Three-valve combined intelligent control method based on water content of fuel cell membrane

Abstract

The present application relates to a kind of three valve combined intelligent control method based on fuel cell membrane moisture content, it relates to fuel cell control technical field, for the unstable operation phenomenon caused by hydrogen pressure fluctuation, flow mismatch and membrane moisture content imbalance of fuel cell, specifically provide a kind of three valve combined intelligent control method based on fuel cell membrane moisture content, method includes: based on artificial neural network model, membrane moisture content is predicted and valve opening is corrected;Current hydrogen pressure fluctuation value in hydrogen discharge process of hydrogen system is obtained, and valve is again fine-tuned using PID control mode, realizes the self-adaptive regulation of system under all operating conditions and decoupling control under multiple outputs, finally achieves the purpose of hydrogen accurate supply and stack stable operation.

Description

Technical Field

[0001] This invention relates to the field of fuel cell control technology, and in particular to a three-valve combined intelligent control method based on the water content of the fuel cell membrane. Background Technology

[0002] Currently, for the hydrogen intake and exhaust process of proton exchange membrane fuel cell systems, hydrogen is generally supplied to the stack through a hydrogen storage tank. When supplying hydrogen to the stack, the high-pressure gas in the hydrogen storage tank must be reduced to a medium-pressure level, and then the pressure is regulated by a proportional valve or hydrogen injector and supplied to the stack. Usually, the hydrogen pressure entering the stack will have large fluctuations. In the face of this situation, the hydrogen system reduces the hydrogen medium-pressure fluctuation value by adjusting the opening of the pressure reducing valve.

[0003] The existing working principle of adjusting the pressure reducing valve opening is based on comparing the hydrogen medium pressure fluctuation value with a threshold. When the fluctuation value is greater than or equal to the threshold, the pressure reducing valve opening is increased. The process continues to collect hydrogen medium pressure fluctuation values ​​for the next time period and compare them with the threshold until the hydrogen medium pressure fluctuation value is less than the threshold. However, this method cannot predict future conditions. For example, when the vehicle is accelerating, the pressure reducing valve opening can only be adjusted after the situation occurs. This will lead to sluggish fuel cell stack power, insufficient acceleration power, and excessive anode pressure fluctuations, which can damage the proton exchange membrane. Summary of the Invention

[0004] In view of the shortcomings of the prior art, the technical problem to be solved by the present invention is to provide a three-valve joint intelligent control method based on the water content of fuel cell membranes. This method solves the problems in the prior art that the inability to predict the vehicle condition in advance when supplying hydrogen to the fuel cell stack leads to insufficient acceleration power and damage to the proton exchange membrane. Based on the predicted information, the method can predict the membrane water content using an artificial neural network model and perform three-valve joint intelligent control. The valve opening is initially adjusted, and after collecting the hydrogen pressure fluctuation value, the valve is finely adjusted again using PID control to more accurately meet the vehicle's operating requirements.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is: a three-valve combined intelligent control method based on the water content of fuel cell membranes. In a first aspect, the present invention provides a formula for calculating the rate of change of fuel cell stack output current, and converts the predicted value into a valve opening target, including:

[0006] The rate of change of the fuel cell stack output current is calculated by using the power supply side parameters;

[0007] In the current fuel cell system, various physical parameters are input based on the mechanistic model to calculate the predicted value of the current change rate.

[0008] The influencing factors include the hydrogen ejector's pressure regulation of hydrogen, the amount of compressed air supplied by the air compressor to the cathode, and the ambient temperature.

[0009] Based on the influencing factors, calculate the current change rate of each influencing factor;

[0010] Calculate the maximum current change rate under the constraint of hydrogen supply to the anode of the fuel cell, the current change rate under the constraint of air compressor at the cathode of the fuel cell, and the current change rate under the constraint of ambient temperature.

[0011] After obtaining the rate of change of current under different influencing factors, the minimum value is taken as the predicted value of the final output current rate of change:

[0012]

[0013] By predicting the output current change rate, the valve opening can be adjusted in advance to regulate the hydrogen pressure value.

[0014] By calculating the rate of change of current under various constraints, the dynamic trend is predicted, the target value is calculated based on the trend, and the target value is used as the model input to obtain the valve opening.

[0015] The target values ​​include target hydrogen flow rate, target anode pressure, target oxygen flow rate, and target cathode pressure.

[0016] Furthermore, in a second aspect, the present invention provides an artificial neural network model for predicting membrane water content and designing a three-valve linkage controller, with the prediction of membrane water content being taken as an example.

[0017] The membrane water content cannot be directly measured online by sensors, but it is closely related to the stack operating parameters. The input layer uses stack current, stack temperature, anode inlet pressure, cathode inlet pressure, stack voltage, and hydrogen exhaust solenoid valve duty cycle as input parameters, with a total of 6 nodes.

[0018] The hidden layer is the core of realizing nonlinear mapping. Here, it is determined to be 1 hidden layer and 8 neurons. The activation function of the hidden layer is the ReLU function, the number of output nodes is 1, and the normalized membrane water content estimate is directly output. The activation function is the Sigmoid function.

[0019] The true value of membrane water content was obtained through offline calibration experiments. The proton exchange membrane was removed by disassembling the fuel cell stack using the weighing method, and the membrane water content was calculated. The offline data acquisition needs to cover all operating conditions. Under steady-state conditions, the current needs to be fixed and the parameters under different temperatures and pressures need to be recorded. Under dynamic conditions, the load change can be simulated. The current can be stepped from 20A to 150A and from 150A to 20A. The rise and fall rate is set to 5A / s, the sampling frequency is set to 1Hz, 10,000 sets of valid data are collected, and some invalid data caused by sensor abnormalities and sudden changes in operating conditions are deleted. All parameters are mapped to the [0,1] interval.

[0020] Randomly initialize weights W and biases b. Based on the loss obtained from forward propagation, in backpropagation, use the Adam optimizer to calculate the partial derivatives of the loss function with respect to the weights and biases respectively, and update the weights and biases in the direction of gradient descent.

[0021] The core formula for parameter update in gradient descent is:

[0022]

[0023] To quantify the fitting accuracy of the neural network model and determine whether the model meets the requirements, three accuracy indicators are calculated: coefficient of determination, mean absolute error, and maximum error.

[0024] After ensuring the model's reliability, the actual membrane water content value is obtained through inverse normalization;

[0025] Furthermore, in a third aspect, the present invention provides an artificial neural network model for designing a three-valve linkage controller;

[0026] The input layer takes five parameters, including membrane water content deviation. fuel cell current Anode hydrogen pressure fuel cell stack temperature Hydrogen cylinder pressure The output layer parameters are the opening degree of the electronically controlled pressure reducing valve, the hydrogen inlet proportional valve, and the hydrogen outlet solenoid valve. The number of hidden layers is selected as two layers, with the number of neurons in hidden layer 1 set to 16 and the number of neurons in hidden layer 2 set to 8.

[0027] Using the collected full-condition data as a dataset, the steps for designing the three-valve coordinator are similar to those for estimating membrane water content. First, the data of the three valves are mapped to the [0,1] interval to obtain three normalized calculation formulas. Then, the input and output formulas of each layer of the neural network are written. Next, the loss function formula is written. Then, in backpropagation, the Adam optimizer is used to calculate the partial derivatives of the loss function with respect to the weights and biases, and the weights and biases are updated according to the gradient descent direction. The accuracy index is calculated in the same way. Finally, the opening size of the three valves is obtained by inverse normalization.

[0028] The weighted loss function is calculated as follows:

[0029]

[0030]

[0031]

[0032] In the formula, Let be the MWC bias weight for the i-th sample (n=20000). The anodic hydrogen pressure threshold weight is the weight of the i-th sample. The normalized opening degree of the j-th valve predicted by the model. Let be the optimal normalized opening degree of the j-th valve;

[0033] After inverse normalization, gradient adjustment verification is performed, taking an electronically controlled pressure reducing valve as an example:

[0034] If the above equation is satisfied, no correction is needed; adjust directly.

[0035] If the above equation is not satisfied, adjust by 1% each time until the above relationship is satisfied. The final valve opening calculation formula is:

[0036] .

[0037] Furthermore, in a fourth aspect, the present invention provides a PID-controlled valve opening scheme, comprising:

[0038] Obtain the current hydrogen pressure feedback value of the hydrogen system at a pipe diameter of 5 to 10 times downstream of the pressure reducing valve outlet. Additionally, MWC estimates and stack currents also need to be obtained. Hydrogen cylinder pressure The sampling frequency is 1Hz;

[0039] Using a positional PID controller, the output is the valve opening adjustment amount, and the formula is:

[0040]

[0041] Set the electronically controlled pressure reducing valve to the first priority, the hydrogen inlet proportional valve to the second priority, and the hydrogen discharge solenoid valve to the third priority.

[0042] First, calculate the hydrogen pressure deviation. The relationship between the deviation and the threshold is compared. If the deviation is greater than the threshold, the deviation is substituted into the above formula to calculate the first priority opening. After the adjustment is completed, a new feedback value is collected again, and the relationship between the deviation and the threshold is compared again. If the deviation is less than the threshold and remains above 0.1s, it is determined to be stable and enters the next priority valve adjustment.

[0043] Next, calculate the hydrogen inlet flow rate deviation. It is necessary to Perform coupling correction, compare the corrected value with the threshold. If it is greater than the threshold, calculate the second priority opening, wait 0.5s for adjustment, re-collect feedback value after adjustment, and continue to compare the relationship between deviation and threshold. If the deviation is less than the threshold and remains above 0.1s, it is determined to be stable and enters the final priority.

[0044] Finally, the membrane water content deviation was calculated. Similar to the second priority adjustment steps, it is necessary to... After performing coupling correction, wait 0.5s before adjusting again. Once stable, start a new round of adjustment, re-acquire feedback parameters, and calculate the deviation. If the hydrogen pressure fluctuation exceeds the threshold, slightly adjust the PID parameters until they stabilize.

[0045] Compared with the prior art, the present invention has the following beneficial effects: by predicting the membrane water content based on an artificial neural network model and correcting the valve opening; by obtaining the current hydrogen pressure fluctuation value of the hydrogen system during the hydrogen discharge process, and by using PID control to fine-tune the valve again, the adaptive control of the system under all operating conditions and the decoupled control under multiple outputs are realized, ultimately achieving the goal of precise hydrogen supply and stable operation of the fuel cell stack.

[0046] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the system architecture of the hydrogen system and the air system in an embodiment of the present invention;

[0048] Figure 2 This is a flowchart of the three-valve combined intelligent control method in the fuel cell control system under different operating conditions in an embodiment of the present invention;

[0049] Figure 3 This is a table showing the changes in valve opening and PWM duty cycle under different operating conditions in this embodiment of the invention.

[0050] In the diagram: 1-Hydrogen cylinder, 2-Solenoid valve, 3-Electrically controlled pressure reducing valve, 4-Hydrogen inlet proportional valve, 5-Ejector, 6-Air filter, 7-Flow meter, 8-Air compressor, 9-Intercooler, 10-Humidifier, 11-Inlet valve, 12-Hydrogen exhaust solenoid valve, 13-Exhaust throttle valve, 14-Gas-water separator. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0052] A proton exchange membrane (PEM) hydrogen fuel cell system is a power generation device that directly converts the chemical energy of hydrogen and oxygen into electrical energy. Its basic principle is that hydrogen gas is introduced into the anode plate, flows through hydrogen channels on the anode plate, and evenly passes through the gas diffusion layer (microporous carbon paper). It comes into full contact with the catalyst (usually Pt or alloy catalyst). Under the action of the catalyst, the hydrogen gas undergoes an oxidation reaction on the proton exchange membrane, decomposing into protons (i.e., hydrogen ions) and electrons. The protons adhere to water molecules on the membrane surface to form hydrated hydrogen ions. These hydrated hydrogen ions pass through the proton exchange membrane to the cathode, while the electrons reach the cathode through an external circuit, forming an electric current. The oxidation reaction is as follows:

[0053]

[0054] At the cathode end, oxygen flows through the gas channels on the cathode plate, uniformly passing through the gas diffusion layer to reach the catalyst layer. Under the action of the catalyst, a reduction reaction occurs, and the oxygen atom is broken down to form weak O-Pt bonds attached to the Pt surface. Then, each oxygen atom leaves the catalyst Pt and combines with electrons from the external circuit and hydrogen ions penetrating from the proton exchange membrane to form water molecules, generating a certain amount of heat. The reduction reaction is as follows:

[0055]

[0056] The overall reaction of a fuel cell is:

[0057]

[0058] Therefore, as long as the fuel required for fuel cells—hydrogen and oxygen—is continuously supplied, fuel cells can generate electricity continuously.

[0059] Hydrogen supply to a proton exchange membrane fuel cell system typically involves delivering hydrogen to the stack via a hydrogen storage tank. To store a larger mass of hydrogen, high pressure (35 MPa / 70 MPa) is usually used in the storage tank to increase the hydrogen density. When delivering hydrogen to the stack, the high-pressure gas must be reduced to a medium pressure of 1–2 MPa, and then the pressure is regulated by a proportional valve or hydrogen injector before being supplied to the stack. During this process, sudden changes in stack operating conditions can cause pressure fluctuations entering the stack, leading to continuous mechanical fatigue damage to the proton exchange membrane and ultimately resulting in an irreversible reduction in the stack's lifespan.

[0060] The following is combined with Figures 1 to 3 The embodiments shown illustrate the solution of the present invention;

[0061] Example 1

[0062] Figure 1 A schematic diagram of a system architecture for a hydrogen supply system and an air supply system according to an embodiment of the present invention is shown.

[0063] A fuel cell system includes a fuel cell stack, a hydrogen supply system, an air supply system, a temperature control system, and an electronic control system. Figure 1 This section primarily introduces the hydrogen supply system and the oxygen supply system. The hydrogen supply system mainly provides hydrogen to the fuel cell stack at a certain pressure, flow rate, temperature, and humidity. The oxygen required for the electrochemical reaction of the fuel cell comes from the air, and the air supply system mainly provides clean air to the fuel cell stack at a certain pressure, flow rate, temperature, and humidity.

[0064] In this embodiment of the invention, the hydrogen supply system includes a hydrogen cylinder 1 and a solenoid valve 2, an electronically controlled pressure reducing valve 3, a hydrogen inlet proportional valve 4, and an ejector 5 arranged sequentially along the hydrogen delivery line. The hydrogen cylinder 1 is connected to the solenoid valve 2 via a pipeline. The air supply system includes an air filter 6, a flow meter 7, an air compressor 8, an intercooler 9, a humidifier 10, and an intake valve 11 arranged sequentially along the air delivery line. The exhaust system includes a hydrogen exhaust solenoid valve 12 and an exhaust throttle valve 13.

[0065] In this embodiment of the invention, the efficient operation of the fuel cell requires a continuous and stable supply of hydrogen, and the hydrogen should enter the fuel cell at a certain pressure, flow rate and humidity. The hydrogen supply system is designed to achieve this function.

[0066] In this embodiment of the invention, the output of the hydrogen source needs to be controlled on and off, and a switch-type solenoid valve should be installed. Hydrogen is a flammable and explosive gas, and the components selected for the entire fuel cell system must not generate sparks; therefore, the solenoid valve must be explosion-proof. The hydrogen output from the hydrogen source is often at a high pressure, so a pressure reducing valve is needed to reduce it to a suitable pressure range. However, the pressure reducing valve's pressure stabilization capability is weak; with changes in hydrogen flow rate, the output hydrogen pressure of the pressure reducing valve will fluctuate significantly. The fuel cell's output current varies with the load, resulting in frequent fluctuations in hydrogen flow rate. If the internal hydrogen pressure of the fuel cell changes drastically as a result, it can lead to fatigue damage to the proton exchange membrane and even accelerate catalyst shedding, thereby reducing the fuel cell's lifespan. Therefore, a pressure sensor should be installed to measure the hydrogen pressure. To achieve precise control of the hydrogen flow rate, a hydrogen inlet proportional valve should be installed. To achieve the recycling of unreacted hydrogen, improve hydrogen utilization, simplify the system structure, and reduce energy consumption and cost, an ejector should be installed. Not all the hydrogen at the stack anode participates in the reaction; unreacted hydrogen will be discharged along with the generated water. Without an ejector, unreacted hydrogen would be directly vented, resulting in wasted hydrogen. After entering the fuel cell stack, it reacts electrochemically with oxygen on the surface of the proton exchange membrane, producing electricity and water. Unreacted hydrogen flows out from the fuel cell hydrogen outlet. Theoretically, the hydrogen outlet could be sealed to prevent gas outflow. However, water generated at the cathode can seep into the anode, meaning there will be generated water and humidifying water within the hydrogen flow path. If not drained, this would flood the stack, reducing the actual active area of ​​the fuel cell and decreasing its performance and lifespan. Therefore, a pressure sensor is installed at the outlet to monitor the outlet hydrogen pressure, and a gas-water separator 14 is installed to allow water from the anode to flow into and be collected by the separator. A solenoid valve is installed at the output of the gas-water separator to connect to the atmosphere, allowing water to be discharged under the control of the electronic control system.

[0067] In this embodiment of the invention, the air supply system feeds air into the fuel cell, causing the oxygen in the air to undergo an electrochemical reaction with hydrogen inside the fuel cell to generate electrical energy.

[0068] In this embodiment of the invention, air is filtered to remove impurities, and the flow rate of the filtered air is measured using a flow meter. The air then enters the air compressor, increasing the oxygen partial pressure and accelerating the reaction rate. The air compression process follows the principle of "adiabatic compression," causing a rapid temperature increase. High temperatures reduce the solubility of oxygen in the gas and accelerate proton exchange membrane dehydration, affecting the fuel cell stack's reaction efficiency. An intercooler lowers the temperature of the compressed air using either air or water cooling. The proton conduction capacity of the proton exchange membrane depends entirely on the moisture within the membrane. If the air is too dry, the membrane will dehydrate, leading to a significant increase in proton conduction resistance and a substantial drop in the fuel cell stack's output voltage. After humidification, the air enters the fuel cell with a certain level of humidity. An intake valve is installed between the humidifier and the fuel cell stack cathode inlet to assist the air compressor in achieving precise airflow control. An exhaust throttle valve is installed between the fuel cell stack cathode outlet and the humidifier (exhaust gas recirculation branch) to regulate the cathode chamber pressure and, in conjunction with the humidifier, achieve moisture recovery and reuse.

[0069] Example 2

[0070] Figure 2 A flowchart illustrating a three-valve combined intelligent control method in a fuel cell system under different operating conditions is shown.

[0071] The specific process is as follows:

[0072] First, perform curve calibration on the three valves to establish the mapping relationship between valve opening and actual output.

[0073] Curve calibration is used to accurately connect theory with practice. Theoretical parameters are measured under idealized conditions, but in reality, they are affected by many unexpected situations and uncontrollable factors, causing deviations between reality and theory. Curve calibration can use measured data to correct theoretical values, making the model closer to the real situation. In addition, each valve has different influencing factors, and curve calibration can create a unique characteristic curve for each valve.

[0074] The core function of the electrically controlled pressure reducing valve is to control the intermediate hydrogen pressure. Therefore, it is necessary to establish an "opening degree - intermediate hydrogen pressure" curve. The hydrogen cylinder pressure and fuel cell stack temperature are used as calibration conditions. The hydrogen cylinder pressure is set to three levels: low (10MPa), medium (35MPa), and high (70MPa). The fuel cell stack temperature is set to three levels: low (25°C), operating temperature (60°C), and high (80°C). The calibration steps for the electrically controlled pressure reducing valve are as follows:

[0075] First, fix a set of boundary conditions (hydrogen tank pressure 35 MPa, fuel cell stack temperature 60 °C). The opening command was increased in increments of 5% from 0% to 100%, with each opening increment running stably for 30 seconds. The frequency was set to 1Hz to sample the hydrogen medium pressure data at the outlet of the pressure reducing valve. The average hydrogen medium pressure data over these 30 seconds was taken as the stable hydrogen medium pressure value at that opening. The boundary conditions were changed, and full-condition data was collected. The data was then fitted using the least squares method, and the linear curve coefficients obtained were:

[0076]

[0077] In the formula, For valve opening, This represents the average valve opening. This is the pressure value of hydrogen. is the average pressure in hydrogen, and k is the fitting coefficient;

[0078] The final linear curve expression for the electrically controlled pressure reducing valve is:

[0079]

[0080] In the formula, Where is the valve opening, P is the hydrogen medium pressure value, and b is the fitting coefficient;

[0081] The core function of the hydrogen proportional valve is to control the hydrogen flow rate. Therefore, it is necessary to establish an "opening degree - hydrogen flow rate" curve, using the hydrogen intermediate pressure and the fuel cell stack temperature as calibration conditions. The hydrogen intermediate pressure is set to a low target value (0.15 MPa), a typical target value (0.2 MPa), and a high target value (0.25 MPa), respectively. The fuel cell stack temperature is fixed at the operating temperature (60°C). The calibration steps for the hydrogen inlet proportional valve are as follows:

[0082] First, fix a set of boundary conditions (hydrogen pressure 0.2 MPa, stack temperature 60 °C). The opening command was increased in increments of 5% from 0% to 100%, with each opening set for 30 seconds of stable operation. Hydrogen flow rate data at the outlet of the hydrogen inlet proportional valve was sampled at a frequency of 1Hz. The average hydrogen flow rate over these 30 seconds was taken as the stable hydrogen flow rate at that opening. The boundary conditions were changed, and full-condition data was collected. The data was fitted using the least squares method, yielding the following nonlinear curve expression:

[0083]

[0084] In the formula, Q represents the hydrogen flow rate. Let be the valve opening degree, and a, b, and c be the fitting coefficients;

[0085] To solve for the fitting coefficients and find the optimal curve, the formula for the sum of squared residuals is:

[0086]

[0087] By taking the partial derivatives with respect to the fitting coefficients a, b, and c, and setting them to 0, the coefficients can be directly calculated. After simplification, a system of three linear equations in three variables is obtained:

[0088]

[0089] The coefficients can be obtained by substituting the data into the variables required for the above system of equations.

[0090] The core function of the hydrogen discharge solenoid valve is to discharge water and hydrogen. Therefore, a "duty cycle - drainage effect" curve can be established. The hydrogen tank pressure is set to 35 MPa, the hydrogen intermediate pressure to 0.2 MPa, and the fuel cell stack temperature to 60°C. The duty cycle PWM is increased in increments of 5% from 0% to 100%, simulating three fuel cell stack conditions: flooded, normal, and dry. The calibration steps for the hydrogen discharge solenoid valve curve are as follows:

[0091] First, fix a set of boundary conditions (PWM duty cycle of 30%, fuel cell stack status as normal), and fix other condition values: hydrogen pressure of 0.2 MPa, hydrogen tank pressure of 35 MPa, and fuel cell stack temperature of 60°C. The simulated fuel cell stack conditions included flooding, normal operation, and dryness. The duty cycle was increased from 0% to 100% in 5% increments, with each increment held for 30 seconds to ensure steady-state flow and pressure. The frequency was set to 1Hz. Water accumulation was collected, boundary conditions were changed, and full-condition data was acquired. The drainage rate was used to reflect the drainage effect, and the formula for calculating this curve is as follows:

[0092]

[0093] In the formula, For drainage rate, This is the upper limit of the saturation range for the drainage rate. Duty cycle, For coefficients;

[0094] This will be the initial Defined as the maximum value in the measured data, and transforming the nonlinear model into a linear model:

[0095]

[0096] In the formula, , ,

[0097] Using the least squares method to find The calculation formula is:

[0098]

[0099] right Find the derivative and set it to 0, then substitute it into the equation. Seeking for:

[0100]

[0101] The answer can be obtained by substituting it into a linear model;

[0102] Once the curve is established, the valve needs to be triggered, and the predicted current change rate can serve as a dynamic trigger for the valve action.

[0103] Second, calculate the predicted value of current change and deduce the target parameters.

[0104] The load demand of a fuel cell is constantly changing, and the rate of change of current directly determines the instantaneous growth rate of the demand for hydrogen and oxygen. It can directly reflect the rate of change of the electrochemical reaction rate, which is the core factor determining the performance and lifespan of a fuel cell.

[0105] The constraints of hydrogen supply to the anode, air compressor at the cathode, and ambient temperature in a fuel cell are dynamically adjustable external factors that determine the real-time upper limit of the rate of change of current under actual operating conditions.

[0106] Calculate the maximum rate of change of current under hydrogen supply constraints at the anode of the fuel cell:

[0107]

[0108] Calculate the rate of change of current under the constraint of the fuel cell cathode air compressor:

[0109]

[0110] Calculate the rate of change of current under ambient temperature constraints:

[0111]

[0112] As the "barrel effect" shows, the amount of water a barrel can hold is determined not by its longest plank, but by its shortest. Similarly, the upper limit of the rate of change of electric current is determined by the most stringent influencing factors.

[0113] Therefore, the predicted value for the final output current change rate is:

[0114]

[0115] Knowing the predicted rate of change of current, we can anticipate the dynamic demands of the fuel cell stack. To achieve the predicted value, we can adjust the valve opening. However, adjusting the valve opening requires a clear "target flow rate" as a reference.

[0116] Calculating the target hydrogen flow rate can match the power generation needs of the fuel cell stack, avoiding hydrogen shortages or waste.

[0117] Determining the target hydrogen flow rate also requires coordination with the supply rate, which can be referenced by calculating the anode pressure value.

[0118] Calculate the target hydrogen flow rate:

[0119]

[0120] Calculate the target anode pressure:

[0121]

[0122] Third, adjust the valve opening based on the predicted and target values ​​of the current change rate.

[0123] In this embodiment of the invention, the valve needs to be opened wider or closed narrower based on the sign of the predicted value. A calibration curve of valve opening degree-flow rate-pressure is established as the basis for converting the target value into the opening degree. The valve opening degree is corrected based on the target value to bring the current measured value closer to the target flow rate value. The final opening degree adjustment is a combination of the two.

[0124] Figure 3 The table shows the changes in valve opening and PWM duty cycle under different operating conditions;

[0125] In this embodiment of the invention, the opening sizes of three valves—the electrically controlled pressure reducing valve, the hydrogen inlet proportional valve, and the hydrogen venting solenoid valve—can be adjusted. The pressure reducing valve reduces the inlet pressure to a desired outlet pressure by adjusting the valve opening, and automatically maintains a stable outlet pressure by relying on the energy of the medium itself. The core function of the hydrogen inlet proportional valve is to precisely match the hydrogen flow requirements of the fuel cell stack and quickly respond to dynamic load changes. The core function of the hydrogen venting solenoid valve is to quickly vent residual hydrogen or abnormal pressure at the anode under specific operating conditions.

[0126] like This indicates an increased load, requiring the electronically controlled pressure reducing valve to open wider in advance to replenish pressure reserves, offset the impact of the sudden increase in flow on pressure, and prevent medium-pressure collapse. The hydrogen inlet proportional valve should open simultaneously, while the hydrogen exhaust solenoid valve should open with a delayed start to ensure a sufficient hydrogen supply to the fuel cell stack.

[0127] like This indicates that the load has not changed, and the current valve status can be maintained.

[0128] like This indicates that the load has decreased, so the electronically controlled pressure reducing valve needs to be closed in advance, the hydrogen inlet proportional valve needs to be closed in sync, and the hydrogen discharge solenoid valve needs to be opened quickly.

[0129] Fourth, based on an artificial neural network model, the membrane water content is predicted and a three-valve coordinated controller is designed to initially adjust the valves.

[0130] Artificial neural networks are machine learning models built by simulating the structure and information transmission of neurons in the human brain. At their core, they achieve fitting, learning, prediction, and decision-making on complex nonlinear relationships through hierarchical connections and parameter optimization of a large number of simple processing units (neurons). Their core essence is to automatically learn the mapping rules between input and output through data training, without relying on precise physical / mathematical analytical models. They are particularly good at solving problems involving multi-parameter coupling, strong nonlinearity, and complex mechanisms that are difficult to model. This is also the key reason why they can be applied in engineering scenarios such as fuel cell membrane water content estimation and three-valve linkage control.

[0131] Example 3

[0132] The membrane water content cannot be directly measured online by sensors, but it is closely related to the stack operating parameters. The input layer uses stack current, stack temperature, anode inlet pressure, cathode inlet pressure, stack voltage, and hydrogen exhaust solenoid valve duty cycle as input parameters, with a total of 6 nodes.

[0133] The hidden layer is the core of the nonlinear mapping; here it is determined to have 1 hidden layer and 8 neurons. The activation function for the hidden layer is the ReLU function, and the formula is:

[0134]

[0135] The output node count is set to 1 layer, directly outputting the normalized membrane water content estimate. The activation function is the Sigmoid function, and the formula is:

[0136]

[0137] The true value of membrane water content was obtained through offline calibration experiments. The proton exchange membrane was removed from the fuel cell stack using a weighing method, and the membrane water content was calculated. Offline data acquisition needed to cover all operating conditions. Under steady-state conditions, a fixed current was required, and parameters were recorded under different temperatures and pressures. Under dynamic conditions, load changes were simulated, with the current stepping from 20A to 150A and then back down to 20A. The rise and fall rate was set to 5A / s, and the sampling frequency was set to 1Hz. 10,000 sets of valid data were collected, and invalid data due to sensor malfunctions or sudden changes in operating conditions were deleted. All parameters were mapped to the [0,1] interval. The calculation formula is as follows:

[0138]

[0139] In the formula, The original data values, This is the minimum value of the parameter across all samples. This is the maximum value of the parameter across all samples. These are the normalized data values;

[0140] Randomly initialize the weights W and bias b. During forward propagation, the hidden layer input formula is:

[0141]

[0142] In the formula, Input data;

[0143] The hidden layer output formula is:

[0144]

[0145] The input formula for the output layer is:

[0146]

[0147] The formula for outputting the estimated water content of the output membrane is:

[0148]

[0149] The formula for the loss function is:

[0150]

[0151] In the formula, The actual value;

[0152] Based on the loss obtained from forward propagation, the Adam optimizer is used to calculate the loss function pairs during backpropagation. The partial derivatives of the weights and biases are used to update the weights and biases in the direction of gradient descent, and the formula is as follows:

[0153]

[0154]

[0155]

[0156]

[0157]

[0158]

[0159] The core formula for parameter update in gradient descent is:

[0160]

[0161] In the formula, For the updated weight parameters, For the weight parameters to be updated, The learning rate is set to 0.001. For the loss function with respect to weights The partial derivatives;

[0162] To quantify the fitting accuracy of the neural network model and determine whether the model meets the requirements, three accuracy metrics are calculated here. The formula for calculating the coefficient of determination is as follows:

[0163]

[0164] The formula for calculating the mean absolute error is:

[0165]

[0166] The formula for calculating the maximum error is:

[0167]

[0168] After ensuring the model's reliability, the actual membrane water content value is obtained through inverse normalization:

[0169]

[0170] The input layer takes five parameters, including membrane water content deviation. fuel cell current Anode hydrogen pressure fuel cell stack temperature Hydrogen cylinder pressure The output layer parameters are the opening degree of the electronically controlled pressure reducing valve, the hydrogen inlet proportional valve, and the hydrogen outlet solenoid valve. The number of hidden layers is selected as two layers, with the number of neurons in hidden layer 1 set to 16 and the number of neurons in hidden layer 2 set to 8.

[0171] Based on the aforementioned three valve calibration curves, the collected full-condition data is used as the dataset. The steps for designing the three-valve coordinator are similar to those for estimating membrane water content. First, the data of the three valves are mapped to the [0,1] interval to obtain three normalized calculation formulas. Then, the input and output formulas of each layer of the neural network are written. Next, the loss function formula is written. Then, in backpropagation, the Adam optimizer is used to calculate the partial derivatives of the loss function with respect to the weights and biases, and the weights and biases are updated according to the gradient descent direction. The accuracy index is calculated in the same way. Finally, the opening size of the three valves is obtained by inverse normalization. Here, only some formulas are simply written.

[0172] The weighted loss function is calculated as follows:

[0173]

[0174]

[0175]

[0176] In the formula, Let be the MWC bias weight for the i-th sample (n=20000). The anodic hydrogen pressure threshold weight is the weight of the i-th sample. The normalized opening degree of the j-th valve predicted by the model. Let be the optimal normalized opening degree of the j-th valve;

[0177] After inverse normalization, gradient adjustment verification is performed, taking an electronically controlled pressure reducing valve as an example:

[0178]

[0179] In the formula, The target value of the valve is obtained after inverse normalization;

[0180] If the above equation is satisfied, no correction is needed; adjust directly.

[0181] If the above equation is not satisfied, adjust by 1% each time until the above relationship is satisfied. The final valve opening calculation formula is:

[0182]

[0183] Step 5: Obtain parameters such as the current hydrogen pressure fluctuation value, and use a PID control strategy to fine-tune the valve again.

[0184] Example 4

[0185] When fine-tuning the valve using a PID control strategy, first obtain the current hydrogen pressure feedback value. Additionally, data and frequencies are collected to determine valve priorities.

[0186] The intermediate hydrogen pressure is a critical pressure at the outlet of the pressure reducing valve and upstream of the hydrogen proportional valve. Its fluctuations are directly transmitted to the anode of the fuel cell stack, threatening the safety of the proton exchange membrane and catalyst. If the fluctuation amplitude is too large, it will cause the anode-cathode pressure difference to exceed the limit instantaneously, and the proton exchange membrane will experience mechanical fatigue or even breakdown due to uneven stress. If the fluctuation frequency is too high, it will cause frequent fluctuations in the anode hydrogen concentration, leading to local hydrogen deficiency or hydrogen excess—hydrogen deficiency will cause carbon corrosion of the catalyst, while hydrogen excess will increase the risk of leakage. Measuring the fluctuation value and setting a threshold is essentially to set up a pressure safety line for the fuel cell stack, ensuring that the fluctuation is always within the membrane's tolerance range. The core purpose of measuring the intermediate hydrogen pressure at a distance of 5 to 10 times the pipe diameter downstream of the outlet of the electrically controlled pressure reducing valve is to obtain a stable and accurate pressure value, avoiding the impact of turbulence disturbances at the outlet of the pressure reducing valve on the measurement accuracy.

[0187] In this embodiment of the invention, the hydrogen pressure measured in real time during the hydrogen discharge process can be achieved by installing a medium-pressure sensor at the exhaust port of the pressure reducing valve. This sensor detects the pressure of the gas discharged from the pressure reducing valve in real time. During normal operation of the hydrogen fuel cell system and during the hydrogen discharge process, the discharge of hydrogen causes a pressure drop on the hydrogen side. However, the mechanical strength of the membrane electrode assembly (MEA) in the fuel cell stack is relatively low. To avoid damage to the proton exchange membrane, it is necessary to ensure that the hydrogen side pressure is equal to or slightly higher than the air side pressure, and that the pressures on both sides rise and fall simultaneously as much as possible. During the hydrogen discharge process, if pressure compensation is not performed, excessively low hydrogen side pressure will cause additional stress on the proton exchange membrane, affecting the power generation efficiency and lifespan of the battery. It should be noted that, to ensure the stability of the hydrogen pressure entering the stack, an entry pressure sensor is installed at the output port of the hydrogen supply unit of the hydrogen system (i.e., the hydrogen inlet of the fuel cell stack). This entry pressure sensor is used to monitor the hydrogen pressure entering the stack in real time.

[0188] Obtain the current hydrogen pressure feedback value of the hydrogen system at a pipe diameter of 5 to 10 times downstream of the pressure reducing valve outlet. Additionally, MWC estimates and stack currents also need to be obtained. Hydrogen cylinder pressure The sampling frequency is 1Hz;

[0189] Using a positional PID controller, the output is the valve opening adjustment amount, and the formula is:

[0190]

[0191] In the formula, Subtract the feedback value from the target value. This is the proportionality coefficient. The integral coefficient is... These are the differential coefficients;

[0192] Set the electronically controlled pressure reducing valve to the first priority, the hydrogen inlet proportional valve to the second priority, and the hydrogen discharge solenoid valve to the third priority.

[0193] Calculate hydrogen pressure deviation The relationship between the deviation and the threshold is compared. If the deviation is less than the threshold, the PID regulation is paused. If the deviation is greater than the threshold, the deviation is substituted into the above formula to calculate the first priority opening. After the regulation is completed, a new feedback value is collected again, and the relationship between the deviation and the threshold is compared again. If the deviation is less than the threshold and remains above 0.1s, it is determined to be stable and enters the next priority valve regulation.

[0194] Calculate hydrogen pressure deviation The relationship between the deviation and the threshold is compared. If the deviation is less than the threshold, the PID regulation is paused. If the deviation is greater than the threshold, the deviation is substituted into the above formula to calculate the first priority opening. After the regulation is completed, a new feedback value is collected again, and the relationship between the deviation and the threshold is compared again. If the deviation is less than the threshold and remains above 0.1s, it is determined to be stable and enters the next priority valve regulation.

[0195] Calculate the hydrogen inlet flow rate deviation When correcting the valve opening, even if the first priority adjustment has stabilized, a slight deviation in hydrogen pressure will still cause it to deviate from the true value. Therefore, adjustments are needed. Coupling correction is performed, and the correction formula is as follows:

[0196]

[0197] In the formula, The hydrogen pressure-hydrogen flow rate coupling coefficient;

[0198] If the value is less than the threshold, PID adjustment is paused. If the value is greater than the threshold, the second priority opening is calculated, and adjustment is performed after 0.5 seconds. After adjustment, the feedback value is collected again, and the relationship between the deviation and the threshold is compared again. If the deviation is less than the threshold and remains above 0.1 seconds, it is determined to be stable and enters the final priority.

[0199] Calculate membrane moisture content deviation Similar to the second priority adjustment steps, it is necessary to... Coupling correction is performed, and the correction formula is as follows:

[0200]

[0201] In the formula, The hydrogen pressure-MWC coupling coefficient is... Hydrogen flow rate - MWC coupling coefficient;

[0202] After waiting 0.5 seconds, the system will adjust again. Once the system is stable, a new round of adjustment will begin. After entering the next control cycle, the feedback parameters will be collected again, and the deviation will be calculated. If the hydrogen pressure fluctuation exceeds the threshold, the PID controller parameters will be slightly adjusted until the parameters stabilize.

[0203] This invention is not limited to the preferred embodiment described above. Anyone inspired by this invention can derive various other forms of intelligent three-valve control methods based on the water content of fuel cell membranes. All equivalent variations and modifications made within the scope of the claims of this invention should be considered within the scope of this invention.

Claims

1. A three-valve combined intelligent control method based on the water content of a fuel cell membrane, comprising a fuel cell system, characterized in that: The fuel cell system includes a stack and a hydrogen supply system, an air supply system, and an exhaust system respectively connected to the stack. The hydrogen supply system includes a hydrogen cylinder and an electronically controlled pressure reducing valve, a hydrogen inlet proportional valve, and an ejector arranged sequentially along the hydrogen delivery line. The exhaust system is equipped with a hydrogen discharge solenoid valve, and the air supply system is equipped with an air compressor.

2. The three-valve combined intelligent control method based on the water content of fuel cell membranes according to claim 1, characterized in that: Includes the following steps: Step 1: Perform curve calibration on the electronically controlled pressure reducing valve, hydrogen inlet proportional valve, and hydrogen discharge solenoid valve respectively to establish the mapping relationship between valve opening degree and actual output; Step 2: Calculate the predicted current change value and deduce the target parameters; Step 3: Adjust the valve opening based on the predicted and target values ​​of the current change rate; Step 4: Based on the artificial neural network model, predict the membrane water content and design a three-valve coordinated controller to adjust the valves.

3. The three-valve combined intelligent control method based on the water content of fuel cell membranes according to claim 2, characterized in that: The hydrogen intermediate pressure is controlled by an electrically controlled pressure reducing valve, and an "opening degree - hydrogen intermediate pressure" curve is established. The hydrogen cylinder pressure and fuel cell stack temperature are used as calibration conditions. The hydrogen cylinder pressure is set to three levels: low (10 MPa), medium (35 MPa), and high (70 MPa), and the fuel cell stack temperature is set to low (25 MPa). ), operating temperature (60) ), high temperature (80) Three gears; The calibration steps for the electrically controlled pressure reducing valve are as follows: First, fix a set of boundary conditions (hydrogen tank pressure 35 MPa, fuel cell stack temperature 60 °C). The opening command was increased in increments of 5% from 0% to 100%, with each opening increment running stably for 30 seconds. The frequency was set to 1Hz to sample the hydrogen medium pressure data at the outlet of the electronically controlled pressure reducing valve. The average hydrogen medium pressure data over these 30 seconds was taken as the stable hydrogen medium pressure value at that opening. The boundary conditions were changed, and full-condition data was collected. The data was then fitted using the least squares method, and the linear curve coefficients obtained were: In the formula, For valve opening, This represents the average valve opening. This is the pressure value of hydrogen. is the average pressure in hydrogen, and k is the fitting coefficient; The final linear curve expression for the electrically controlled pressure reducing valve is: In the formula, Where is the valve opening, P is the hydrogen medium pressure value, and b is the fitting coefficient; The hydrogen flow rate is controlled by a proportional hydrogen inlet valve, and an "opening degree - hydrogen flow rate" curve is established. The hydrogen intermediate pressure and stack temperature are used as calibration conditions. The hydrogen intermediate pressure is set to a low target value (0.15 MPa), a typical target value (0.2 MPa), and a high target value (0.25 MPa), respectively. The stack temperature is fixed at the operating temperature (60°C). ); The calibration steps for the hydrogen inlet proportional valve are as follows: First, fix a set of boundary conditions (hydrogen pressure 0.2 MPa, stack temperature 60 °C). The opening command was increased in increments of 5% from 0% to 100%, with each opening set for 30 seconds of stable operation. Hydrogen flow rate data at the outlet of the hydrogen inlet proportional valve was sampled at a frequency of 1Hz. The average hydrogen flow rate over these 30 seconds was taken as the stable hydrogen flow rate at that opening. The boundary conditions were changed, and full-condition data was collected. The data was fitted using the least squares method, yielding the following nonlinear curve expression: In the formula, Q represents the hydrogen flow rate. Let be the valve opening degree, and a, b, and c be the fitting coefficients; To solve for the fitting coefficients and find the optimal curve, the formula for the sum of squared residuals is: By taking the partial derivatives of the fitted coefficients a, b, and c respectively and setting them to 0, the coefficients can be directly calculated. The three linear equations are as follows: Simplifying, we obtain a system of three linear equations in three variables: The coefficients can be obtained by substituting the data into the variables required for the above system of equations. Hydrogen is drained using a hydrogen venting solenoid valve, and a "duty cycle - drainage effect" curve is established. The hydrogen cylinder pressure is set to 35 MPa, the hydrogen intermediate pressure to 0.2 MPa, and the fuel cell stack temperature to 60°C. The duty cycle PWM is increased in increments of 5% from 0% to 100% to calibrate the curves under three fuel cell stack conditions: flooded, normal, and dry. The calibration steps for the hydrogen discharge solenoid valve are as follows: First, fix a set of boundary conditions (PWM duty cycle of 30%, fuel cell stack status as normal), and fix other condition values: hydrogen pressure of 0.2 MPa, hydrogen tank pressure of 35 MPa, and fuel cell stack temperature of 60°C. The simulated fuel cell stack conditions included flooding, normal operation, and dryness. The duty cycle was increased from 0% to 100% in 5% increments, with each increment held for 30 seconds to ensure steady-state flow and pressure. The frequency was set to 1Hz. Water accumulation was collected, boundary conditions were changed, and full-condition data was acquired. The drainage rate was used to reflect the drainage effect, and the formula for calculating this curve is as follows: In the formula, For drainage rate, This is the upper limit of the saturation range for the drainage rate. Duty cycle, For coefficients; This will be the initial Defined as the maximum value in the measured data, and transforming the nonlinear model into a linear model: In the formula, , , Using the least squares method to find The calculation formula is: right Find the derivative and set it to 0, then substitute it into the equation. Seeking for: The answer can be obtained by substituting it into a linear model.

4. The three-valve combined intelligent control method based on the water content of fuel cell membranes according to claim 3, characterized in that: The predicted value of current change is calculated based on the input of various physical parameters into the mechanism model. The factors affecting the predicted value of current change include: the pressure regulation of hydrogen by the hydrogen ejector, the amount of compressed air supplied to the cathode by the air compressor, and the ambient temperature. Based on the above influencing factors, the current change rate of each factor was calculated, and the minimum value was taken as the predicted value of the output current change rate of the fuel cell stack.

5. The three-valve combined intelligent control method based on the water content of fuel cell membranes according to claim 4, characterized in that: Calculate the rate of change of current for each influencing factor, including: Calculate the maximum rate of change of current under hydrogen supply constraints at the anode of the fuel cell: In the formula, It is Faraday's constant. It is the rate of change of hydrogen flow rate in the ejector. It is the anodic hydrogen pressure. It's the ejector efficiency. It is the hydrogen gas constant. It is the operating temperature of the fuel cell stack. It is hydrogen utilization rate. It is the number of individual battery cells; Calculate the rate of change of current under the constraint of the fuel cell cathode air compressor: In the formula, It is the air volume change rate of the air compressor. It refers to the air compressor efficiency. It is the volume fraction of oxygen in the air. It is oxygen utilization rate. It is the air volume constant. It is the number of individual battery cells; Calculate the rate of change of current under ambient temperature constraints: In the formula, It is the maximum rate of change of current at room temperature. It is the ambient temperature; The predicted rate of change of the fuel cell output current is: 。 6. The three-valve combined intelligent control method based on the water content of fuel cell membranes according to claim 5, characterized in that: Adjusting the valve opening includes: converting the predicted value into a target valve opening through reverse calculation using a formula. Calculate the target hydrogen flow rate: In the formula, It is the target current. It is the target anode pressure; Calculate the target anode pressure: In the formula, It is the safety pressure difference between the anode and cathode. It is the hydrogen flow rate versus the flow channel resistance coefficient. It is the target cathode pressure; Calculate the target oxygen flow rate: Calculate the target cathode pressure: In the formula, It is the airflow-channel resistance coefficient.

7. The three-valve combined intelligent control method based on the water content of fuel cell membranes according to claim 6, characterized in that: Based on the predicted value of the current change rate, the valve can be adjusted in advance: like This indicates an increased load, requiring the electronically controlled pressure reducing valve to open earlier, the intake proportional valve to open simultaneously, and the hydrogen exhaust solenoid valve to open later. like This indicates that the load has not changed, and the current valve status can be maintained. like This indicates that the load has decreased, requiring the electronically controlled pressure reducing valve to be closed in advance, the hydrogen inlet proportional valve to be closed simultaneously, and the hydrogen discharge solenoid valve to be opened quickly.

8. The three-valve combined intelligent control method based on the water content of fuel cell membranes according to claim 7, characterized in that: Intelligent valve regulation can be achieved by incorporating artificial neural network models, including: Artificial neural network models can be applied in intelligent control, including: Accurate estimation of membrane water content (WMC); The membrane water content cannot be directly measured online by sensors. It is closely related to the stack operating parameters. The input layer uses stack current, stack temperature, anode inlet pressure, cathode inlet pressure, stack voltage, and hydrogen exhaust solenoid valve duty cycle as input parameters, with a total of 6 nodes. The hidden layer is the core of the nonlinear mapping. It is determined to have one hidden layer and eight neurons. The activation function of the hidden layer is the ReLU function, and the formula is: The output node count is set to 1 layer, directly outputting the normalized membrane water content estimate. The activation function is the Sigmoid function, and the formula is: The true value of membrane water content was obtained through offline calibration experiments. The proton exchange membrane was removed from the fuel cell stack using a weighing method, and the membrane water content was calculated. Offline data acquisition needed to cover all operating conditions. Under steady-state conditions, a fixed current was required, and parameters were recorded under different temperatures and pressures. Under dynamic conditions, vehicle acceleration and deceleration were simulated, with the current stepping from 20A to 150A and then back down to 20A. The rise and fall rate was set to 5A / s, and the sampling frequency was set to 1Hz. 10,000 sets of valid data were collected, and invalid data due to sensor malfunctions or sudden changes in operating conditions were deleted. All parameters were mapped to the [0,1] interval. The calculation formula is as follows: In the formula, The original data values, This is the minimum value of the parameter across all samples. This is the maximum value of the parameter across all samples. These are the normalized data values; Randomly initialize the weights W and bias b. During forward propagation, the hidden layer input formula is: In the formula, Input data; The hidden layer output formula is: The input formula for the output layer is: The formula for outputting the estimated water content of the output membrane is: The formula for the loss function is: In the formula, The actual value; Based on the loss obtained from forward propagation, the Adam optimizer is used to calculate the loss function pairs during backpropagation. The partial derivatives of the weights and biases are used to update the weights and biases in the direction of gradient descent, and the formula is as follows: The core formula for parameter update in gradient descent is: In the formula, For the updated weight parameters, For the weight parameters to be updated, The learning rate is set to 0.

001. For the loss function with respect to weights The partial derivatives; To quantify the fitting accuracy of the neural network model and determine whether the model meets the requirements, three accuracy metrics are calculated here. The formula for calculating the coefficient of determination is as follows: The formula for calculating the mean absolute error is: The formula for calculating the maximum error is: After ensuring the model's reliability, the actual membrane water content value is obtained through inverse normalization: Design a three-valve coordinated controller; The input layer takes five parameters, including membrane water content deviation. fuel cell current Anode hydrogen pressure fuel cell stack temperature Hydrogen cylinder pressure The output layer parameters are the opening degree of the electronically controlled pressure reducing valve, the hydrogen inlet proportional valve, and the hydrogen outlet solenoid valve. Two hidden layers are selected, with the number of neurons in hidden layer 1 set to 16 and the number of neurons in hidden layer 2 set to 8.

9. The three-valve combined intelligent control method based on the water content of fuel cell membranes according to claim 8, characterized in that: Using the collected full-condition data as a dataset, the steps for designing a three-valve coordinated controller are similar to those for estimating membrane water content. First, the data of the three valves are mapped to the [0,1] interval to obtain three normalized calculation formulas. Then, the input and output formulas of each layer of the neural network are written. Next, the loss function formula is written. Then, in backpropagation, the Adam optimizer is used to calculate the partial derivatives of the loss function with respect to the weights and biases, and the weights and biases are updated according to the gradient descent direction. The accuracy index is also calculated. Finally, the opening size of the three valves is obtained by inverse normalization. Only some formulas are written here. The weighted loss function is calculated as follows: In the formula, Let be the MWC bias weight for the i-th sample (n=20000). The anodic hydrogen pressure threshold weight is the weight of the i-th sample. The normalized opening degree of the j-th valve predicted by the model. Let be the optimal normalized opening degree of the j-th valve; After inverse normalization, gradient adjustment verification is performed, taking an electronically controlled pressure reducing valve as an example: In the formula, This is the true value of the valve target obtained after inverse normalization; If the above equation is satisfied, no correction is needed; adjust directly. If the above equation is not satisfied, adjust by 1% each time until the above relationship is satisfied. The final valve opening calculation formula is: 。 10. The three-valve combined intelligent control method based on the water content of fuel cell membranes according to claim 9, characterized in that: The valve adjustment process includes: since a single adjustment cannot completely eliminate all interference, the hydrogen pressure in the fuel cell needs to be measured, and a PID control strategy is used to fine-tune the valve again, specifically including: Obtain the current hydrogen pressure feedback value of the hydrogen system at a pipe diameter of 5 to 10 times downstream of the pressure reducing valve outlet. Additionally, MWC estimates and stack currents also need to be obtained. Hydrogen cylinder pressure The sampling frequency is 1Hz; Using a positional PID controller, the output is the valve opening adjustment amount, and the formula is: In the formula, Subtract the feedback value from the target value. This is the proportionality coefficient. The integral coefficient is... These are the differential coefficients; Set the electronically controlled pressure reducing valve to the first priority, the hydrogen inlet proportional valve to the second priority, and the hydrogen discharge solenoid valve to the third priority. First, calculate the hydrogen pressure deviation. The relationship between the deviation and the threshold is compared. If the deviation is less than the threshold, the PID regulation is paused. If the deviation is greater than the threshold, the deviation is substituted into the above formula to calculate the first priority opening. After the regulation is completed, a new feedback value is collected again, and the relationship between the deviation and the threshold is compared again. If the deviation is less than the threshold and remains above 0.1s, it is determined to be stable and enters the next priority valve regulation. Next, calculate the hydrogen inlet flow rate deviation. When correcting the valve opening, even if the first priority adjustment has stabilized, a slight deviation in hydrogen pressure will still cause it to deviate from the true value. Therefore, adjustments are needed. Coupling correction is performed, and the correction formula is as follows: In the formula, The hydrogen pressure-hydrogen flow rate coupling coefficient; If the value is less than the threshold, PID adjustment is paused. If the value is greater than the threshold, the second priority opening is calculated, and adjustment is performed after 0.5s. After adjustment, the feedback value is collected again, and the relationship between the deviation and the threshold is compared again. If the deviation is less than the threshold and remains above 0.1s, it is determined to be stable and enters the last priority. Finally, the membrane water content deviation was calculated. Similar to the second priority adjustment steps, it is necessary to... Coupling correction is performed, and the correction formula is as follows: In the formula, The hydrogen pressure-MWC coupling coefficient is... Hydrogen flow rate - MWC coupling coefficient; Wait another 0.5 seconds before adjusting, and then proceed with a new round of adjustments once the condition is stable. After entering the next control cycle, the feedback parameters are collected again and the deviation is calculated. If the hydrogen pressure fluctuation exceeds the threshold, the PID parameters are slightly adjusted until the parameters stabilize.

Images