Theoretical voltage V nernst for: Where T st is the stack temperature, is the partial pressure of the corresponding substance; Activation loss V act for: α is the charge transfer coefficient, i cell is the stack current, i0 is the exchange current density; Ohm loss V ohmic for: V ohmic =i cell ·Rohm (19) Where Rohm is the resistance, and its value is T mem / σ, σ is the conductivity of the membrane, which is related to the stack temperature T st , relative humidity RH, and the water content λ of the membrane; Concentration loss V conc for: where i L is the limiting current density; The energy represented by electrochemical heat is expressed by the difference between the theoretical voltage and the actual voltage, i.e. Heat=(V theorv -V stack )·i cell ·area_cell (21) Where area_cell is the single cell reaction area, V stack Calculated by formula (16), V theorv The calculation of takes into account the high calorific value of hydrogen and the heat of evaporation of water, that is, Where N cell is the number of single batteries, is the higher calorific value of hydrogen, 286 kJ / mol, h w0 is the heat of vaporization of water, which is determined by temperature and pressure. is the molar mass of water, which is 18.015 kg per mole; Water transfer Water Including water diffusion drag and electroosmotic drag of water drag ,Right now n Water =n drag -n diff (23) The specific calculation formula is: Where D H2O is the diffusion coefficient of water, C ccl , C acl are the concentrations of water at the cathode and anode, respectively, calculated by the following formulas: where ρ mem is the density of the membrane, M mem is the molar mass of the membrane, λ is the water content of the membrane; The electroosmotic drag of water is calculated by the following formula: n drag =0.0029·l mem 2 +0.05 min mem 。 6. The method for predicting variable-scale performance of a fuel cell system in combination with a neural network according to claim 1, characterized in that: In step 5, the method is: Use neural network fitting to build a function U der =f(I,Φ1,Φ2,...), where Φ represents the external influencing factor. Based on experimental data, a function of the influence degree of the external influencing factor on the voltage at different current densities that is difficult to simulate is established, thereby achieving the ability of rapid response; Set I, Φ1, Φ2, ... as the "response" in the neural network fitting, set U as the "predictor variable", determine the percentage of training data Tr and the number of training layers N, and after the training is completed, you can get the fitting function U = f (I, Φ1, Φ2, ...), and its quotient with U0 = f (I, 0, 0, ...) is U der :

Images