A control method for a fuel cell system

Abstract

The present invention relates to the field of fuel cell control technology and discloses a fuel cell system control method, comprising: step 1, establishing a dynamic model of the fuel cell system, defining the system's state variables as hydrogen concentration, oxygen concentration, membrane moisture content, and stack temperature, and control inputs as hydrogen supply rate, oxygen supply rate, and cooling water flow rate, and describing the dynamic behavior of the fuel cell system through the relationship between the state variables and the control inputs; and step 2, defining the performance targets of the fuel cell system based on the dynamic model established in step 1. By introducing a dynamic feedback gain matrix, based on the output of a deep learning model, the control input is dynamically corrected according to the deviation between the state variables and a reference state variable, thereby achieving adaptive adjustment of the control strategy to the real-time state of the fuel cell system, thereby improving the real-time and accuracy of the control input under dynamic load conditions.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fuel cell control, in particular to a control method of a fuel cell system. BACKGROUND

[0002] Fuel cell system is an important clean energy technology, widely used in transportation, fixed power generation and portable power supply and other fields. The internal dynamic behavior of fuel cell is complex, including nonlinear coupling phenomena of electrochemical reaction, gas transport, heat and mass transfer, and the running state is highly sensitive to the dynamic changes of external load. In order to ensure the stable operation and efficient energy utilization of fuel cell system under different working conditions, the design of control strategy becomes a key technical problem.

[0003] At present, the traditional fuel cell control method mostly adopts control strategy based on linear model, such as PID control and feedback control. The method shows certain stability under static working condition, but under dynamic load condition, it cannot effectively deal with the nonlinear coupling behavior and complex dynamic characteristics of fuel cell system, showing slow response and low fuel utilization rate. In addition, the traditional control method cannot balance the accuracy of power output and the economy of energy utilization, and the resource allocation efficiency is low, which further limits the operation performance of fuel cell system.

[0004] In view of the above problems, in recent years, control technology based on model predictive control and machine learning method is proposed, trying to improve the control accuracy through optimization algorithm. The model predictive control method is strongly dependent on the model, and cannot respond quickly under complex dynamic conditions, while the machine learning method has good nonlinear expression ability, but there are bottlenecks in training data and real-time performance. Therefore, how to efficiently and accurately adjust the control input of fuel cell under complex dynamic conditions, further optimize the power output and improve the energy utilization efficiency, is a problem to be solved in the field of fuel cell control technology.

[0005] The present application aims at the control problem of fuel cell system under dynamic load, introduces dynamic feedback gain matrix, deep learning model and efficiency function, combines with the dynamic adjustment mechanism of weight matrix, and proposes a control method of fuel cell system to solve the above problems. SUMMARY

[0006] In view of the deficiencies of the prior art, the present application provides a control method of a fuel cell system to solve the problems proposed in the background art.

[0007] To achieve the above purpose, the present application realizes the following technical scheme: a control method of a fuel cell system, comprising:

[0008] Step 1, establishing a dynamic model of the fuel cell system, defining the state variables of the system as hydrogen concentration, oxygen concentration, membrane water content and stack temperature, the control inputs as hydrogen supply rate, oxygen supply rate and cooling water flow rate, and describing the dynamic behavior of the fuel cell system through the relationship between the state variables and the control inputs;

[0009] Step 2, defining the performance target of the fuel cell system based on the dynamic model established in Step 1, the performance target being used to measure the relationship between the control inputs and the output power of the fuel cell system, being defined by the difference between the output power and the reference power, and being comprehensively evaluated in combination with the resource consumption and allocation of the control inputs, the reference power being determined according to the load demand, and the output power being calculated through the state variables in the dynamic model;

[0010] Step 3, determining the control strategy of the fuel cell system based on the performance target defined in Step 2, the control strategy being designed according to the difference between the output power and the reference power in the performance target and the changing relationship of the state variables in the dynamic model, and determining the allocation scheme of the control inputs in combination with the influence of the hydrogen supply rate, the oxygen supply rate and the cooling water flow rate on the performance target;

[0011] Step 4, solving the allocation scheme of the control inputs through the deep learning model based on the control strategy determined in Step 3, the input of the deep learning model being the state variables defined in Step 1, and the output being the specific allocation result of the control inputs, the deep learning model being trained through the relationship between the state changes in the dynamic model and the performance target;

[0012] Step 5, passing the allocation result of the control inputs calculated by the deep learning model in Step 4 to the actuators of the fuel cell system, and adjusting the hydrogen supply rate, the oxygen supply rate and the cooling water flow rate defined in Step 1 according to the allocation result, so as to realize the dynamic changes of the state variables of the fuel cell system and complete the control of the fuel cell system.

[0013] Preferably, in Step 1, the dynamic model of the fuel cell system includes the following relationship:

[0014]

[0015] wherein x1(t) is the hydrogen concentration, x2(t) is the oxygen concentration, x3(t) is the membrane water content, x4(t) is the stack temperature, u1, u2, u3 are the hydrogen supply rate, the oxygen supply rate and the cooling water flow rate in turn, k1, k2, k3, k4 are the hydrogen consumption constant, the oxygen consumption constant, the water loss rate and the heat loss rate in turn,

[0016] g(x1, x2, u3) represents the water generation function, and h(x1, x2, x3) represents the heat generation function.

[0017] Preferably, the performance objective in step 2 is defined by the following relationship:

[0018]

[0019] where J is the performance objective, t0is the initial time of fuel cell system control, t f is the final time of fuel cell system control, x(t) is the state variable vector of fuel cell system, u(t) is the control input vector of fuel cell system

[0020] q(x(t), u(t)) represents the difference between the output power y(t) and the reference power y ref (t),

[0021] r(u(t)) represents the cost of control input, and R is a diagonal weight matrix.

[0022] Preferably, the output power in the performance objective is calculated by the following equation:

[0023] y(t) = n(x1(t), x2(t), x3(t), x4(t)),

[0024] where y(t) represents the output power of fuel cell system,

[0025] n(x1(t), x2(t), x3(t), x4(t)) is an efficiency function,

[0026] x1(t) is the hydrogen concentration, x2(t) is the oxygen concentration, x3(t) is the membrane water content, x4(t) is the stack temperature, and n(x1(t), x2(t), x3(t), x4(t)) represents the output power efficiency of fuel cell system under different state variables.

[0027] Preferably, the efficiency function n(x1(t), x2(t), x3(t), x4(t)) is represented by the following equation:

[0028]

[0029] where k e represents the efficiency factor, c e is the compensation coefficient of stack temperature,

[0030] x1•x2 represents the coupling effect of hydrogen and oxygen concentrations, reflecting the influence of fuel cell electrochemical reaction intensity on efficiency,

[0031] x4+c erepresents the overall effect of stack temperature on efficiency, x1(t) is hydrogen concentration, x2(t) is oxygen concentration, x3(t) is membrane moisture content, and x4(t) is stack temperature.

[0032] Preferably, the elements of the weight matrix R of the control input are dynamically adjusted according to the operating conditions of the fuel cell system, and the adjustment method is:

[0033]

[0034] wherein r i is the i-th diagonal element of the control input weight matrix R,

[0035] k r is an adjustment coefficient, represents the sensitivity of the reference power to the control input u i .

[0036] Preferably, the allocation scheme of the hydrogen supply rate, the oxygen supply rate, and the cooling water flow rate in the control strategy is determined by the following rules:

[0037]

[0038] wherein a is an allocation proportion coefficient, x1(t) is hydrogen concentration, x2(t) is oxygen concentration, x4(t) is stack temperature, u1, u2, u3 are the hydrogen supply rate, the oxygen supply rate, and the cooling water flow rate in sequence, and t is the control input time point of the fuel cell system at a certain time.

[0039] Preferably, the deep learning model is used to approximately solve the allocation scheme in the control strategy, the input of the deep learning model includes the state variable x(t) and the reference power y ref (t) in step 2, the output is the allocation result of the control input u(t), and the deep learning model is trained according to the relationship between the dynamic model and the performance target.

[0040] Preferably, the training process of the deep learning model includes the following steps:

[0041] Generating training data of the fuel cell system under various dynamic load conditions, the training data including the state variable x(t), the reference power y ref (t), and the control input u(t);

[0042] Designing a loss function, defining the loss based on the deviation between the allocation result of the control input and the performance target;

[0043] Optimizing the model parameters by using the gradient descent method, the model parameters including the deep learning network weights and biases used to approximate the control strategy.

[0044] Preferably, the output of the deep learning model controls the input distribution result to be dynamically corrected by the following formula:

[0045] u(t) = u NN (t) + K(x(t) - x ref ),

[0046] Wherein, u(t) = [u1(t), u2(t), u3(t)] T , u(t) represents a control input vector, which is a hydrogen supply rate, an oxygen supply rate and a cooling water flow rate,

[0047] u NN (t) represents the control input distribution result output by the deep learning model;

[0048] K is a dynamic feedback gain matrix,

[0049] x(t) = [x1(t), x2(t), x3(t), x4(t)] T , x(t) represents a current state variable vector, which is a hydrogen concentration, an oxygen concentration, a membrane moisture content and a stack temperature in turn;

[0050] x ref represents a reference state variable vector, which is used to represent a target operating state of the system and is determined according to a load demand.

[0051] The application provides a control method of a fuel cell system. The following advantages are provided:

[0052] 1. The application introduces a dynamic feedback gain matrix, dynamically corrects the control input according to the deviation between the state variable and the reference state variable on the basis of the output of the deep learning model, realizes the adaptive adjustment of the control strategy to the real-time state of the fuel cell system, and improves the real-time performance and accuracy of the control input under dynamic load conditions.

[0053] 2. The application realizes efficient calculation of the control input under complex nonlinear dynamic conditions by combining the relationship between the dynamic model and the performance target to train the approximate solution of the control strategy distribution scheme by the deep learning model, and greatly reduces the calculation complexity and improves the accuracy of the control strategy.

[0054] 3. The application quantifies the nonlinear relationship between the output power and the state variables such as hydrogen concentration, oxygen concentration, membrane moisture content and stack temperature by defining an efficiency function, optimizes the electrochemical reaction efficiency of the fuel cell system, and improves the output power and energy utilization efficiency.

[0055] 4、The application realizes the priority dynamic distribution of the fuel cell system resources, and improves the operation efficiency and stability of the fuel cell system by dynamically adjusting the control input weight matrix, and optimizing the weight distribution according to the change of the control input sensitivity of the reference power. BRIEF DESCRIPTION OF DRAWINGS

[0056] Figure 1 The flow chart of the application. DETAILED DESCRIPTION

[0057] In order for those skilled in the art to understand the application scheme, the technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, other embodiments obtained by those skilled in the art without creative labor should be within the protection scope of the application.

[0058] The application will be described in detail below with reference to the drawings:

[0059] Embodiment:

[0060] Please refer to the accompanying drawings Figure 1 The embodiment of the application provides a control method of a fuel cell system, which comprises:

[0061] Step 1, a dynamic model of the fuel cell system is established, the state variables of the system are defined as the hydrogen concentration, the oxygen concentration, the membrane water content and the stack temperature, the control inputs are the hydrogen supply rate, the oxygen supply rate and the cooling water flow rate, and the dynamic behavior of the fuel cell system is described through the relationship between the state variables and the control inputs;

[0062] Step 2, the performance target of the fuel cell system is defined based on the dynamic model established in step 1, the performance target is used to measure the relationship between the control input and the output power of the fuel cell system, is defined through the difference between the output power and the reference power, and is comprehensively evaluated in combination with the resource consumption and distribution of the control input, the reference power is determined according to the load demand, and the output power is calculated through the state variables in the dynamic model;

[0063] Step 3, the control strategy of the fuel cell system is determined based on the performance target defined in step 2, the control strategy is designed according to the difference between the output power and the reference power in the performance target and the change relationship of the state variables in the dynamic model, and the distribution scheme of the control input is determined in combination with the influence of the hydrogen supply rate, the oxygen supply rate and the cooling water flow rate on the performance target;

[0064] Step 4, based on the control strategy determined in step 3, solve the allocation scheme of control input through a deep learning model, the input of which is the state variable defined in step 1, and the output is the specific allocation result of the control input, the deep learning model is trained by using the relationship between the state change in the dynamic model and the performance target;

[0065] Step 5, the control input allocation result calculated by the deep learning model in step 4 is transmitted to the actuator of the fuel cell system, and the hydrogen supply rate, oxygen supply rate and cooling water flow rate defined in step 1 are adjusted according to the allocation result to realize the dynamic change of the state variable of the fuel cell system, and the control of the fuel cell system is completed.

[0066] The advantage of step 1 is that by clearly defining the hydrogen concentration, oxygen concentration, membrane water content and stack temperature as state variables, and defining the control input as the hydrogen supply rate, oxygen supply rate and cooling water flow rate, the dynamic behavior of the fuel cell system is accurately described, which lays a scientific foundation for the design of subsequent performance targets and control strategies, and solves the problem of describing complex dynamic characteristics in traditional models.

[0067] The advantage of step 2 is that by combining the difference between the output power and the reference power, and the resource consumption and allocation of the control input, the performance target is comprehensively evaluated, the quantitative measurement of the control effect is realized, the output of the system is closely related to the control input, and it is ensured that the control strategy can meet the output power demand and optimize the resource utilization.

[0068] The advantage of step 3 is that by combining the performance target, the difference between the output power and the reference power, and the dynamic relationship of the state variable, the control strategy is designed and the allocation scheme of the control input is determined, so that the fuel cell system can quickly respond under different dynamic load conditions, and the problem of unreasonable allocation scheme and slow response of traditional control methods under complex dynamic conditions is solved.

[0069] The advantage of step 4 is that the deep learning model is used to solve the allocation scheme of the control input, which is trained by combining the dynamic model and the performance target, effectively handles the nonlinear characteristics and high-dimensional state changes of the fuel cell system, significantly reduces the computational complexity, improves the real-time performance and solving efficiency of the control strategy, and solves the problem of low computational efficiency of traditional analytical methods in high-dimensional nonlinear problems.

[0070] The advantage of step 5 is that by transmitting the control result output by the deep learning model to the actuator, the hydrogen supply rate, oxygen supply rate and cooling water flow rate are adjusted in real time to realize accurate control of the state variable of the fuel cell system, so that the control scheme can be truly implemented, and the system operation can meet the requirements of the performance target.

[0071] In step 1, the dynamic model of the fuel cell system includes the following relationships:

[0072]

[0073] wherein x1(t) is the hydrogen concentration, x2(t) is the oxygen concentration, x3(t) is the membrane moisture content, x4(t) is the stack temperature, u1, u2, u3 are the hydrogen supply rate, the oxygen supply rate and the cooling water flow rate in sequence, k1, k2, k3, k4 are the hydrogen consumption constant, the oxygen consumption constant, the moisture loss rate and the heat loss rate in sequence,

[0074] g(x1, x2, u3) represents the moisture generation function, and h(x1, x2, x3) represents the heat generation function.

[0075] The dynamic model accurately reflects the dynamic behavior of the fuel cell system by explicitly defining the change relationship of the hydrogen concentration x1, the oxygen concentration x2, the membrane moisture content x3 and the stack temperature x4, links the change of the key state variables to the corresponding input, and can accurately model the operating characteristics of the fuel cell, thereby providing a basis for the control strategy;

[0076] The model relationship includes the hydrogen consumption constant k1, the oxygen consumption constant k2, the moisture loss rate k3 and the heat loss rate k4, and introduces the moisture generation function and the heat generation function. The introduction of the parameters and the functions comprehensively reflects the electrochemical reaction, material transport and heat and mass transfer process inside the fuel cell system, and helps to capture the nonlinear characteristics and coupling relationship of the system;

[0077] The dynamic model combines the control input u1, u2, u3 with the dynamic change of the state variables, thereby providing a mathematical basis for the design of the control strategy. The dynamic model clearly identifies the influence path of the control input on each state variable, so that the control strategy can achieve fine adjustment of the operating state of the fuel cell system;

[0078] The moisture generation function and the heat generation function in the model explicitly show the influence of water management and heat management on the operation of the fuel cell system, which helps to maintain the dynamic balance of the membrane humidity and the stack temperature under different load conditions, and improves the adaptability and robustness of the fuel cell system.

[0079] The performance target in step 2 is defined by the following relationship:

[0080]

[0081] wherein J is the performance target, t0 is the initial time of the fuel cell system control, t f is the end time of the fuel cell system control, x(t) is the state variable vector of the fuel cell system, and u(t) is the control input vector of the fuel cell system

[0082] q(x(t), u(t)) represents the difference between the output power y(t) and the reference power y ref (t),

[0083] r(u(t)) represents the cost of control input, and R is a diagonal weight matrix.

[0084] The performance target quantifies the difference between the output power y(t) and the reference power y ref (t) to provide an evaluation standard for the effectiveness of fuel cell system control. Through quantitative means, it can be directly determined whether the control strategy achieves accurate tracking of the target power, ensuring that the power output of the fuel cell system meets the load demand.

[0085] The performance target focuses on the accuracy of the output power, and the cost of control input reflects the efficiency of resource utilization. By adding the cost of control input, the problem of resource waste caused by simply pursuing power output is avoided, and the overall operation efficiency of the fuel cell system is improved by balancing system performance and resource conservation.

[0086] The reference power in the performance target changes dynamically according to the load demand, allowing the fuel cell system to flexibly adjust the control strategy under different operating conditions, ensuring the matching of system output and actual load demand, and improving the adaptability and operation stability of the fuel cell system.

[0087] The performance target provides an optimization direction for the control strategy by combining the dynamic relationship between the state variables and the control input. By minimizing the performance target, the control strategy can dynamically adjust the input parameters under the trade-off between output power and control input, ensuring accurate power output and minimum energy consumption.

[0088] The performance target is defined in the form of a dynamic function of state variables and control input, providing a mathematical framework for model-based optimization methods.

[0089] The output power in the performance target is calculated by the following formula:

[0090] y(t) = n(x1(t), x2(t), x3(t), x4(t)),

[0091] where y(t) represents the output power of the fuel cell system,

[0092] n(x1(t), x2(t), x3(t), x4(t)) is an efficiency function,

[0093] x1(t) is the hydrogen concentration, x2(t) is the oxygen concentration, x3(t) is the membrane water content, and x4(t) is the stack temperature, representing the output power efficiency of the fuel cell system under different state variables.

[0094] By the definition of the output power formula, the output power is directly related to the key state variables of the fuel cell system, i.e. hydrogen concentration, oxygen concentration, membrane water content and stack temperature, which captures the core dynamic characteristics of the fuel cell system, i.e. electrochemical reaction, gas diffusion and heat and mass transfer, and can accurately characterize the output power of the system;

[0095] The efficiency function explicitly describes how the output power efficiency of the fuel cell system depends on the hydrogen concentration, oxygen concentration, membrane water content and stack temperature, and through the function, the control strategy can be dynamically adjusted to optimize the power efficiency;

[0096] Through the definition of the efficiency function, the control strategy can understand the contribution of each state variable to the output power, providing a clear physical basis for optimizing the state variables and guiding the adjustment of the control input;

[0097] The state variable dependent performance of the output power formula can reflect the dynamic characteristics of the fuel cell system in real time, and through real-time monitoring and updating, the control system can quickly adjust the hydrogen supply rate, oxygen supply rate and cooling water flow rate according to the actual situation to meet the power demand under dynamic load conditions;

[0098] The efficiency function is used for the calculation of the output power, providing core input for the definition of performance targets and the design of control strategies, and through the function, the control system can comprehensively evaluate the contribution weight of each state variable, prioritize the optimization of key variables, and improve the output power under resource-limited conditions, thereby improving the energy utilization efficiency of the fuel cell.

[0099] The efficiency function n(x1(t), x2(t), x3(t), x4(t)) is represented by the following formula:

[0100]

[0101] where k e represents the efficiency factor, c e is the compensation coefficient of the stack temperature,

[0102] x1·x2 represents the coupling effect of hydrogen and oxygen concentrations, reflecting the influence of the strength of the fuel cell electrochemical reaction on efficiency,

[0103] x4+c e represents the comprehensive influence of the stack temperature on efficiency, x1(t) is the hydrogen concentration, x2(t) is the oxygen concentration, x3(t) is the membrane water content, and x4(t) is the stack temperature.

[0104] The efficiency function formula explicitly describes the dynamic relationship between fuel cell efficiency and hydrogen concentration, oxygen concentration, membrane water content and stack temperature, and through quantifying the influence of state variables on efficiency, it can reflect the efficiency variation law of the fuel cell system under different operating conditions, providing a basis for optimization control;

[0105] x1·x2 in the efficiency function represents the coupling effect of hydrogen and oxygen concentration, directly reflects the influence of electrochemical reaction intensity on efficiency, and through the introduction, the supply rate of hydrogen and oxygen can be dynamically adjusted to keep the electrochemical reaction rate within the optimal range at all times;

[0106] x4+c in the efficiency function e describes the comprehensive influence of stack temperature on efficiency, where c e temperature compensation coefficient, used to adjust the nonlinear effect of stack temperature. High stack temperature will cause water loss and membrane drying, and low stack temperature will reduce the electrochemical reaction rate. The influence of stack temperature on fuel cell efficiency can be accurately quantified to guide the adjustment of cooling water flow rate and ensure that the system operates within the optimal temperature range;

[0107] By introducing the efficiency factor k e , the efficiency function can adapt to the characteristics of different fuel cell systems, and the specific system parameters are determined by experiments to make the function fit the actual operating conditions, providing a physical basis for the optimization of control inputs, and improving the universality of the model;

[0108] The efficiency function quantifies the comprehensive influence of multiple key variables on efficiency through a simple and effective form, and the clear expression method facilitates the control strategy to directly use the efficiency function as the optimization objective, and dynamically adjusts the supply rates of hydrogen and oxygen and the cooling water flow rate to improve power output and fuel utilization efficiency.

[0109] The elements of the weight matrix R of control inputs are dynamically adjusted according to the operating conditions of the fuel cell system, and the adjustment method is:

[0110]

[0111] where r i is the i-th diagonal element of the control input weight matrix R,

[0112] k r is the adjustment coefficient, represents the sensitivity of the reference power to the control input u i .

[0113] The dynamic adjustment method of the weight matrix R realizes the accurate allocation and dynamic adaptability of control resources by combining the sensitivity of the reference power to the control input and the adjustment coefficient, can optimize the priority of the control input in real time under different load demands and operating conditions, effectively improves the operating efficiency and stability of the fuel cell system, and at the same time maintains a low calculation complexity and simple engineering implementation. The dynamic adjustment of the weight matrix provides an efficient and flexible solution for resource allocation of complex nonlinear systems, enabling the fuel cell system to achieve optimal performance under dynamic conditions.

[0114] The allocation scheme of the hydrogen supply rate, the oxygen supply rate, and the cooling water flow rate in the control strategy is determined by the following rules:

[0115]

[0116] wherein a is an allocation proportionality coefficient, x1(t) is the hydrogen concentration, x2(t) is the oxygen concentration, x4(t) is the stack temperature, u1, u2, u3 are the hydrogen supply rate, the oxygen supply rate, and the cooling water flow rate in sequence, and t is the control input time point of the fuel cell system at a certain time.

[0117] The allocation rules of the control strategy guarantee the resource supply proportion balance of the fuel cell system through the dynamic coupling of the hydrogen, oxygen, and cooling water supply rates with the state variables, optimize the electrochemical reaction efficiency and the adjustment ability of the stack temperature, and provide support for the flexibility of the allocation strategy through the introduction of the allocation proportionality coefficient a, so as to adapt to the resource demand under different working conditions, improve the operation efficiency and stability of the fuel cell system, realize the efficient adjustment of the real-time control input with low computational complexity, and be suitable for the control demand under complex dynamic load conditions.

[0118] The deep learning model is used to approximately solve the allocation scheme in the control strategy, the input of the deep learning model includes the state variables x(t) and the reference power y ref (t) in step 2, the output is the allocation result of the control input u(t), and the deep learning model is trained according to the relationship between the dynamic model and the performance target.

[0119] The training process of the deep learning model includes the following steps:

[0120] Training data of the fuel cell system under multiple dynamic load conditions is generated, the training data includes the state variables x(t), the reference power y ref (t), and the control input u(t);

[0121] A loss function is designed, and the loss is defined based on the deviation between the allocation result of the control input and the performance target;

[0122] The gradient descent method is used to optimize the model parameters, and the model parameters include the deep learning network weights and biases used to approximate the control strategy.

[0123] The approximate solution of the control strategy allocation scheme and the training process design of the deep learning model solve the problems of low efficiency, poor real-time performance and insufficient adaptability of traditional methods in complex nonlinear problems. The deep learning model can use multi-dimensional data of state variables, reference power and control input to dynamically adjust the control input allocation result. At the same time, through loss function optimization, the model output is close to the performance target requirement. The gradient descent method in the training process further ensures the rapid convergence and engineering realizability of the model. Overall, the scheme provides an efficient, flexible and accurate control strategy solving method under complex dynamic load conditions, significantly improving the operation performance and control quality of the fuel cell system.

[0124] The output control input allocation result of the deep learning model is dynamically corrected by the following formula:

[0125] u(t) = u NN (t) + K(x(t) - x ref ),

[0126] Where u(t) = [u1(t), u2(t), u3(t)] T , u(t) represents the control input vector, which is the hydrogen supply rate, oxygen supply rate and cooling water flow rate,

[0127] u NN (t) represents the control input allocation result output by the deep learning model;

[0128] K is a dynamic feedback gain matrix,

[0129] x(t) = [x1(t), x2(t), x3(t), x4(t)] T , x(t) represents the current state variable vector, which is the hydrogen concentration, oxygen concentration, membrane moisture content and stack temperature in turn;

[0130] x ref represents the reference state variable vector, which is used to represent the target operating state of the system and is determined according to the load demand.

[0131] The dynamic correction formula introduces a dynamic feedback gain matrix, combines the deviation of the current state variable and the reference state variable, and dynamically adjusts the control input allocation result output by the deep learning model, so that the control input can quickly respond to the real-time changes of the fuel cell system operating state, effectively compensating for the deviation of the deep learning model under some unexpected conditions;

[0132] By the dynamic feedback adjustment formula, the hydrogen supply rate, oxygen supply rate and cooling water flow rate can be precisely controlled. The feedback gain matrix provides sensitivity information between state variables and control inputs, so that the corrected control inputs can better match the operation target of the fuel cell, ensuring that the output power and system efficiency meet the expectations;

[0133] The adjustment of the dynamic feedback gain matrix can be flexibly optimized according to the actual working conditions and load requirements. By designing appropriate feedback gains, the control inputs corresponding to key variables are preferentially adjusted, thereby realizing efficient allocation and priority optimization of control resources and meeting the operation requirements under different working conditions;

[0134] The dynamic correction formula uses the deviation between state variables and reference state variables for feedback, which can quickly correct the situation where the system deviates from the target state. Especially in the case of sudden load changes and operating disturbances, it effectively alleviates the risk of system instability and ensures that the fuel cell system always operates in a stable and efficient state;

[0135] The deep learning model provides the ability to solve complex nonlinear dynamic problems, but there are errors in some extreme working conditions. The dynamic correction formula adds feedback terms, which significantly improves the robustness and reliability of the overall control strategy while maintaining the computational efficiency of the deep learning model.

[0136] Although embodiments of the present application have been shown and described, it will be understood by those having ordinary skill in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for controlling a fuel cell system, characterized in that: include: Step 1: Establish a dynamic model of the fuel cell system. Define the system's state variables as hydrogen concentration, oxygen concentration, membrane moisture content, and stack temperature. Define the control inputs as hydrogen supply rate, oxygen supply rate, and cooling water flow rate. Describe the dynamic behavior of the fuel cell system through the relationship between the state variables and the control inputs. The dynamic model of the fuel cell system includes the following relationship: Where x1(t) is the hydrogen concentration, x2(t) is the oxygen concentration, x3(t) is the membrane moisture content, x4(t) is the stack temperature, u1, u2, and u3 are the hydrogen supply rate, oxygen supply rate, and cooling water flow rate, respectively; k1, k2, k3, and k4 are the hydrogen consumption constant, oxygen consumption constant, moisture loss rate, and heat loss rate, respectively. g(x1, x2, u3) represents the moisture generation function, and h(x1, x2, x3) represents the heat generation function; Step 2: Define the performance target of the fuel cell system based on the dynamic model established in Step 1. The performance target is used to measure the relationship between the control input and the output power of the fuel cell system. It is defined by the difference between the output power and the reference power. The performance target is comprehensively evaluated in combination with the resource consumption and allocation of the control input. The reference power is determined according to the load demand, and the output power is calculated using the state variables in the dynamic model. The performance goal is defined by the following relationship: Where J is the performance target, t0 is the initial time of fuel cell system control, t f is the end time of the fuel cell system control, x(t) is the state variable vector of the fuel cell system, and u(t) is the control input vector of the fuel cell system. q(x(t), u(t)) represents the difference between the output power y(t) and the reference power y ref (t) the difference between r(u(t)) represents the cost of the control input, and R is the diagonal weight matrix; The output power in the performance target is calculated using the following formula: y(t)=n(x1(t),x2(t),x3(t),x4(t)), Where y(t) represents the output power of the fuel cell system, n(x1(t), x2(t), x3(t), x4(t)) is the efficiency function, x1(t) is the hydrogen concentration, x2(t) is the oxygen concentration, x3(t) is the membrane moisture content, and x4(t) is the stack temperature, which represents the output power efficiency of the fuel cell system under different state variables; The efficiency function n(x1(t), x2(t), x3(t), x4(t)) is expressed by the following formula: Among them, k e represents the efficiency factor, c e is the compensation coefficient of stack temperature, x1·x2 represents the coupling effect of hydrogen and oxygen concentrations, reflecting the effect of the electrochemical reaction intensity of the fuel cell on the efficiency. x4+c e represents the comprehensive effect of stack temperature on efficiency, x1(t) is the hydrogen concentration, x2(t) is the oxygen concentration, x3(t) is the membrane moisture content, and x4(t) is the stack temperature; Step 3: Determine the control strategy of the fuel cell system based on the performance target defined in Step 2. The control strategy is designed based on the difference between the output power and the reference power in the performance target and the changing relationship between the state variables in the dynamic model. The control input allocation scheme is determined by combining the influence of the hydrogen supply rate, oxygen supply rate and cooling water flow rate on the performance target. Step 4: Based on the control strategy determined in Step 3, the control input allocation scheme is solved using a deep learning model. The input of the deep learning model is the state variables defined in Step 1, and the output is the specific allocation result of the control input. The deep learning model is trained by leveraging the relationship between state changes and performance objectives in the dynamic model. Step 5: The control input allocation result calculated by the deep learning model in step 4 is transmitted to the actuator of the fuel cell system. The hydrogen supply rate, oxygen supply rate and cooling water flow rate defined in step 1 are adjusted according to the allocation result to realize the dynamic change of the state variables of the fuel cell system and complete the control of the fuel cell system.

2. The method for controlling a fuel cell system according to claim 1, wherein: The elements of the control input weight matrix R are dynamically adjusted according to the operating conditions of the fuel cell system, and the adjustment method is: i＝1，2，3， Among them, r i is the i-th diagonal element of the control input weight matrix R, k r is the adjustment coefficient, Indicates the reference power to the control input u i sensitivity.

3. The control method of a fuel cell system according to claim 1, characterized in that: The allocation scheme of hydrogen supply rate, oxygen supply rate and cooling water flow rate in the control strategy is determined by the following rules: Where a is the distribution ratio coefficient, x1(t) is the hydrogen concentration, x2(t) is the oxygen concentration, x4(t) is the stack temperature, u1, u2, and u3 are the hydrogen supply rate, oxygen supply rate, and cooling water flow rate, respectively, and t is the control input time point of the fuel cell system at a certain moment.

4. The method for controlling a fuel cell system according to claim 1, wherein: The deep learning model approximates the allocation scheme in the control strategy. The input of the deep learning model includes the state variable x(t) and the reference power y in step 2. ref (t), the output is the distribution result of the control input u(t), and the deep learning model is trained according to the relationship between the dynamic model and the performance target.

5. The method for controlling a fuel cell system according to claim 1, wherein: The training process of the deep learning model includes the following steps: Generate training data of the fuel cell system under various dynamic load conditions, including state variables x(t), reference power y ref (t) and control input u(t); Design a loss function that defines the loss based on the deviation between the distribution of control inputs and the performance target; Gradient descent is used to optimize model parameters, including the weights and biases of the deep learning network used to approximate the control strategy. The output control input allocation result of the deep learning model is dynamically modified by the following formula: u(t)=u NN (t)+K(x(t)-x ref ), Where, u(t0=[u1(t),u2(t),u3(t)] T , u(t) represents the control input vector, which is the hydrogen supply rate, oxygen supply rate and cooling water flow rate, u NN (t) represents the control input allocation result output by the deep learning model; K is the dynamic feedback gain matrix, x(t)=[x1(t), x2(t), x3(t), x4(t)] T , x(t) represents the current state variable vector, which is hydrogen concentration, oxygen concentration, membrane moisture content and stack temperature in turn; x ref Represents the reference state variable vector, which is used to represent the target operating state of the system and is determined according to the load demand.

Images