An adaptive control method, system, and electronic equipment for a water electrolysis hydrogen production system.
By employing adaptive control methods, multiphysics coupling prediction models, and online fine-tuning techniques, the response hysteresis and coupling interference problems of the water electrolysis hydrogen production system under renewable energy fluctuations were solved, achieving efficient and safe operation of the system.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA THREE GORGES CORPORATION
- Filing Date
- 2026-05-08
- Publication Date
- 2026-07-24
AI Technical Summary
The existing control strategies for water electrolysis hydrogen production systems are difficult to adapt to the volatility of renewable energy, resulting in response lag and mismatch, control dimension fragmentation and strong coupling interference, and lack of time-varying adaptability, which affects the safety and energy efficiency of the system.
An adaptive control method is adopted, which uses a multi-physics coupled prediction model to perform rolling prediction, constructs a multi-objective optimization problem, adjusts the control input in real time, and combines recursive least squares method or extended Kalman filter for online fine-tuning to achieve coordinated control of electrochemical, temperature field and gas-fluid processes.
Maintaining optimal system performance throughout its entire lifecycle improves adaptability to renewable energy fluctuations and enhances system security and energy efficiency.
Smart Images
Figure CN122446264A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of hydrogen production technology, and specifically relates to an adaptive control method, system and electronic equipment for a water electrolysis hydrogen production system. Background Technology
[0002] Currently, the control strategies for commercial water electrolysis hydrogen production systems are generally quite basic and difficult to adapt to the direct coupling requirements of fluctuating renewable energy sources. Most mainstream solutions employ traditional proportional-integral-derivative (PID) control combined with pre-set logic rules.
[0003] A typical existing technical solution is a combination of power point tracking (PPT) and current mapping with independent closed-loop control. That is, the system first monitors the instantaneous output power of the renewable energy source, and then maps this power value to the DC current setpoint of the electrolyzer using a fixed or piecewise linear lookup table. Temperature control of the electrolyzer is handled by an independent PID loop, which uses the cooling water outlet temperature or the average temperature at a fixed point in the electrolyzer as feedback to adjust the speed of the cooling water pump or the opening of the three-way valve to maintain the temperature at the setpoint. The pressure of the gas-liquid separator is maintained constant by adjusting the back pressure valve through another independent PID controller.
[0004] Another slightly improved existing technique might introduce feedforward compensation. For example, when the current setpoint changes abruptly, a feedforward signal is injected into the temperature control loop to adjust the cooling water flow rate in advance. However, this feedforward relationship is usually based on simple linear assumptions or empirical formulas, and the parameters are fixed.
[0005] The main drawback of existing technology is that: Hysteresis and Mismatch: PID and rule-based control based on current power or deviation is essentially a hysteresis response. When faced with random and drastic fluctuations in wind and solar power ranging from seconds to minutes, this control method cannot predict future trends. Its response speed is severely mismatched with the fluctuation speed, causing large oscillations in system operating conditions (such as current and temperature), frequently triggering safety limits, forcing the system to reduce load or shut down, and severely limiting the absorption rate of renewable energy.
[0006] The fragmented control dimension and strong coupling interference: Multiple physical processes, such as electricity, heat, and gas (fluid), are strongly coupled within the electrolyzer. Existing technologies decouple these processes into multiple independent single-loop control, neglecting the coupling effect. For example, when the current is increased to track a sudden power surge, a large amount of reaction heat is instantaneously generated. The independent, lagging temperature control loop cannot synchronously counteract this thermal disturbance, easily causing localized hot spots in the electrolyzer and leading to irreversible damage such as membrane dry burning and catalyst sintering. The existing approach sacrifices the overall safety and optimization of the system.
[0007] Lack of time-varying adaptability: Control system parameters (such as PID gain, feedforward coefficients, and mapping tables) and potential simple model parameters are usually fixed after system commissioning. However, the performance of the electrolyzer drifts with operating time (e.g., catalyst activity decay, decrease in membrane ionic conductivity, and changes in heat exchange efficiency due to fouling in the flow channels). Fixed control strategies cannot adapt to this slow time-varying characteristic, causing control performance to gradually deteriorate over operating time, and neither energy efficiency nor safety can be guaranteed in the long term.
[0008] Therefore, there is an urgent need in this field for an advanced control method that can predict ahead, coordinate multiple variables, and has self-learning capabilities to solve the challenges of dynamic adaptation and safety optimization in renewable energy hydrogen production scenarios.
[0009] In summary, the problem to be solved is how to provide a control method that can coordinate multiple variables and has self-learning capabilities to address the dynamic adaptation problem in renewable energy hydrogen production scenarios. Summary of the Invention
[0010] To address the aforementioned issues, this application provides an adaptive control method, system, storage medium, and electronic device for a water electrolysis hydrogen production system, which maintains optimal performance throughout its entire lifecycle through adaptive control.
[0011] In a first aspect, embodiments of this application provide an adaptive control method for a water electrolysis hydrogen production system, the method comprising: Based on the current state vector and the assumed future control input sequence, the prediction model is invoked to perform forward simulation rolling prediction, resulting in multiple prediction data, including: predicted electrochemical data, predicted temperature field data, and predicted gas-fluid data. The coupled model library used by the prediction model includes: electrochemical voltage-current relationship sub-model, non-uniform temperature field dynamic sub-model, and gas-fluid network sub-model. Under dynamic constraints based on multiple constraints, a multi-objective optimization problem is constructed. The objective function of the multi-objective optimization problem is to maximize the cumulative sum of the composite operating efficiency index function in the prediction time domain in the next N steps, where N is a natural number greater than or equal to 1. In the rolling time domain, the multi-objective optimization problem is optimized and solved to obtain the set of optimal control input sequences within a preset control period; The first optimal control input sequence in the optimal control input sequence set is used as the actual control command, and the actual control command is sent to the actuator to execute the actual control command and obtain multiple actual data, wherein the multiple actual data include: actual electrochemical data, actual temperature field data and actual gas-flow data; Compare any one of the multiple predicted data with its matching actual data to obtain the corresponding comparison result; Based on the comparison results and the preset method, multiple different time-varying parameters in the key time-varying parameter set in the prediction model are fine-tuned online to perform adaptive control. The preset method includes: recursive least squares method or extended Kalman filter method.
[0012] Optionally, before invoking the prediction model to perform forward simulation rolling prediction based on the current state vector and the assumed future control input sequence to obtain multiple prediction data, the method further includes: Acquire multi-source heterogeneous data and state data, wherein the multi-source heterogeneous data includes: first correlation data associated with electrical quantities, second correlation data associated with thermodynamic quantities, third correlation data associated with fluid and gas quantities, and fourth correlation data associated with environmental quantities; the state data includes: membrane water content and catalyst surface temperature; The multi-source heterogeneous data and the state data are fused together to obtain the current state vector.
[0013] Optionally, the step of optimizing the multi-objective optimization problem in the rolling time domain includes: Obtain the multi-objective optimization problem, wherein the multi-objective optimization problem is a constrained nonlinear optimization problem; The multi-objective optimization problem is transformed into a corresponding sequential quadratic programming problem; or... The multi-objective optimization problem is transformed into a form solvable by the interior point method.
[0014] Optionally, before fusing the multi-source heterogeneous data and the state data to obtain the current state vector, the method further includes: The first associated data is collected in real time, including: DC bus current, total voltage, instantaneous active power from the converter and its rate of change; The second associated data is collected in real time. The second associated data includes: a group of temperature sensors arranged sequentially at the inlet, outlet and preset typical positions of the electrolytic cell; the inlet temperature of the main coolant pipeline and key branches; and the flow rate of the main coolant pipeline and key branches. The third associated data is collected in real time, including: electrolyte circulation pump frequency and outlet pressure, hydrogen / oxygen online purity analyzer and mass flow meter readings, and gas-liquid separator pressure; The fourth associated data is collected in real time, and the fourth associated data includes: ambient temperature and humidity.
[0015] Optionally, before constructing the multi-objective optimization problem under dynamic constraints based on multiple constraints, the method further includes: The various constraints are set, including: current upper and lower limit constraints, absolute temperature and temperature difference constraints, pressure constraints, valve opening constraints, and state equation constraints defined by the model itself.
[0016] Optionally, before fusing the multi-source heterogeneous data and the state data to obtain the current state vector, the method further includes: The first associated data is collected in real time, including: DC bus current, total voltage, instantaneous active power from the converter and its rate of change; The second associated data is collected in real time. The second associated data includes: a group of temperature sensors arranged sequentially at the inlet, outlet and preset typical positions of the electrolytic cell; the inlet temperature of the main coolant pipeline and key branches; and the flow rate of the main coolant pipeline and key branches. The third associated data is collected in real time, including: electrolyte circulation pump frequency and outlet pressure, hydrogen / oxygen online purity analyzer and mass flow meter readings, and gas-liquid separator pressure; The fourth associated data is collected in real time, and the fourth associated data includes: ambient temperature and humidity.
[0017] Secondly, embodiments of this application provide an adaptive control system for a water electrolysis hydrogen production system based on multi-physics field coupling, the system comprising: The prediction module is used to call the prediction model to perform forward simulation rolling prediction based on the current state vector and the assumed future control input sequence, and obtain multiple prediction data. The multiple prediction data include: predicted electrochemical data, predicted temperature field data and predicted gas-fluid data. The coupling model library used by the prediction model includes: electrochemical voltage-current relationship sub-model, non-uniform temperature field dynamic sub-model and gas-fluid network sub-model. The module is used to construct a multi-objective optimization problem under dynamic constraints based on multiple constraints. The objective function of the multi-objective optimization problem is to maximize the cumulative sum of the composite operating efficiency index function in the prediction time domain in the next N steps, where N is a natural number greater than or equal to 1. The optimization solution module is used to optimize and solve the multi-objective optimization problem in the rolling time domain to obtain the set of optimal control input sequences within a preset control period; The processing module is used to take the first optimal control input sequence in the optimal control input sequence set as the actual control instruction, and to send the actual control instruction to the actuator so that the actuator can execute the actual control instruction to obtain multiple actual data, wherein the multiple actual data include: actual electrochemical data, actual temperature field data and actual gas-flow data; The comparison module is used to compare any one of the multiple predicted data with its matching actual data to obtain the corresponding comparison result; The online fine-tuning module is used to perform online fine-tuning of multiple different time-varying parameters in the key time-varying parameter set of the prediction model based on the comparison results and preset methods, so as to perform adaptive control.
[0018] Optionally, the system further includes: The acquisition module is used to: acquire multi-source heterogeneous data and state data before the rolling prediction of forward simulation based on the current state vector and the assumed future control input sequence, and to obtain multiple prediction data. The multi-source heterogeneous data includes: first correlation data associated with electrical quantities, second correlation data associated with thermodynamic quantities, third correlation data associated with fluid and gas quantities, and fourth correlation data associated with environmental quantities; the state data includes: membrane water content and catalyst surface temperature. The fusion processing module is used to fuse the multi-source heterogeneous data and the state data to obtain the current state vector.
[0019] Optionally, the optimization solution module is specifically used for: Obtain the multi-objective optimization problem, wherein the multi-objective optimization problem is a constrained nonlinear optimization problem; The multi-objective optimization problem is transformed into a corresponding sequential quadratic programming problem; or... The multi-objective optimization problem is transformed into a form solvable by the interior point method.
[0020] Thirdly, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the method of the first aspect.
[0021] Fourthly, an electronic device is provided, including a memory and a processor, wherein the memory stores executable code, and the processor executes the executable code to implement the method of the first aspect.
[0022] Compared with the prior art, this application has the following advantages: It can maintain optimal performance throughout its entire lifecycle through adaptive control.
[0023] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description
[0024] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0025] Figure 1 A flowchart of an adaptive control method for a water electrolysis hydrogen production system according to an embodiment of this application is shown; Figure 2 The diagram shows the control architecture of the adaptive control method for the water electrolysis hydrogen production system according to an embodiment of this application. Figure 3 A diagram of a dedicated heterogeneous hardware acceleration platform architecture is shown. Figure 4 A schematic diagram of the structure of an adaptive control system 200 for a water electrolysis hydrogen production system according to an embodiment of this application is shown. Detailed Implementation
[0026] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0027] This application provides an adaptive control method and system for a water electrolysis hydrogen production system, a computer-readable medium, and an electronic device, which will be described below with reference to the accompanying drawings.
[0028] The adaptive control method for a water electrolysis hydrogen production system based on multiphysics coupling provided in this application lies in constructing a real-time rolling "electric-thermal-gas" multiphysics coupling predictive optimization control framework. A hierarchical predictive control architecture is adopted. The bottom layer is the data acquisition and execution layer, the middle layer is the core multiphysics coupling predictive model and rolling optimizer, and the upper layer is the dynamic objective and constraint management module. In each sampling period, the optimizer predicts the future system behavior in the time domain based on the model and solves a constrained multi-objective optimization problem, applying the first element of the calculated optimal control sequence to the controlled object.
[0029] Please refer to Figure 1 It illustrates a flowchart of an adaptive control method for a water electrolysis hydrogen production system provided in some embodiments of this application, such as... Figure 1 As shown, the adaptive control method for a water electrolysis hydrogen production system may include the following steps: Step S101: Based on the current state vector and the assumed future control input sequence, call the prediction model to perform forward simulation rolling prediction to obtain multiple prediction data, including: predicted electrochemical data, predicted temperature field data and predicted gas-fluid data. The coupling model library used by the prediction model includes: electrochemical voltage-current relationship sub-model, non-uniform temperature field dynamic sub-model and gas-fluid network sub-model.
[0030] In practical applications, the algorithms used in the aforementioned electrochemical voltage-current relationship sub-model, non-uniform temperature field dynamic sub-model, and gas-fluid network sub-model are described in detail below: (1) Electrochemical voltage-current relationship sub-model: Formula (1); In the formula, V_cell: the actual operating voltage (V) of a single cell in the electrolyzer; E_rev: Reversible open-circuit voltage (V), i.e., the lowest electrolysis voltage according to thermodynamic theory; η_act: Activation overpotential (V), the additional voltage required to overcome the activation energy of an electrochemical reaction; η_ohm: Ohmic overpotential (V), the voltage required to overcome the resistance of the components inside the electrolytic cell.
[0031] Formula (2); In the formula, E_rev,0: the reversible voltage under standard conditions (typically 1.229 V); ΔS: Standard molar entropy change of reaction (J / (mol·K)); n: The number of electrons transferred in the reaction (n=2 for water electrolysis); F: Faraday constant (approximately 96485 C / mol); T: The actual absolute operating temperature (K) of the electrolyzer; T_ref: Reference absolute temperature (usually 298.15 K); R: Ideal gas constant (8.314 J / (mol·K)); p_H2: The actual partial pressure of hydrogen on the cathode side (atm or bar); p_O2: The actual partial pressure of oxygen on the anode side (atm or bar); a_H2O: The activity of liquid water (usually approximately 1 in pure water or dilute solutions).
[0032] Formula (3); In the formula, α: charge transfer coefficient (symmetry factor), which usually takes a value between 0 and 1; i: Actual operating current density (A / cm²); i_ex: Exchange current density (A / cm²), which is a function that strongly correlates catalyst activity with temperature; asinh: Inverse hyperbolic sine function.
[0033] Formula (4); In the formula, R_mem: the equivalent area resistivity (Ω·cm²) of the proton exchange membrane (PEM) or diaphragm, is a time-varying function of membrane water content, temperature and aging state; R_ele: Electronic conductivity resistance of the electrode and current collector (Ω·cm²); R_c: Contact resistance between components (Ω·cm²).
[0034] (2) Dynamic sub-model of non-uniform temperature field (discretized using lumped parameter method): The reaction zone of the electrolyzer is discretized in three dimensions to generate M control volumes. For the j-th control volume: Formula (5); In the formula, C_j: the equivalent heat capacity (J / K) of the j-th control volume; T_j: Real-time temperature (K) of the j-th control volume; t: time variable (s); Q_gen, j: Internal heat generation rate (W) of the j-th control volume; Q_cool, j: The heat dissipation rate (W) of the j-th control unit carried away by the coolant; Q_loss, j: The rate (W) of heat loss from the j-th control volume to the external environment; λ_jk: Equivalent thermal conductivity (W / K) between the j-th control volume and the adjacent k-th control volume; T_k: Real-time temperature (K) of the kth adjacent control volume.
[0035] Formula (6); In the formula, Heat generation includes ohmic joule heat and heat of irreversible reaction.
[0036] i_j: Local current density of the j-th control volume (A / cm²); A_j: Effective electrochemical reaction area (cm²) of the j-th control body; V_cell, j: Local voltage (V) of the j-th control cell; E_rev, j: Local reversible voltage (V) of the j-th control volume; η_act, j: Local activation overpotential (V) of the j-th control body; ζ (Zeta): Thermal neutral voltage conversion factor (or irreversible heat dissipation factor), used to characterize the proportion of the activation overpotential converted into heat energy.
[0037] Formula (7); In the formula, : Convection heat dissipation item.
[0038] h_j(v_cool): Local convective heat transfer coefficient (W / (m²·K)), which is a function of the local coolant flow rate v_cool; A_s, j: Effective convective heat transfer area (m²) between the j-th control volume and the coolant; T_coolant, j: Real-time temperature (K) of the coolant flowing through the j-th control unit.
[0039] (3) Gas-fluid network sub-model: Formula (8); In the formula, ρ_l: density of liquid water / electrolyte in the gas-liquid separator (kg / m³). V_sep: Effective gas volume (m³) of the gas-liquid separator. K: The bulk modulus of the gas or the proportionality constant in the equation of state; P_sep: Real-time pressure (Pa or bar) of the gas-liquid separator. β_in: The mass or volume fraction ratio coefficient of hydrogen in the inlet gas-liquid mixture; F_H2, in: Hydrogen mass flow rate (kg / s) flowing into the gas-liquid separator. F_H2, out: Mass flow rate of pure hydrogen gas (kg / s) flowing out of the gas-liquid separator. f_valve: The nonlinear characteristic function of the back pressure valve in terms of flow rate, pressure, and opening degree; u_valve: The actual control opening command of the back pressure valve (0~100%).
[0040] The pressure dynamics of the separator are described by the above formula (8), where, This is the flow-pressure-opening characteristic function of the back pressure valve.
[0041] It should be noted that the above prediction model is a high-fidelity multiphysics coupling prediction model.
[0042] High-fidelity multiphysics coupled prediction model: A coupled prediction model library integrating an electrochemical Gibbs-activation-ohmic overpotential model, a dynamic model of a non-uniform temperature field based on control volume discreteness, and a gas-fluid network model has been constructed. This prediction model can more accurately simulate the complex coupled processes inside the electrolyzer, providing reliable future state predictions for nonlinear model predictive control (MPC), and is the foundation for achieving precise feedforward control.
[0043] In one example, before invoking the prediction model to perform forward simulation and rolling prediction based on the current state vector and the assumed future control input sequence to obtain multiple prediction data, the adaptive control method for the water electrolysis hydrogen production system provided in this application embodiment further includes the following steps: Acquire multi-source heterogeneous data and state data. The multi-source heterogeneous data includes: first correlation data related to electrical quantities, second correlation data related to thermodynamic quantities, third correlation data related to fluid and gas quantities, and fourth correlation data related to environmental quantities. The state data includes: membrane water content and catalyst surface temperature. The current state vector is obtained by fusing multi-source heterogeneous data and state data.
[0044] In one example, before fusing multi-source heterogeneous data and state data to obtain the current state vector, the adaptive control method for the water electrolysis hydrogen production system provided in this application embodiment may further include the following steps: Real-time acquisition of the first associated data, which includes: DC bus current, total voltage, instantaneous active power from the converter and its rate of change; The second correlation data is collected in real time. The second correlation data includes: temperature sensor groups arranged sequentially at the inlet, outlet and preset typical positions of the electrolytic cell, inlet temperature of the main coolant pipeline and key branches, and flow rate of the main coolant pipeline and key branches. Real-time acquisition of third-party correlation data, including: electrolyte circulation pump frequency and outlet pressure, hydrogen / oxygen online purity analyzer and mass flow meter readings, and gas-liquid separator pressure; The fourth correlation data is collected in real time, including ambient temperature and humidity.
[0045] Step S102: Under dynamic constraints based on multiple constraints, construct a multi-objective optimization problem, wherein the objective function of the multi-objective optimization problem is: to maximize the cumulative sum of the composite operating efficiency index function in the prediction time domain in the next N steps, where N is a natural number greater than or equal to 1.
[0046] In one example, before constructing a multi-objective optimization problem under dynamic constraints based on multiple constraints, the adaptive control method for the water electrolysis hydrogen production system provided in this application embodiment may further include the following steps: Multiple constraints are set, including: upper and lower limits of current, absolute temperature and temperature difference, pressure, valve opening, and state equation constraints defined by the model itself.
[0047] Step S103: In the rolling time domain, optimize the multi-objective optimization problem to obtain the set of optimal control input sequences within the preset control period.
[0048] The objective function of the optimization problem is to maximize the cumulative sum of the composite operational efficiency index function Φ over the next N prediction steps in the time domain: Formula (9); In the formula, N: the prediction time-domain step size (natural number) of model predictive control (MPC). k: The kth discrete time step in the future prediction time domain.
[0049] Formula (10); In the formula, η_H2(k): the instantaneous hydrogen production efficiency predicted at the k-th step (e.g., hydrogen production / electricity consumption). ΔT_std(k): Spatial standard deviation of the temperature of all control volumes of the electrolyzer predicted in step k, characterizing the uniformity of the temperature field; P_dev(k): The absolute deviation between the predicted pressure of the separator tank in step k and the set value; ω1(k), ω2(k), ω3(k): These are dynamic weighting factors for energy efficiency, temperature uniformity, and pressure stability, respectively; determined by the upper-level management module based on... The predicted trends and system operating modes (e.g., efficiency priority, safety priority) are generated in real time. For example, when When the value is a large negative value, immediately increase ω2.
[0050] ξ(k): Thermal inertia feedforward compensation coefficient.
[0051] Formula (11); In the formula, t0: the initial time of the current control cycle; |dP / dt(t0)|: The absolute value of the volatility of the input renewable energy power at the current moment; ΔT(t0): The maximum temperature difference inside the electrolytic cell at the current moment (i.e., the difference between the highest and lowest temperatures); γ, δ: Empirically adjustable parameters used to adjust the sensitivity and saturation of thermal inertia compensation; tanh: Hyperbolic tangent function, used to limit the compensation coefficient within a reasonable range to prevent the control quantity from diverging.
[0052] Constraints include: upper and lower current limits Absolute temperature value and temperature difference constraint , Pressure constraints The constraints include valve opening constraints and state equation constraints defined by the model itself. The boundary values of these constraints can also be dynamically fine-tuned according to the operating conditions.
[0053] It should be noted that the multi-objective dynamic trade-off optimization mechanism proposes an online optimization framework centered on the "composite operating efficiency index function Φ". This function innovatively unifies and quantifies heterogeneous objectives such as hydrogen production efficiency, temperature field uniformity, and pressure stability, and introduces dynamic weighting factors (ω1, ω2, ω3), which can be adjusted online according to the operating mode and external fluctuation predictions, realizing real-time, flexible trade-offs and collaborative optimization among multiple objectives.
[0054] Thermal inertia active feedforward compensation design: A "thermal inertia compensation coefficient ξ" is constructed. This coefficient nonlinearly combines the fluctuation rate of input power (dP / dt) with the real-time thermal state (ΔT) of the system and embeds it into the objective function. This enables the nonlinear model predictive control (MPC) optimization algorithm to proactively compensate for the temperature response lag caused by the huge heat capacity of the electrolyzer. This upgrades the traditional temperature feedback control to "thermal inertia feedforward-feedback fusion" control, significantly improving the stability and anticipation of thermal management.
[0055] In one example, the optimization solution for a multi-objective optimization problem in the rolling time domain includes the following steps: Obtain a multi-objective optimization problem, wherein the multi-objective optimization problem is a constrained nonlinear optimization problem; The multi-objective optimization problem is transformed into a corresponding sequential quadratic programming problem.
[0056] In another example, the optimization solution for a multi-objective optimization problem in the rolling time domain includes the following steps: Obtain a multi-objective optimization problem, wherein the multi-objective optimization problem is a constrained nonlinear optimization problem; The multi-objective optimization problem is transformed into a form solvable by the interior point method.
[0057] Step S104: Take the first optimal control input sequence in the optimal control input sequence set as the actual control command, and send the actual control command to the actuator to execute the actual control command and obtain multiple actual data, including: actual electrochemical data, actual temperature field data and actual gas-flow data.
[0058] Step S105: Compare any one of the predicted data from the multiple predicted data with its matching actual data to obtain the corresponding comparison result.
[0059] Step S106: Based on the comparison results and preset methods, perform online fine-tuning of multiple different time-varying parameters in the key time-varying parameter set of the prediction model to achieve adaptive control. The preset methods include: recursive least squares method or extended Kalman filter method.
[0060] like Figure 2 The diagram shows the control architecture of the adaptive control method for the water electrolysis hydrogen production system according to an embodiment of this application.
[0061] The adaptive control method for a water electrolysis hydrogen production system based on multiphysics coupling provided in this application uses a control algorithm that cyclically executes the following steps within each fixed sampling period (e.g., Tc = 100ms): Step 1: Fusion processing of the perceived multi-source heterogeneous data and the estimated state data.
[0062] Data is collected in real time through a sensor network, and key states (such as membrane water content and catalyst surface temperature) that are difficult to measure directly or have high noise levels are estimated using software modules based on Kalman filters or sliding window observers, forming a complete current state vector x0. The collected data includes: Electrical quantities: DC bus current I_dc, total voltage V_stack, instantaneous active power P_in from the converter and its rate of change dP / dt.
[0063] Thermodynamic quantities: Temperature sensor groups T_1…T_n arranged at the inlet, outlet and typical locations of the electrolytic cell; inlet temperature T_cool_in and flow rate F_cool of the main coolant pipeline and key branches.
[0064] Fluid and gas quantities: frequency and outlet pressure of electrolyte circulating pump, readings of hydrogen / oxygen online purity analyzer and mass flow meter F_H2, F_O2, and pressure of gas-liquid separator P_sep.
[0065] Environmental parameters: ambient temperature T_amb, humidity.
[0066] Step 2: Predict the rolling state based on the high-fidelity coupling model.
[0067] Step 2 is based on the prediction engine of nonlinear model predictive control (MPC). Based on the current state vector x0 and the assumed future control input sequence, a coupled model library is invoked for forward simulation.
[0068] (1) Electrochemical voltage-current relationship sub-model: Formula (1); In the formula, V_cell: the actual operating voltage (V) of a single cell in the electrolyzer; E_rev: Reversible open-circuit voltage (V), i.e., the lowest electrolysis voltage according to thermodynamic theory; η_act: Activation overpotential (V), the additional voltage required to overcome the activation energy of an electrochemical reaction; η_ohm: Ohmic overpotential (V), the voltage required to overcome the resistance of the components inside the electrolytic cell.
[0069] Formula (2); In the formula, E_rev,0: the reversible voltage under standard conditions (typically 1.229 V); ΔS: Standard molar entropy change of reaction (J / (mol·K)); n: The number of electrons transferred in the reaction (n=2 for water electrolysis); F: Faraday constant (approximately 96485 C / mol); T: The actual absolute operating temperature (K) of the electrolyzer; T_ref: Reference absolute temperature (usually 298.15 K); R: Ideal gas constant (8.314 J / (mol·K)); p_H2: The actual partial pressure of hydrogen on the cathode side (atm or bar); p_O2: The actual partial pressure of oxygen on the anode side (atm or bar); a_H2O: The activity of liquid water (usually approximately 1 in pure water or dilute solutions).
[0070] Formula (3); In the formula, α: charge transfer coefficient (symmetry factor), which usually takes a value between 0 and 1; i: Actual operating current density (A / cm²); i_ex: Exchange current density (A / cm²), which is a function that strongly correlates catalyst activity with temperature; asinh: Inverse hyperbolic sine function.
[0071] Formula (4); In the formula, R_mem: the equivalent area resistivity (Ω·cm²) of the proton exchange membrane (PEM) or diaphragm, is a time-varying function of membrane water content, temperature and aging state; R_ele: Electronic conductivity resistance of the electrode and current collector (Ω·cm²); R_c: Contact resistance between components (Ω·cm²).
[0072] (2) Dynamic sub-model of non-uniform temperature field (discretized using lumped parameter method): The reaction zone of the electrolyzer is discretized in three dimensions to generate M control volumes. For the j-th control volume: Formula (5); In the formula, C_j: the equivalent heat capacity (J / K) of the j-th control volume; T_j: Real-time temperature (K) of the j-th control volume; t: time variable (s); Q_gen, j: Internal heat generation rate (W) of the j-th control volume; Q_cool, j: The heat dissipation rate (W) of the j-th control unit carried away by the coolant; Q_loss, j: The rate (W) of heat loss from the j-th control volume to the external environment; λ_jk: Equivalent thermal conductivity (W / K) between the j-th control volume and the adjacent k-th control volume; T_k: Real-time temperature (K) of the kth adjacent control volume.
[0073] Formula (6); In the formula, Heat generation includes ohmic joule heat and heat of irreversible reaction.
[0074] i_j: Local current density of the j-th control volume (A / cm²); A_j: Effective electrochemical reaction area (cm²) of the j-th control body; V_cell, j: Local voltage (V) of the j-th control cell; E_rev, j: Local reversible voltage (V) of the j-th control volume; η_act, j: Local activation overpotential (V) of the j-th control body; ζ (Zeta): Thermal neutral voltage conversion factor (or irreversible heat dissipation factor), used to characterize the proportion of the activation overpotential converted into heat energy.
[0075] Formula (7); In the formula, : Convection heat dissipation item.
[0076] h_j(v_cool): Local convective heat transfer coefficient (W / (m²·K)), which is a function of the local coolant flow rate v_cool; A_s, j: Effective convective heat transfer area (m²) between the j-th control volume and the coolant; T_coolant, j: Real-time temperature (K) of the coolant flowing through the j-th control unit.
[0077] (3) Gas-fluid network sub-model: Formula (8); In the formula, ρ_l: density of liquid water / electrolyte in the gas-liquid separator (kg / m³). V_sep: Effective gas volume (m³) of the gas-liquid separator. K: The bulk modulus of the gas or the proportionality constant in the equation of state; P_sep: Real-time pressure (Pa or bar) of the gas-liquid separator. β_in: The mass or volume fraction ratio coefficient of hydrogen in the inlet gas-liquid mixture; F_H2, in: Hydrogen mass flow rate (kg / s) flowing into the gas-liquid separator. F_H2, out: Mass flow rate of pure hydrogen gas (kg / s) flowing out of the gas-liquid separator. f_valve: The nonlinear characteristic function of the back pressure valve in terms of flow rate, pressure, and opening degree; u_valve: The actual control opening command of the back pressure valve (0~100%).
[0078] The pressure dynamics of the separator are described by the above formula (8), where, This is the flow-pressure-opening characteristic function of the back pressure valve.
[0079] Step 3: Dynamic constraints and multi-objective optimization problem construction.
[0080] The objective function of the optimization problem is to maximize the cumulative sum of the composite operational efficiency index function Φ over the next N prediction steps in the time domain: Formula (9); In the formula, N: the prediction time-domain step size (natural number) of model predictive control (MPC). k: The kth discrete time step in the future prediction time domain.
[0081] Formula (10); In the formula, η_H2(k): the instantaneous hydrogen production efficiency predicted at the k-th step (e.g., hydrogen production / electricity consumption). ΔT_std(k): Spatial standard deviation of the temperature of all control volumes of the electrolyzer predicted in step k, characterizing the uniformity of the temperature field; P_dev(k): The absolute deviation between the predicted pressure of the separator tank in step k and the set value; ω1(k), ω2(k), ω3(k): These are dynamic weighting factors for energy efficiency, temperature uniformity, and pressure stability, respectively; determined by the upper-level management module based on... The predicted trends and system operating modes (e.g., efficiency priority, safety priority) are generated in real time. For example, when When the value is a large negative value, immediately increase ω2.
[0082] ξ(k): Thermal inertia feedforward compensation coefficient.
[0083] Formula (11); In the formula, t0: the initial time of the current control cycle; |dP / dt(t0)|: The absolute value of the volatility of the input renewable energy power at the current moment; ΔT(t0): The maximum temperature difference inside the electrolytic cell at the current moment (i.e., the difference between the highest and lowest temperatures); γ, δ: Empirically adjustable parameters used to adjust the sensitivity and saturation of thermal inertia compensation; tanh: Hyperbolic tangent function, used to limit the compensation coefficient within a reasonable range to prevent the control quantity from diverging.
[0084] Constraints include: upper and lower current limits Absolute temperature value and temperature difference constraint , Pressure constraints The constraints include valve opening constraints and state equation constraints defined by the model itself. The boundary values of these constraints can also be dynamically fine-tuned according to the operating conditions.
[0085] Step 4: Optimize the solution in the rolling time domain.
[0086] The constrained nonlinear optimization problem constructed in step 3 is transformed into a solvable form using sequential quadratic programming (SQP) or the interior-point method. In each control cycle, the solver calculates the optimal control input sequence. Where u includes I_set, F_cool_set, valve_openings, etc. Only... As the actual control command output.
[0087] Step 5: Instruction execution and online adaptation of model parameters.
[0088] Will The data is then sent to the execution mechanism. Simultaneously, at the end of each cycle, the actual system output y_real is compared with the model's predicted output y_pred. Recursive Least Squares (RLS) or Extended Kalman Filter (EKF) is used to fine-tune different time-varying parameters (such as baseline values for R_mem, i_ex, and h_j) in the key time-varying parameter set θ of the prediction model online. Formula (12); In the formula, θ_{new}: the set of key time-varying parameters updated in the current period (such as the updated membrane resistance R_mem); θ_{old}: The set of key time-varying parameters used in the previous cycle; K: Kalman gain or adaptive gain matrix calculated by an adaptive algorithm (such as RLS or EKF); y_real: The actual output vector of the system acquired by the sensor in the current cycle (such as actual voltage, actual temperature); y_pred: The prediction output vector calculated by the current cycle prediction model.
[0089] Figure 3 A diagram of a dedicated heterogeneous hardware acceleration platform architecture is shown.
[0090] To ensure the real-time operation of the above algorithm, the adaptive control method provided in this application proposes the following dedicated hardware architecture: Overall architecture: It adopts a three-layer heterogeneous computing architecture of "edge awareness - centralized decision-making - distributed execution".
[0091] Edge sensing layer: Composed of multiple industrial-grade IO modules, each module is based on an ARM Cortex-M7 core microcontroller and is responsible for high-speed acquisition, filtering and preliminary processing of sensor signals in a specific area / type. It sends data packets to the decision layer with deterministic latency via EtherCAT or TSN (Time-Sensitive Network) industrial Ethernet.
[0092] Centralized decision-making layer (core): Employs a high-performance heterogeneous system-on-a-chip, such as the Xilinx Zynq UltraScale+ MPSoC. This chip contains programmable logic (FPGA) and a multi-core processing system (PS, such as ARM Cortex-A53 / A72).
[0093] FPGA Component: Responsible for all computationally intensive, high real-time tasks. Implements hardware acceleration for the coupled prediction model in step 2. Maps the thermal differential equations of M control volumes to M parallel computing units (Processing Elements, PEs) within the FPGA, enabling synchronous parallel solving of each prediction step, far exceeding the speed of sequential CPU execution. Implements hardware acceleration for optimizing the core linear algebra operations of the solver in step 4 (such as matrix multiplication, inversion, and interior point iteration in QP solving). Through customized pipeline design, solution time can be significantly reduced.
[0094] Multi-core ARM PS section: runs a real-time operating system.
[0095] Core 0: Responsible for communication management, data fusion, state estimation (step 1), and dynamic weight calculation (step 3 part of the logic).
[0096] Core 1: Runs the upper-level logic of the optimization solver, the parameter adaptive algorithm (step 5), and system monitoring. PS and PL (FPGA) are interconnected via a high-speed AXI bus for extensive data exchange.
[0097] Distributed execution layer: Receives control commands from the decision layer via Profinet IRT or Ethernet / IP, and controls actuators such as rectifiers, frequency converters, and proportional valves through high-precision analog output modules or servo drives.
[0098] It should be noted that the fully embedded heterogeneous hardware acceleration platform features a dedicated hardware implementation scheme based on a heterogeneous SoC of "FPGA + multi-core ARM CPU". By utilizing the parallel pipeline of the FPGA to accelerate the prediction model and matrix operations at the hardware level, and using the CPU for logic management and parameter adaptation, the complex algorithm can be reliably completed within a millisecond-level control cycle, making it possible to implement high-order MPC algorithms in real time in industrial settings.
[0099] The adaptive control method for the water electrolysis hydrogen production system provided in this application embodiment can maintain optimal performance throughout its entire life cycle through adaptive control. Furthermore, the adaptive control method provided in this application embodiment significantly improves dynamic response quality and energy absorption capacity. Through model prediction and rolling optimization, a fundamental shift from "passive delayed response" to "active forward adjustment" is achieved. When facing fluctuations in renewable energy power, it can adjust control quantities in advance, smoothly transitioning and effectively avoiding severe oscillations in operating conditions. This significantly shortens the transition time of power tracking, thereby greatly improving the immediate absorption ratio of fluctuating energy and the overall energy efficiency of the system. Secondly, the adaptive control method provided in this application embodiment fundamentally ensures operational safety and lifespan under multi-physics field collaboration. A unified optimization framework collaboratively handles all key variables such as electricity, heat, and gas, completely solving the problem of single-loop control's incomplete coverage. It can simultaneously optimize the cooling flow field under high current density operation, effectively suppressing the generation of local hot spots; while simultaneously controlling gas pressure smoothly. This greatly reduces the risk of core components (such as membrane electrodes and catalysts) deteriorating due to thermal and mechanical stress, providing a key guarantee for the long-term, high-reliability operation of the electrolyzer. Furthermore, the adaptive control method provided in this application enhances the system's robustness to changes in its own state and its long-term performance. The online parameter adaptive fine-tuning mechanism enables the predictive model to continuously track characteristic drift caused by system aging, ensuring that the control strategy "keeps pace with the times." This allows the control system to maintain near-optimal performance throughout its entire lifecycle, overcoming the persistent problem of performance degradation in fixed-parameter strategies. In addition, the adaptive control method provided in this application offers a highly reliable engineering implementation path. The proprietary heterogeneous computing hardware platform design solves the bottleneck of real-time computing resources for complex algorithms, ensuring that the advanced control strategy described in this application can operate stably and at high speed in harsh industrial environments, demonstrating clear engineering feasibility and application value.
[0100] In the above embodiments, an adaptive control method for a water electrolysis hydrogen production system is provided. Correspondingly, this application also provides an adaptive control system for a water electrolysis hydrogen production system. The adaptive control system for the water electrolysis hydrogen production system provided in this application can implement the above-described adaptive control method for the water electrolysis hydrogen production system. This adaptive control system for the water electrolysis hydrogen production system can be implemented through software, hardware, or a combination of both. For example, the adaptive control system for the water electrolysis hydrogen production system may include integrated or separate functional modules or units to execute the corresponding steps in the above methods.
[0101] Please refer to Figure 4This illustration shows a schematic diagram of an adaptive control system for a water electrolysis hydrogen production system provided by some embodiments of this application. Since the system embodiments are basically similar to the method embodiments, the description is relatively simple; relevant details can be found in the description of the method embodiments. The system embodiments described below are merely illustrative.
[0102] like Figure 4 As shown, the adaptive control system 400 of the water electrolysis hydrogen production system may include: The prediction module 401 is used to call the prediction model to perform forward simulation rolling prediction based on the current state vector and the assumed future control input sequence, and obtain multiple prediction data. Among them, the multiple prediction data include: predicted electrochemical data, predicted temperature field data and predicted gas-fluid data. The coupling model library used by the prediction model includes: electrochemical voltage-current relationship sub-model, non-uniform temperature field dynamic sub-model and gas-fluid network sub-model. Module 402 is used to construct a multi-objective optimization problem under dynamic constraints based on multiple constraints. The objective function of the multi-objective optimization problem is to maximize the cumulative sum of the composite operating efficiency index function in the prediction time domain in the next N steps, where N is a natural number greater than or equal to 1. The optimization solution module 403 is used to optimize and solve a multi-objective optimization problem in the rolling time domain to obtain the set of optimal control input sequences within a preset control period. The processing module 404 is used to take the first optimal control input sequence in the optimal control input sequence set as the actual control command, and to send the actual control command to the actuator so that the actuator can execute the actual control command to obtain multiple actual data, including: actual electrochemical data, actual temperature field data and actual gas-flow data; The comparison module 405 is used to compare any one of the predicted data from multiple predicted data with its matching actual data to obtain the corresponding comparison result; The online fine-tuning module 406 is used to perform online fine-tuning of multiple different time-varying parameters in the key time-varying parameter set of the prediction model based on the comparison results and preset methods, so as to carry out adaptive control.
[0103] In some embodiments of this application, the adaptive control system 400 of the water electrolysis hydrogen production system may further include: Get module (in) Figure 4(Not shown in the image) is used to: acquire multi-source heterogeneous data and state data before calling the prediction model to perform forward simulation rolling prediction based on the current state vector and the assumed future control input sequence, and obtaining multiple prediction data. The multi-source heterogeneous data includes: first correlation data associated with electrical quantities, second correlation data associated with thermodynamic quantities, third correlation data associated with fluid and gas quantities, and fourth correlation data associated with environmental quantities; the state data includes: membrane water content and catalyst surface temperature. Fusion processing module (in) Figure 4 (Not shown in the image), used to fuse multi-source heterogeneous data and state data to obtain the current state vector.
[0104] In some embodiments of this application, the optimization solution module 403 is specifically used for: Obtain a multi-objective optimization problem, wherein the multi-objective optimization problem is a constrained nonlinear optimization problem; The multi-objective optimization problem can be transformed into a corresponding sequential quadratic programming problem; or... The multi-objective optimization problem is transformed into a form solvable by the interior point method.
[0105] In some embodiments of this application, the adaptive control system 400 of the water electrolysis hydrogen production system may further include: The acquisition module (in) Figure 4 (not shown in the image), specifically used for: Before fusing multi-source heterogeneous data and state data to obtain the current state vector, the first associated data is collected in real time. The first associated data includes: DC bus current, total voltage, instantaneous active power from the converter and its rate of change. The second correlation data is collected in real time. The second correlation data includes: temperature sensor groups arranged sequentially at the inlet, outlet and preset typical positions of the electrolytic cell, inlet temperature of the main coolant pipeline and key branches, and flow rate of the main coolant pipeline and key branches. Real-time acquisition of third-party correlation data, including: electrolyte circulation pump frequency and outlet pressure, hydrogen / oxygen online purity analyzer and mass flow meter readings, and gas-liquid separator pressure; The fourth correlation data is collected in real time, including ambient temperature and humidity.
[0106] In some embodiments of this application, the adaptive control system 400 of the water electrolysis hydrogen production system may further include: Settings module (in) Figure 4 (not shown in the image), specifically used for: Before constructing a multi-objective optimization problem based on dynamic constraints with multiple constraints, various constraints are set, including: upper and lower limits of current, absolute temperature and temperature difference constraints, pressure constraints, valve opening constraints, and state equation constraints defined by the model itself.
[0107] According to another embodiment, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed in a computer, causes the computer to execute the adaptive control method of the above-described water electrolysis hydrogen production system.
[0108] According to another embodiment, an electronic device is also provided, including a memory and a processor, wherein the memory stores executable code, and when the processor executes the executable code, it implements the adaptive control method of the above-described water electrolysis hydrogen production system.
[0109] Those skilled in the art will recognize that, in one or more of the examples above, the functions described in this application can be implemented using hardware, software, firmware, or any combination thereof. When implemented in software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium.
[0110] Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.
Claims
1. An adaptive control method for a water electrolysis hydrogen production system, characterized in that, The method includes: Based on the current state vector and the assumed future control input sequence, the prediction model is invoked to perform forward simulation rolling prediction, resulting in multiple prediction data, including: predicted electrochemical data, predicted temperature field data, and predicted gas-fluid data. The coupled model library used by the prediction model includes: electrochemical voltage-current relationship sub-model, non-uniform temperature field dynamic sub-model, and gas-fluid network sub-model. Under dynamic constraints based on multiple constraints, a multi-objective optimization problem is constructed. The objective function of the multi-objective optimization problem is to maximize the cumulative sum of the composite operating efficiency index function in the prediction time domain in the next N steps, where N is a natural number greater than or equal to 1. In the rolling time domain, the multi-objective optimization problem is optimized and solved to obtain the set of optimal control input sequences within a preset control period; The first optimal control input sequence in the optimal control input sequence set is used as the actual control command, and the actual control command is sent to the actuator to execute the actual control command and obtain multiple actual data, wherein the multiple actual data include: actual electrochemical data, actual temperature field data and actual gas-flow data; Compare any one of the multiple predicted data with its matching actual data to obtain the corresponding comparison result; Based on the comparison results and the preset method, multiple different time-varying parameters in the key time-varying parameter set in the prediction model are fine-tuned online to perform adaptive control. The preset method includes: recursive least squares method or extended Kalman filter method.
2. The adaptive control method according to claim 1, characterized in that, Before invoking the prediction model to perform forward simulation and rolling prediction based on the current state vector and the assumed future control input sequence to obtain multiple prediction data, the method further includes: Acquire multi-source heterogeneous data and state data, wherein the multi-source heterogeneous data includes: first correlation data associated with electrical quantities, second correlation data associated with thermodynamic quantities, third correlation data associated with fluid and gas quantities, and fourth correlation data associated with environmental quantities; the state data includes: membrane water content and catalyst surface temperature; The multi-source heterogeneous data and the state data are fused together to obtain the current state vector.
3. The adaptive control method according to claim 1, characterized in that, The optimization solution of the multi-objective optimization problem in the rolling time domain includes: Obtain the multi-objective optimization problem, wherein the multi-objective optimization problem is a constrained nonlinear optimization problem; The multi-objective optimization problem is transformed into a corresponding sequential quadratic programming problem; or... The multi-objective optimization problem is transformed into a form solvable by the interior point method.
4. The adaptive control method according to claim 2, characterized in that, Before fusing the multi-source heterogeneous data and the state data to obtain the current state vector, the method further includes: The first associated data is collected in real time, including: DC bus current, total voltage, instantaneous active power from the converter and its rate of change; The second associated data is collected in real time. The second associated data includes: a group of temperature sensors arranged sequentially at the inlet, outlet and preset typical positions of the electrolytic cell; the inlet temperature of the main coolant pipeline and key branches; and the flow rate of the main coolant pipeline and key branches. The third associated data is collected in real time, including: electrolyte circulation pump frequency and outlet pressure, hydrogen / oxygen online purity analyzer and mass flow meter readings, and gas-liquid separator pressure; The fourth associated data is collected in real time, and the fourth associated data includes: ambient temperature and humidity.
5. The adaptive control method according to claim 1, characterized in that, Before constructing the multi-objective optimization problem under dynamic constraints based on multiple constraints, the method further includes: The various constraints are set, including: current upper and lower limit constraints, absolute temperature and temperature difference constraints, pressure constraints, valve opening constraints, and state equation constraints defined by the model itself.
6. An adaptive control system for a water electrolysis hydrogen production system, characterized in that, The system includes: The prediction module is used to call the prediction model to perform forward simulation rolling prediction based on the current state vector and the assumed future control input sequence, and obtain multiple prediction data. The multiple prediction data include: predicted electrochemical data, predicted temperature field data and predicted gas-fluid data. The coupling model library used by the prediction model includes: electrochemical voltage-current relationship sub-model, non-uniform temperature field dynamic sub-model and gas-fluid network sub-model. The module is used to construct a multi-objective optimization problem under dynamic constraints based on multiple constraints. The objective function of the multi-objective optimization problem is to maximize the cumulative sum of the composite operating efficiency index function in the prediction time domain in the next N steps, where N is a natural number greater than or equal to 1. The optimization solution module is used to optimize and solve the multi-objective optimization problem in the rolling time domain to obtain the set of optimal control input sequences within a preset control period; The processing module is used to take the first optimal control input sequence in the optimal control input sequence set as the actual control instruction, and to send the actual control instruction to the actuator so that the actuator can execute the actual control instruction to obtain multiple actual data, wherein the multiple actual data include: actual electrochemical data, actual temperature field data and actual gas-flow data; The comparison module is used to compare any one of the multiple predicted data with its matching actual data to obtain the corresponding comparison result; The online fine-tuning module is used to perform online fine-tuning of multiple different time-varying parameters in the key time-varying parameter set of the prediction model based on the comparison results and preset methods, so as to perform adaptive control.
7. The adaptive control system according to claim 6, characterized in that, The system also includes: The acquisition module is used to: acquire multi-source heterogeneous data and state data before the rolling prediction of forward simulation based on the current state vector and the assumed future control input sequence, and to obtain multiple prediction data. The multi-source heterogeneous data includes: first correlation data associated with electrical quantities, second correlation data associated with thermodynamic quantities, third correlation data associated with fluid and gas quantities, and fourth correlation data associated with environmental quantities; the state data includes: membrane water content and catalyst surface temperature. The fusion processing module is used to fuse the multi-source heterogeneous data and the state data to obtain the current state vector.
8. The adaptive control system according to claim 6, characterized in that, The optimization solution module is specifically used for: Obtain the multi-objective optimization problem, wherein the multi-objective optimization problem is a constrained nonlinear optimization problem; The multi-objective optimization problem is transformed into a corresponding sequential quadratic programming problem; or... The multi-objective optimization problem is transformed into a form solvable by the interior point method.
9. A computer-readable storage medium, characterized in that, It contains a computer program that, when executed in a computer, causes the computer to perform the method described in any one of claims 1 to 5.
10. An electronic device, characterized in that, The method includes a memory and a processor, wherein the memory stores executable code, and the processor executes the executable code to implement the method according to any one of claims 1 to 5.