Fuel cell system thermal management adaptive control method based on online learning algorithm

Abstract

The application provides a fuel cell system thermal management adaptive control method based on an online learning algorithm. The fuel cell system thermal management adaptive control method based on the online learning algorithm can improve the control precision and stability of fuel cell system thermal management, reduce energy consumption, and enhance the adaptability of the system to different radiator models and environmental temperatures.

Description

Technical Field

[0001] This invention relates to the field of fuel cells, and more particularly to an adaptive control method for thermal management of fuel cell systems based on online learning algorithms. Background Technology

[0002] Fuel cell systems, as a core technology in the field of clean energy, are widely used in transportation, distributed energy, and industrial power generation. With the continuous improvement of fuel cell stack power density and the increasing complexity of application scenarios, their thermal management control technology has evolved from traditional single PID temperature control to a joint control system of thermostat and radiator. Specifically, modern cooling subsystems, through the coordinated operation of water pumps, electric heaters, thermostats, and radiators, construct a multi-physics coupled control architecture covering flow distribution, temperature regulation, and energy recovery. Among these, thermostat angle adjustment and radiator fan control, as key links in the temperature closed loop, need to achieve precise matching of flow rate and heat dissipation capacity under dynamic operating conditions to maintain the stack operating temperature within the optimal range, ensuring electrochemical reaction efficiency and material structural stability.

[0003] However, the thermostat-radiator combined control scheme commonly used in existing technologies has systemic defects. Specifically, traditional PID control strategies are unable to cope with temperature fluctuations caused by frequent load changes within the 10%-100% Pe operating range, and the nonlinear relationship between the thermostat angle and flow distribution is prone to causing temperature control oscillations. These technical limitations make existing systems susceptible to temperature runaway risks when there are sudden changes in ambient temperature or frequent load adjustments. At the same time, the over-reliance on initial parameter calibration and hardware redundancy configuration significantly increases research and development costs and energy consumption. Summary of the Invention

[0004] The present invention aims to at least partially solve one of the technical problems in the related art.

[0005] Therefore, the first objective of this invention is to propose an adaptive control method for thermal management of fuel cell systems based on online learning algorithms.

[0006] The second objective of this invention is to propose an adaptive control device for thermal management of fuel cell systems based on online learning algorithms.

[0007] To achieve the above objectives, a first aspect of the present invention proposes an adaptive control method for thermal management of a fuel cell system based on an online learning algorithm, comprising: S1, determine the linear range of the relationship between the thermostat angle and the large circulation flow rate; S2, calculate the cooling flow rate and water pump speed based on the heat generation of the fuel cell stack at different operating conditions, and control the thermostat angle within the linear range; S3, divide the radiator fan speed into N speeds evenly according to the heat dissipation capacity, and establish a two-dimensional fan speed lookup table for ambient temperature and system operating power; S4 uses an online learning algorithm to perform steady-state load tests on the full power points in a two-dimensional lookup table and dynamically adjusts the fan speed parameters according to the temperature stability. S5 stores the current optimal fan speed parameters to the control device hardware when the system is shut down, and iteratively updates the initial calibration parameters to form a closed-loop learning.

[0008] In one embodiment of the present invention, S1 includes: S11, calculates the nonlinear relationship curve between thermostat angle and large circulation flow rate through multiphysics simulation; S12, Fit the nonlinear relationship curve based on the experimental test data, and determine the upper and lower limits of the linear interval based on the fitting results.

[0009] In one embodiment of the present invention, S2 includes: S21 uses a thermodynamic model to calculate the real-time heat generation of the fuel cell stack at the target temperature and dynamically adjusts the pump speed based on the heat generation. S22 corrects the thermostat angle by real-time feedback of the coolant temperature deviation from the target temperature to maintain it within the linear range.

[0010] In one embodiment of the present invention, S3 includes: S31, based on the fan speed-heat dissipation power characteristic curve, divides the fan speed into N+1 levels, where level 0 corresponds to the fan stopping and level N+1 corresponds to the maximum heat dissipation power; S32 optimizes the configuration parameters of the initial gear table based on the combined operating conditions of ambient temperature and system operating power using a genetic algorithm.

[0011] In one embodiment of the present invention, S4 includes: S41 uses a differential temperature detection method to determine the temperature change trend. When the temperature continues to rise and the thermostat opening exceeds the upper limit of the linear range, it triggers an upgrade operation. The S42 uses an intelligent counting algorithm to quantitatively evaluate temperature trends. When the count value exceeds a preset threshold, it executes fan speed switching and records steady-state operating data.

[0012] In one embodiment of the present invention, S5 includes: S51, based on the genetic algorithm, performs crossover and mutation operations on the fan speed parameters of this run to generate the next generation of optimized parameter set; S52 compares the optimized parameter set with the initial calibration parameters and selects the parameter with the highest fitness value as the initial value for the next run.

[0013] To achieve the above objectives, a second aspect of the present invention provides an adaptive control device for thermal management of a fuel cell system based on an online learning algorithm, comprising: The linear interval determination module is used to determine the linear interval range of the relationship between the thermostat angle and the large circulation flow rate; The operating condition control module is used to calculate the cooling flow rate and water pump speed based on the heat generation of the fuel cell stack at different operating conditions, and to control the thermostat angle within the linear range. The fan speed configuration module is used to evenly divide the heat sink fan speed into N speeds according to the heat dissipation capacity, and to establish a two-dimensional fan speed lookup table of ambient temperature and system operating power. The online learning adjustment module is used to perform steady-state load tests on the full power points in the two-dimensional lookup table through an online learning algorithm, and dynamically adjust the fan speed parameters according to the temperature stability state. The parameter storage and update module is used to store the current optimal fan speed parameters to the control device hardware when the system is shut down, and iteratively update the initial calibration parameters to form a closed-loop learning.

[0014] The adaptive control method for thermal management of fuel cell systems based on online learning algorithms in this invention can improve the thermal management response speed and control stability of fuel cell systems under complex operating conditions. By using online learning algorithms to achieve adaptive adjustment of fan speed, it reduces redundant heat dissipation configuration and enhances system economy and robustness.

[0015] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0016] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of an adaptive control method for thermal management of a fuel cell system based on an online learning algorithm, according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the cooling subsystem loop of a fuel cell system according to an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the relationship between the thermostat angle and the large circulation flow rate according to an embodiment of the present invention; Figure 4 This is a block diagram of the fan speed adaptive online learning function according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the online learning control process for thermal management of a fuel cell system based on a genetic algorithm, according to an embodiment of the present invention. Figure 6 This is a structural diagram of an adaptive control device for thermal management of a fuel cell system based on an online learning algorithm, according to an embodiment of the present invention. Detailed Implementation

[0017] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0018] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0019] The adaptive control method for thermal management of a fuel cell system based on an online learning algorithm, proposed according to an embodiment of the present invention, is described below with reference to the accompanying drawings.

[0020] Example 1 Figure 1 This is a flowchart of an adaptive control method for thermal management of a fuel cell system based on an online learning algorithm, according to an embodiment of the present invention. Figure 1 As shown, it includes: S1 determines the linear range of the relationship between the thermostat angle and the large circulation flow rate.

[0021] Specifically, in some implementations, confirming the linear range of the relationship between the thermostat angle and the large circulation flow rate is a key preliminary step in the thermal management adaptive control method of this invention. Its technical implementation principle is based on the analysis of the flow response characteristics of the thermostat at different opening degrees. As the core component controlling the distribution of large and small circulation flow rates in the cooling system, the thermostat's angle change directly affects the coolant flow rate in the large circulation. Within the nonlinear range, a nonlinear relationship exists between the thermostat angle and the flow rate, leading to decreased control accuracy and unstable system response. Therefore, this invention uses experimental or simulation methods to obtain the large circulation flow rate response curves of the thermostat at different angles, thereby identifying the range within which the flow rate change has a linear relationship with the thermostat angle.

[0022] Furthermore, under steady-state conditions, the large-circulation flow rate data at the corresponding angle is collected by adjusting the thermostat angle. Flow rate measurement can be performed using a mass flow meter or the differential pressure method, with a sampling frequency recommended to be no less than 10Hz to ensure data continuity and accuracy.

[0023] This step is typically completed during the system debugging phase in practical applications and is applicable to the formulation of thermal management control strategies for different thermostat models. By defining the linear range of the thermostat, basic constraints can be provided for subsequent control strategies based on online learning algorithms, ensuring that the thermostat always operates within a high linearity range, thereby improving the stability and response speed of the system temperature control. Furthermore, this step provides boundary conditions for parameter optimization of the genetic algorithm, which helps to improve the adaptability and robustness of the control device and is an important technical support for achieving efficient thermal management control of fuel cell systems.

[0024] Furthermore, S1 includes: S11 uses multiphysics simulation to calculate the nonlinear relationship curve between the thermostat angle and the large circulation flow rate.

[0025] Specifically, calculating the nonlinear relationship curve between the thermostat angle and the large circulation flow rate through multiphysics simulation is one of the key technical steps in this invention for achieving adaptive control of the fuel cell system's thermal management. This step aims to establish a precise mapping relationship between the thermostat angle and the large circulation flow rate of the coolant, thereby providing a basic model support for subsequent control strategies.

[0026] Furthermore, this simulation process utilizes a multiphysics modeling method that couples CFD (Computational Fluid Dynamics) and thermodynamics to create a three-dimensional model of the flow channel structure of the thermostat at different opening angles, and sets boundary conditions to simulate the flow state of the coolant inside the thermostat. The simulation needs to consider the physical properties of the coolant (such as density, viscosity, specific heat capacity, etc.) and the geometric parameters of the flow channels inside the thermostat (such as channel cross-sectional area, orifice diameter, valve opening angle, etc.). By changing the opening angle of the thermostat (typically discretized in 5° increments within the range of 0° to 90°), the large circulation flow rate under the corresponding operating conditions is calculated, thereby constructing a nonlinear angle-flow rate relationship curve.

[0027] Furthermore, the relationship between the thermostat angle and the large circulation flow rate needs to be simulated within a specific temperature range (e.g., 40℃~85℃) to cover typical operating conditions of the fuel cell system. Flow rate calculations must meet the accuracy requirements for cooling system flow rates in the ISO 15193 standard, with errors controlled within ±5%. In addition, the linear range of the thermostat is typically defined as the region where angle changes and flow rate changes have an approximately linear relationship, generally between 20° and 70°. The specific range needs to be determined through simulation data fitting, such as using least squares or polynomial fitting, with a correlation coefficient R² ≥ 0.95 to ensure model reliability.

[0028] Furthermore, this step is primarily used for preliminary modeling and parameter calibration of the thermal management control strategy for the fuel cell system. Through simulation results, the control device can identify the flow response characteristics of the thermostat at different angles. Therefore, in actual operation, by adjusting the thermostat angle, the coolant flow rate is kept within the linear range, avoiding control lag and temperature fluctuations caused by nonlinear regions. Especially under conditions of system variable load operation and significant ambient temperature variations, this curve provides crucial input for subsequent adaptive fan speed control.

[0029] Furthermore, this step, through high-precision multiphysics simulation, effectively reveals the nonlinear relationship between the thermostat angle and the large circulation flow rate, providing an accurate mathematical model for the control algorithm. This not only improves the response speed and control accuracy of the system's thermal management but also lays a data foundation for subsequent online learning control based on genetic algorithms, enhancing the system's adaptability and robustness under different radiator models and environmental conditions.

[0030] S12, Fit the nonlinear relationship curve based on the experimental test data, and determine the upper and lower limits of the linear interval based on the fitting results.

[0031] Specifically, this step, which involves fitting the nonlinear relationship curve based on experimental test data and determining the upper and lower limits of the linear interval based on the fitting results, is one of the key steps in realizing adaptive control of thermal management of fuel cell systems. Its technical implementation principle is based on system identification and nonlinear modeling methods, aiming to extract the nonlinear relationship between the thermostat angle and the large circulation flow rate through experimental data and identify the control interval with high linearity, thereby providing accurate adjustment boundaries for subsequent control strategies.

[0032] Furthermore, this step begins by constructing an experimental platform for the cooling circuit of a fuel cell system to collect large circulation flow data of the thermostat at different opening angles. During the experiment, the thermostat angle is typically adjusted within a range of 0° to 90°, with a step interval of 5°. This is combined with the operation of the water pump at different speeds to record key parameters such as coolant flow rate, temperature, and pressure.

[0033] Furthermore, the experimental data were fitted using the least squares method or nonlinear regression algorithm to construct a nonlinear functional relationship model between the thermostat angle and the large circulation flow rate. By calculating the rate of change of the local slope of the fitted curve in different angle intervals (i.e., the fluctuation range of the derivative), the intervals with higher linearity can be identified.

[0034] Optionally, to improve fitting accuracy and robustness, regularization or cross-validation mechanisms can be introduced to prevent overfitting. Finally, based on the fitting results, the upper and lower limits of the linear range of the thermostat angle are determined as constraints for thermostat angle adjustment in subsequent control strategies, ensuring that it always operates within a high linearity range, thereby improving the control accuracy and stability of the thermal management system.

[0035] Furthermore, this is typically executed during the initial system run or when control parameters are updated, and is applicable to different types of fuel cell systems and radiator configurations. Its technical value lies in providing a reliable control boundary for subsequent online learning algorithms, avoiding temperature oscillations caused by the nonlinear characteristics of the thermostat, and improving the system's adaptability and control robustness under complex operating conditions.

[0036] S2 calculates the cooling flow rate and water pump speed based on the heat generation at different operating points of the fuel cell stack, and controls the thermostat angle within the linear range.

[0037] Specifically, this step calculates the cooling flow rate and water pump speed based on the heat generation at different operating points of the fuel cell stack, and controls the thermostat angle within the linear range. This is a key control link in the adaptive control method for thermal management of fuel cell systems. Its core lies in achieving high-precision and high-stability control of system temperature through real-time operating condition analysis and matching with the thermostat flow characteristics.

[0038] Furthermore, this step first uses a heat generation model of the fuel cell stack under different loads (10%~100% Pe) and ambient temperatures. Combining the thermodynamic characteristics of the stack with the heat transfer efficiency of the cooling circuit, the required cooling flow rate at each operating point is calculated. Through this model, the system can determine the coolant flow rate required to maintain the stack temperature within a set range under specific operating conditions.

[0039] Furthermore, based on the cooling flow rate and pump characteristic curves (such as flow rate-speed relationship, head-power curve, etc.), the system calculates the corresponding pump speed. In some implementations, pump speed control can employ PID control or a lookup-based open-loop control strategy to ensure stable coolant circulation within the system.

[0040] Furthermore, there is a non-linear relationship between the thermostat angle and the large circulation flow rate. Typically, within a certain angle range (e.g., 20°~60°), the flow rate change is approximately linearly related to the angle; this range is the linear range of the thermostat. This step, by controlling the thermostat angle within this linear range, ensures the linearity and response speed of its flow regulation capability, thereby improving the accuracy and stability of the system temperature control.

[0041] Furthermore, the linear range of the thermostat is usually determined experimentally or through simulation, and its angle range is closely related to the rate of change of flow (e.g., 0.5~1.2 L / min / °). Cooling flow rate calculations need to consider key parameters such as fuel cell stack heat generation (e.g., 100~500 W), coolant specific heat capacity (e.g., 3.93 kJ / kg·K), and inlet / outlet temperature difference (e.g., 5~15℃). Pump speed control requires mapping based on its flow-speed characteristic curve.

[0042] Furthermore, this step is applicable to fuel cell systems operating under complex conditions, such as frequent vehicle load changes and ambient temperature fluctuations (-20℃ to 40℃). By controlling the thermostat angle within a high linear range, the system can effectively cope with temperature fluctuations caused by sudden load changes or environmental variations, avoid temperature control oscillations, and improve the robustness of system operation.

[0043] Furthermore, this step enables dynamic matching between the cooling system and the stack's thermal load, ensuring that the thermostat operates within its optimal adjustment range. This improves temperature control accuracy (within ±1℃), reduces system energy consumption, extends the service life of the thermostat and water pump, and enhances the overall thermal management performance of the fuel cell system.

[0044] Furthermore, S2 includes: S21 uses a thermodynamic model to calculate the real-time heat generation of the fuel cell stack at the target temperature and dynamically adjusts the pump speed based on the heat generation.

[0045] Specifically, this step uses a thermodynamic model to calculate the real-time heat generation of the fuel cell stack at the target temperature and dynamically adjusts the water pump speed based on the heat generation. This is a key control link in the adaptive control method for thermal management of fuel cell systems. Its core lies in combining real-time thermodynamic modeling with feedback control to achieve precise regulation of the cooling system flow rate, thereby maintaining the fuel cell stack operating temperature within the set range and improving system operating efficiency and stability.

[0046] Furthermore, this step first calculates the heat generation of the fuel cell stack in real time based on a thermodynamic model of the fuel cell stack and the current operating conditions (such as output power, reactant gas flow rate, ambient temperature, etc.). The thermodynamic model typically employs the principle of energy conservation, combining the heat generation formula of the electrochemical reaction of the stack with a heat conduction model to estimate the heat generation rate inside the stack. Further, the system calculates the required coolant flow rate based on the deviation between the target temperature and the current temperature, combined with the heat exchange capacity of the radiator, and maps this flow rate to the water pump speed to achieve dynamic matching of cooling capacity.

[0047] Furthermore, the calculation of the fuel cell stack's heat output needs to consider key parameters such as the stack's rated power (e.g., 50kW), current density (0.8~1.2A / cm²), reactant gas temperature (typically 60~80℃), and ambient temperature (-20℃~45℃). The pump speed adjustment range is typically 500~3000 RPM, with a control cycle of 100~500 ms to ensure response speed and control accuracy. The temperature deviation threshold is generally set at ±2℃ to trigger pump speed adjustment.

[0048] Furthermore, this step is applicable to the operation control of fuel cell systems under frequent load changes (10%~100% Pe) and complex environmental temperature variations. Through real-time calculation and dynamic adjustment, the system can maintain stable stack temperature under different load and environmental conditions, avoiding degradation of membrane electrode performance due to excessively high temperature or impact on reaction efficiency due to excessively low temperature.

[0049] Furthermore, this step, through the combination of thermodynamic modeling and real-time feedback control, achieves precise matching of coolant flow rate, thereby improving the response speed and control accuracy of the system's thermal management. Simultaneously, by dynamically adjusting the water pump speed, it avoids the energy waste or insufficient heat dissipation problems associated with fixed flow control, improving the system's economic efficiency and reliability. This method provides fundamental data support for subsequent fan speed adaptive learning and is a key prerequisite for achieving overall thermal management adaptive control.

[0050] S22 corrects the thermostat angle by real-time feedback of the coolant temperature deviation from the target temperature to maintain it within the linear range.

[0051] Specifically, this step corrects the thermostat angle by real-time feedback of the coolant temperature deviation from the target temperature to maintain it within the linear range. This is a key control link in the adaptive control method for thermal management of fuel cell systems based on online learning algorithms in this invention. Its technical implementation principle is based on closed-loop feedback control and linear range identification of thermostat flow characteristics.

[0052] Furthermore, as the core component for flow distribution in the cooling circuit, the thermostat's angle adjustment directly affects the coolant flow ratio between the large and small circulation loops. The thermostat exhibits non-linear flow characteristics at different angles, typically displaying good linearity within a certain angle range (e.g., 15°~75°), which is the thermostat's linear control range. This step involves real-time acquisition of the temperature signal output from the coolant temperature sensor and comparison with a preset target temperature to calculate the temperature deviation. This deviation signal serves as feedback input, driving the control algorithm to dynamically correct the thermostat angle, ensuring it always operates within the linear flow range, thereby improving the accuracy and stability of temperature control.

[0053] Furthermore, the calculation period for temperature deviation is typically set to 100ms~500ms to balance real-time performance and system stability. When the temperature deviation exceeds a set threshold (e.g., ±2℃) and persists for a certain period (e.g., As=500ms, Bs=1000ms, etc.), the system will trigger the thermostat angle adjustment mechanism. The adjustment range is controlled in stages based on the magnitude and trend of the deviation, for example, using PID or fuzzy control strategies, combined with the thermostat's opening-flow response curve, to calculate the optimal angle increment, and then adjust the angle through an actuator (e.g., a stepper motor or servo valve).

[0054] Furthermore, this step is applicable to dynamic operating scenarios of fuel cell systems under different ambient temperatures (-20℃~45℃) and load conditions (10%~100%Pe). By maintaining the thermostat in a high linearity range, temperature oscillations and control lag caused by nonlinear regulation can be effectively avoided, thereby improving the overall thermal management response speed and robustness of the system.

[0055] Furthermore, through real-time feedback and angle correction mechanisms, adaptive optimization of thermostat control is achieved, providing a stable temperature reference for the subsequent online learning algorithm of fan speed. This is a fundamental link in the entire thermal management control strategy to achieve high precision, low energy consumption, and strong adaptability.

[0056] S3, divide the radiator fan speed into N speeds evenly according to the heat dissipation capacity, and establish a two-dimensional fan speed lookup table for ambient temperature and system operating power.

[0057] Specifically, this step divides the radiator fan speed into N levels evenly according to its heat dissipation capacity and establishes a two-dimensional fan speed lookup table for ambient temperature and system operating power. This is a key parameter initialization step in the adaptive control method for thermal management of fuel cell systems. Its technical implementation principle is based on the quantitative modeling of radiator fan performance and the matching strategy for thermal management requirements under multiple operating conditions.

[0058] Furthermore, the cooling capacity of the radiator fan must first be calibrated. Through experiments or simulations, the cooling power (in W) of the fan at different speeds or duty cycles is obtained and discretized into N levels (0-N+1), where level 0 represents the fan being off and level N+1 represents the maximum cooling power. The division of intermediate levels must satisfy a linear increase in cooling capacity or a non-linear optimization allocation based on the cooling efficiency curve to ensure the continuity and controllability of the system's cooling capacity when switching between different levels. The fan level division should be based on the system's thermal management control accuracy requirements; typically, N ranges from 5 to 10 levels to balance control resolution and system response efficiency.

[0059] Furthermore, a two-dimensional fan speed lookup table is established, with the horizontal axis representing ambient temperature (°C) and the vertical axis representing system operating power (%Pe). Each cell in the table corresponds to a fan speed value. The initial speed configuration of this lookup table can be based on system simulation results or empirical calibration, but it must at least meet the minimum heat dissipation requirements to ensure the thermal management safety of the system under extreme conditions. The construction of the lookup table needs to consider the steady-state heat dissipation requirements of the fuel cell system under different power outputs and ambient temperatures. Typically, the ambient temperature range is set to -20°C to 50°C, and the system power range is 10% to 100%Pe. The step size can be set to 5%Pe and 5°C to ensure the coverage density and control accuracy of the lookup table.

[0060] Furthermore, this step provides an initial control strategy framework for subsequent online learning algorithms. During system operation, the control device quickly retrieves the corresponding fan speed from a lookup table based on the current ambient temperature and system power, achieving initial matching of heat dissipation capacity. This lookup table mechanism effectively reduces the dependence on initial system control parameters and improves the adaptability and robustness of the control strategy.

[0061] Furthermore, this step, through the quantitative division of fan speeds and the establishment of a lookup table mechanism, provides the system with a structured and scalable control parameter space, offering a foundational data structure for subsequent online learning based on genetic algorithms. Simultaneously, uniform speed division avoids abrupt changes in heat dissipation capacity caused by fan start-stop cycles, thereby improving the stability and economy of system temperature control.

[0062] Furthermore, S3 includes: S31 divides the fan speed into N+1 levels based on the fan speed-heat dissipation power characteristic curve, where level 0 corresponds to the fan stopping and level N+1 corresponds to the maximum heat dissipation power.

[0063] Specifically, this step involves dividing the fan speed into N+1 levels based on the radiator fan's speed-heat dissipation power characteristic curve, where level 0 corresponds to the fan stopping and level N+1 corresponds to the maximum heat dissipation power. This operation is a key parameter configuration step in the adaptive control method for thermal management of fuel cell systems, aiming to achieve refined hierarchical control of fan heat dissipation capacity, thereby improving the accuracy and economy of thermal management under different ambient temperatures and operating conditions.

[0064] Further, this step first requires obtaining the fan's speed-heat dissipation power characteristic curve. This curve is usually obtained through experimental testing or CFD (Computational Fluid Dynamics) simulation, reflecting the nonlinear relationship between the fan's heat dissipation power (in W) and airflow (in m³ / s) at different speeds. In some implementations, the fan speed range is divided into N speeds using linear interpolation, with each speed range (e.g., 0~6000 RPM) proportionally divided into N speeds. Each speed corresponds to a fixed PWM duty cycle (e.g., 0%~100%), which is then mapped to the corresponding heat dissipation power value. Speed ​​0 corresponds to a 0% duty cycle, meaning the fan is completely off; speed N+1 corresponds to a 100% duty cycle, meaning the fan runs at full speed, achieving maximum heat dissipation power.

[0065] Furthermore, the value of N is typically between 5 and 10 to balance control accuracy and system complexity. For example, if N=8, the fan speed range is 0-9, with each speed corresponding to approximately 11.1% duty cycle step. The fan's cooling capacity must meet the system's cooling requirements under different ambient temperatures (e.g., -20℃ to 50℃) and operating power (e.g., 10% to 100% Pe), and the cooling capacity of each speed range must be verified in the system simulation model to ensure sufficient adjustment margin when operating within the thermostat's linear range.

[0066] Furthermore, this fan speed setting strategy is applicable to fuel cell systems operating under frequent load variations and complex ambient temperature changes. By matching the fan speed with system power and ambient temperature using a two-dimensional lookup table, the control device can call the optimal speed in real time, ensuring that the coolant temperature remains stable within the set range, while avoiding energy waste and system oscillation caused by frequent fan starts and stops or speed switching.

[0067] Furthermore, by finely dividing the fan speed settings, the cooling capacity becomes continuously adjustable, thereby improving the response speed and control accuracy of the system's thermal management. Simultaneously, the standardized and modular design of the speed settings enhances the compatibility and adaptability of the control algorithm to different fan models, providing a structured parameter space for subsequent online learning control based on genetic algorithms. This forms the foundation for achieving adaptive control and iterative parameter optimization.

[0068] S32 optimizes the configuration parameters of the initial gear table based on the combined operating conditions of ambient temperature and system operating power using a genetic algorithm.

[0069] Specifically, this step, based on the combined operating conditions of ambient temperature and system operating power, optimizes the configuration parameters of the initial fan speed table using a genetic algorithm. This is one of the core components of achieving adaptive thermal management control of the fuel cell system in this invention. The technical principle behind this is to construct a two-dimensional fan speed lookup table model with ambient temperature and system operating power as input variables, and then use a genetic algorithm (GA) to optimize the fan control parameters in the initial speed table online, thereby improving the accuracy and robustness of the system's thermal management control under different operating conditions.

[0070] Furthermore, the initial gear selection table configuration parameters include fan speed (0~N+1), fan speed / duty cycle, heat dissipation capacity threshold, etc. Their initial values ​​can be set based on multiphysics simulation results or empirical calibration, but these are usually difficult to perfectly match the complex operating conditions in actual operation. In this process, a genetic algorithm serves as an optimization tool, iteratively optimizing the parameters in the initial gear selection table by simulating biological evolution mechanisms. The specific operation includes: first, dividing the ambient temperature and system operating power into several discrete intervals to form a two-dimensional operating condition grid; second, using the fan speed corresponding to each operating point as a chromosome code to construct an initial population; and further, evaluating whether the thermostat angle under the current gear configuration is within its linear range of flow distribution (e.g., ...). Figure 2 (as shown in the figure), and a comprehensive evaluation is conducted in conjunction with indicators such as cooling temperature stability and system energy consumption.

[0071] Furthermore, the fitness function typically includes temperature deviation (ΔT), the magnitude of the thermostat angle deviating from the linear region, and the fan energy consumption coefficient (such as the linear relationship between speed and power consumption). The temperature deviation threshold can be set to ±Z℃ (Z is a preset value, such as ±1.5℃), and the thermostat linear region angle range can be determined based on experimental data, such as setting it between 15° and 75°. Parameters such as the number of iterations, population size, crossover rate (e.g., 0.8~0.95), and mutation rate (e.g., 0.01~0.1) of the genetic algorithm need to be reasonably configured according to the system response speed and computational resources to ensure the convergence and real-time performance of the optimization process.

[0072] Furthermore, this step applies to the dynamic operation of fuel cell systems under different ambient temperatures (e.g., -20℃ to 40℃) and load power (e.g., 10% to 100% Pe). Through the online learning mechanism of the genetic algorithm, the system can automatically update the gear table parameters after each run, achieving continuous optimization and adaptive improvement of the control strategy.

[0073] Furthermore, by introducing a genetic algorithm to optimize the initial gear table, the problems of temperature control response lag caused by changes in ambient temperature and system power fluctuations, and temperature oscillation caused by nonlinear adjustment of the thermostat, which are common in traditional control methods, are effectively solved. At the same time, this method improves the system's compatibility and adaptability to different types of radiators, reduces the accuracy requirements of initial calibration, and enhances the robustness and economy of thermal management control.

[0074] S4 uses an online learning algorithm to perform steady-state load tests on the full-power points in a two-dimensional lookup table, and dynamically adjusts the fan speed parameters based on the stable temperature state. Specifically, in some implementations, the step of performing steady-state load tests on the full power point in the two-dimensional lookup table through an online learning algorithm, and dynamically adjusting the fan speed parameters according to the temperature stability state is the core execution link in the thermal management adaptive control method of this invention. Its technical implementation is based on an online learning mechanism of multi-parameter fusion, combined with real-time feedback of fuel cell system operating conditions and ambient temperature, to achieve adaptive optimization of fan speed parameters.

[0075] Further, this step first constructs a two-dimensional fan speed lookup table (FSLT) with ambient temperature and system operating power as input dimensions, where each power point corresponds to an initial fan speed value. During system operation, the control device collects real-time data such as fuel cell outlet temperature, ambient temperature, and system output power to perform steady-state load tests on each power point. The criteria for judging the steady-state load test are: the deviation between the fuel cell outlet temperature and the target temperature is less than ±Z℃ (e.g., ±1℃), and the fan speed remains stable under the current operating conditions without frequent switching. When the steady-state conditions are met, the system enters the next operating point test, forming a complete self-learning process.

[0076] Furthermore, the fan speed is divided into N levels (e.g., N=8), with level 0 corresponding to fan stop, level N+1 corresponding to maximum heat dissipation capacity, and intermediate levels allocating fan speed or duty cycle linearly (e.g., 0%~100% duty cycle). The temperature deviation threshold Z is typically set to 1℃, and the temperature change trend is determined using a differential detection algorithm with a sampling interval of Xms (e.g., 100ms). The temperature change threshold Y is set to 0.2℃. When the temperature continues to rise or fall beyond the set threshold, the system triggers the fan speed increase or decrease logic to maintain the thermostat angle within the linear flow range.

[0077] Furthermore, this step is applicable to dynamic operating scenarios of fuel cell systems under different ambient temperatures (e.g., -20℃ to 40℃) and load variations (10% to 100% Pe). Through online learning algorithms, the system can automatically identify the nonlinear relationship between fan speed and heat dissipation requirements. Especially when the fan model is changed or the ambient temperature changes abruptly, parameter adaptive adjustment can be achieved without the need for extensive re-simulation calibration.

[0078] Furthermore, this step effectively solves the temperature control instability problem caused by traditional PID control or fixed lookup table methods within the nonlinear range of the thermostat. By optimizing the fan speed point by point through an online learning algorithm, the system can achieve stable control of the thermostat angle under different operating conditions, improving thermal management accuracy and response speed. At the same time, this method has parameter inheritance and iterative optimization capabilities, significantly enhancing the robustness and adaptability of the system control, and reducing the complexity of initial calibration and maintenance costs.

[0079] Furthermore, S4 includes: S41 uses a differential temperature detection method to determine the temperature change trend. When the temperature continues to rise and the thermostat opening exceeds the upper limit of the linear range, it triggers an upgrade operation.

[0080] Specifically, this step employs a differential temperature detection method to determine the temperature change trend and triggers an upgrade operation when the temperature continues to rise and the thermostat opening exceeds the upper limit of the linear range. This is one of the key steps in achieving adaptive thermal management control of the fuel cell system in this invention. Its technical implementation principle is based on real-time monitoring of dynamic temperature changes in the cooling system and boundary judgment of the thermostat's flow regulation capability. This allows for timely adjustment of the fan speed when the system's heat dissipation capacity is insufficient, ensuring that the stack temperature remains stable within a safe range.

[0081] Furthermore, the differential temperature detection method collects the difference between the coolant outlet temperature and the target temperature, and performs differential calculations at fixed time intervals (e.g., Xms) to determine the temperature change trend. When the counter value exceeds the set temperature rise judgment threshold, the system determines that the temperature is in a continuous upward trend.

[0082] Furthermore, the temperature rise threshold is set between 10 and 30 based on system stability requirements. The upper limit of the linear range of the thermostat opening is usually determined experimentally or through simulation. For example, when the thermostat angle is between 60° and 80°, its flow regulation characteristics have good linearity. Beyond this range, it enters the nonlinear region, leading to a decrease in control accuracy.

[0083] Furthermore, this step applies to fuel cell systems operating under high load or high temperature conditions. When system power increases or ambient temperature rises, the heat generated by the fuel cell stack increases. If the thermostat is already at its maximum linear opening and still cannot meet the heat dissipation requirements, the fan speed is increased or more fans are turned on by upgrading the speed setting to improve heat dissipation capacity and prevent the fuel cell stack temperature from exceeding the limit.

[0084] Furthermore, by jointly determining the differential temperature detection and thermostat opening, intelligent upshifting control of the fan speed is achieved, avoiding temperature fluctuations caused by frequent thermostat adjustments in the nonlinear range, and improving the stability and response speed of the system's thermal management. Simultaneously, this method enhances the adaptive capability of the control strategy, improving the robustness and economy of the fuel cell system under complex operating conditions.

[0085] The S42 uses an intelligent counting algorithm to quantitatively evaluate temperature trends. When the count value exceeds a preset threshold, it executes fan speed switching and records steady-state operating data.

[0086] Specifically, this step uses an intelligent counting algorithm to quantitatively evaluate the temperature trend. When the count value exceeds a preset threshold, the fan speed is switched and steady-state operating data is recorded. This is a key dynamic adjustment link in the thermal management adaptive control method of this invention. Its core lies in achieving intelligent switching of the fan speed through real-time analysis of temperature change trends, thereby maintaining the stable operation of the fuel cell stack within the linear range of the thermostat flow.

[0087] Furthermore, this step employs an intelligent counting algorithm based on differential temperature detection. The system collects the coolant outlet temperature of the fuel cell stack at a fixed sampling period (e.g., Xms) and compares it with the target temperature to calculate the temperature deviation. The counter's baseline value is set to 50. When the count value exceeds a preset threshold (e.g., 60), it is determined that the temperature is continuously rising, triggering the fan to increase its speed; when the count value is below the threshold (e.g., 40), it is determined that the temperature is continuously falling, triggering the fan to decrease its speed. This algorithm effectively avoids interference from instantaneous temperature fluctuations on control decisions by quantifying the continuity of temperature changes.

[0088] Furthermore, this step is applicable to the dynamic operation of fuel cell systems under different ambient temperatures (e.g., -20℃ to 45℃) and load variations (10% to 100% Pe). By monitoring temperature trends in real time and combining this with a fan speed switching strategy, the system can maintain the stack temperature within the set range under complex operating conditions, while ensuring that the thermostat always operates within the flow linear range, thus improving control accuracy and stability.

[0089] Furthermore, this step, by introducing an intelligent counting mechanism, achieves robust identification of temperature trends, effectively improving the timing control capability of fan speed switching. Combined with an online learning algorithm, the system can optimize fan speed parameters in each run, thereby enhancing the adaptability and economy of thermal management control, reducing dependence on initial calibration parameters, and improving the overall intelligence level of the control system.

[0090] S5 stores the current optimal fan speed parameters to the control device hardware when the system is shut down, and iteratively updates the initial calibration parameters to form a closed-loop learning.

[0091] Specifically, storing the current optimal fan speed parameters to the control device hardware when the system shuts down, and iteratively updating the initial calibration parameters to form a closed-loop learning process, is a key step in this invention for achieving adaptive optimization of thermal management control parameters. This step, based on an online learning mechanism using a genetic algorithm, involves the coordinated operation of the control device's hardware and software after system operation. The fan speed parameters optimized through the adaptive algorithm during operation are stored and iteratively updated with the initial calibration parameters, thereby constructing a closed-loop learning system and continuously improving the adaptability and robustness of the control strategy.

[0092] Furthermore, during system shutdown, the control device initiates a parameter storage process, acquiring fan speed parameters under current operating conditions via the CAN bus or internal communication interface. This includes key variables such as fan speed at each power point, thermostat angle feedback value, and coolant temperature deviation. These parameters are written into the control device's non-volatile memory (such as EEPROM or Flash) to ensure retention after power failure. Simultaneously, the system uses a genetic algorithm to compare and analyze current operating parameters with historical parameters, employing crossover, mutation, and selection operations to generate a new initial fan speed calibration table for initializing the control strategy upon the next system startup.

[0093] Furthermore, the storage and updating of fan speed parameters must meet certain accuracy and response requirements. For example, the fan speed may be divided into N speeds (e.g., N=8), each corresponding to a specific PWM duty cycle or speed setting. Speed ​​switching must meet a minimum temperature difference threshold (e.g., ±0.5℃) and a duration condition (e.g., duration ≥30s) to avoid system disturbances caused by frequent switching. The iteration update frequency of the genetic algorithm is usually set to once per running cycle. The convergence speed of adaptive learning is closely related to the population size (e.g., 20~50 individuals) to ensure the efficiency and stability of parameter optimization.

[0094] Furthermore, this step is applicable to the long-term operation of fuel cell systems under different ambient temperatures (e.g., -20°C to 50°C) and load variations (e.g., 10% to 100% Pe). Through a closed-loop learning mechanism, the system can automatically adapt to factors such as fan aging, environmental changes, or stack performance drift, without the need for manual recalibration, significantly reducing maintenance costs and debugging complexity.

[0095] The technical advantage of this step lies in enabling the control strategy to have self-learning capabilities through continuous iterative updates of parameters, thereby improving the system's thermal management stability and economy under complex operating conditions. Simultaneously, the parameter storage mechanism ensures that the control device accumulates optimization experience through multiple runs, enhancing the system's adaptability to different radiator models and fan configurations, thus achieving intelligent and adaptive thermal management control of the fuel cell system.

[0096] Furthermore, S5 includes: S51, based on a genetic algorithm, performs crossover and mutation operations on the fan speed parameters of this run to generate the next generation of optimized parameter set. Specifically, in this invention, performing crossover and mutation operations on the fan speed parameters obtained in the current operation based on a genetic algorithm to generate the next generation of optimized parameter sets is one of the core steps in realizing adaptive thermal management control of the fuel cell system. This step optimizes and iterates the fan speed parameters obtained in the current operation by simulating a biological evolution mechanism, thereby improving the temperature control accuracy and robustness of the system under different ambient temperatures and operating power.

[0097] Furthermore, the crossover operation of the genetic algorithm employs a simulated binary crossover (SBX) strategy with real-number encoding, exchanging gene fragments of the fan speed parameters that perform well in the current population to generate new parameter combinations. Specifically, each fan speed parameter can be represented as a real-number vector, corresponding to each power point in a two-dimensional lookup table of ambient temperature and system operating power.

[0098] Furthermore, the fan speed parameter set includes, but is not limited to, fan speed (RPM), duty cycle (%), and heat dissipation power (W), and its optimization objective function is a weighted sum between minimizing system temperature deviation and minimizing fan energy consumption.

[0099] Furthermore, this step is typically performed after the system completes a full steady-state load test, i.e., during the system shutdown phase. The control device uses an embedded processor to process the fan speed parameters collected during this operation using a genetic algorithm. This process iteratively updates the parameters in the control device's non-volatile memory (such as Flash or EEPROM) to ensure that better initial parameters are used for the next system run.

[0100] Furthermore, this step, through crossover and variation operations, effectively avoids the control lag and nonlinear oscillation problems that occur in traditional PID control or fixed lookup table methods under complex operating conditions. Simultaneously, its online learning capability significantly improves the system's adaptability to different fan models, ambient temperature changes, and fluctuations in stack heat generation, thereby enhancing the thermal management stability and economy of the fuel cell system.

[0101] S52 compares the optimized parameter set with the initial calibration parameters and selects the parameter with the highest fitness value as the initial value for the next run.

[0102] Specifically, in the thermal management adaptive control method of this invention, comparing the optimized parameter set with the initial calibration parameters and selecting the parameter with the highest fitness value as the initial value for the next run is a key step in achieving iterative optimization of control parameters and improving system robustness. This step is based on the online learning mechanism of a genetic algorithm (GA), which quantitatively evaluates the control effect of different parameter combinations in actual operation, thereby achieving dynamic updating and inheritance of parameters.

[0103] Further, this step first calculates the fitness function value of the current parameter set using real-time data collected during system operation (such as fuel cell stack temperature, coolant flow rate, thermostat opening, fan speed, etc.). The fitness function typically employs a multi-objective optimization strategy, for example, using the integral absolute value (IAE) of temperature deviation, thermostat angle fluctuation amplitude, fan energy consumption, etc., as evaluation indicators to construct a weighted comprehensive scoring model. After completing the steady-state load test at the current operating point, the system compares the current operating parameters with the initial calibration parameters, and judges its control performance based on the fitness value. In some implementations, fuzzy logic or PID control error integrals can be introduced into the calculation of the fitness value to enhance the adaptability to complex operating conditions.

[0104] Furthermore, the calculation of fitness values ​​requires the establishment of clear quantitative standards, such as temperature deviation threshold (±Z℃), thermostat angle linear range (θ_low to θ_high), and fan speed switching frequency (≤X times / minute). After each run, the system will record the fan speed, thermostat angle, and temperature response curve corresponding to each operating point, and score them based on the preset fitness function. The parameter set with the highest fitness value will be marked as the optimal solution and used as the initial parameter input for the next run.

[0105] Furthermore, this step is applicable to multiple start-up and shutdown operations of fuel cell systems under varying ambient temperatures (e.g., -20°C to 45°C) and load changes (10% to 100% Pe). During system shutdown, the control device writes the current optimal parameters into a non-volatile memory (e.g., Flash or EEPROM) via embedded software, ensuring that the parameters can be retrieved upon the next startup. This mechanism is particularly suitable for multi-fan radiator systems, automatically adapting to different fan models and cooling capacities without requiring manual recalibration.

[0106] Furthermore, this step, through iterative parameter updates, enables the system to continuously optimize its control strategy under different operating conditions, thereby improving the response speed, control accuracy, and system economy of thermal management. Simultaneously, through the self-learning capability of the genetic algorithm, the system can gradually converge to the optimal combination of control parameters, enhancing its robustness to uncertainties such as fan failures and environmental disturbances, and significantly reducing the dependence on initial calibration parameters and simulation workload.

[0107] The adaptive control method and apparatus for thermal management of fuel cell systems based on online learning algorithms in this invention can improve the thermal management response speed and control stability of fuel cell systems under complex operating conditions. By using online learning algorithms to achieve adaptive adjustment of fan speed, the redundant configuration of heat dissipation configuration is reduced, thereby enhancing the economy and robustness of the system.

[0108] Example 2 This invention proposes an adaptive control system for thermal management of fuel cell systems based on an online learning algorithm. Utilizing online learning adaptive control of multiple fan speeds on the radiator, it ensures the thermostat adjustment angle remains within a high linearity range, improving the accuracy and stability of coolant temperature control. Furthermore, it enhances the algorithm's versatility and the system's robustness in thermal management, reducing the workload of system heat dissipation simulation and the accuracy requirements of initial calibration parameters. This system is applicable to radiators of different models and heat dissipation capacities. The online learning control based on a genetic algorithm adjusts, corrects, and iteratively stores fan speed parameters in real time to address application scenarios with varying ambient temperatures and structural layouts, thereby improving the system's thermal management stability and economy.

[0109] This invention relates to an adaptive control method and control device for thermal management of a fuel cell system based on an online learning algorithm. Compared to the joint control of fixed fan numbering or random grouping based on heat dissipation capacity with thermostats, which relies heavily on the accuracy of simulation calculations, this invention emphasizes automatically adjusting and adapting fan speeds based on an online learning algorithm to meet the heat dissipation requirements of the fuel cell system under the current ambient temperature, while ensuring that the thermostat angle always operates within the linear region of flow distribution. This ensures the stability of system thermal management temperature control, improves control accuracy, avoids excessive heat dissipation capacity configuration, and enhances system operating economy. After system operation, the adaptive parameters of the fan speed for each run are iteratively replaced with the initial calibration parameters through the software and hardware functions of the control device, and the online-learned parameters are stored in the hardware storage space of the control device. The control device, based on a genetic algorithm, continuously and automatically adapts and iterates to the impact of changes in ambient temperature or fan-related issues on the system's fan speed parameters, thereby improving the robustness of the system's thermal management control.

[0110] In one embodiment of the present invention, the linear range of the relationship between the thermostat angle and the large circulation flow rate is confirmed; the required cooling flow rate and water pump speed are calculated based on the heat generation of the fuel cell stack at different operating points, and it is ensured that each flow rate is within the linear range of the thermostat angle; the fan speed is evenly divided into N speeds according to the heat dissipation capacity; a two-dimensional fan speed lookup table of ambient temperature and system operating power is compiled (the initial speed can be arbitrarily configured, at least ensuring the minimum heat dissipation capacity of the radiator); an adaptive online learning algorithm for fan speed is imported, and steady-state load tests are performed on each of the full power points of the two-dimensional lookup table of ambient temperature and system operating power (the next operating point test can be carried out once the cooling temperature and fan speed are stable, i.e., the online self-learning process of fan speed); after all operating point tests are completed, the system is shut down. During the shutdown process, the self-learning parameters of the fan speed are iteratively stored through the software and hardware functions of the control device as the initial fan speed for the next system operation.

[0111] Specifically, Figure 2 This is a schematic diagram of a cooling subsystem loop in a fuel cell system. The control device is the main actuator for realizing online learning control of thermal management in the fuel cell cooling subsystem, and it is implemented through a combination of software and hardware.

[0112] Specifically, Figure 3 This is a schematic diagram showing the relationship between the thermostat angle and the large circulation flow rate. The linearity of the flow rate change between the lower limit and the upper limit of the linear zone is good, which is conducive to improving the temperature control accuracy of thermal management and is the control target range.

[0113] Specifically, Figure 4This is a functional block diagram of an adaptive online learning fan speed control system for a fuel cell system. The radiator speed control employs an N-level (0-N+1) fine-tuning scheme, where level 0 is the off state, level N+1 corresponds to the maximum heat dissipation power, and the intermediate levels linearly distribute the fan speed / duty cycle. The design of this control system is based on multiphysics simulation results, comprehensively considering parameters such as ambient temperature, system power, and radiator characteristics to determine the initial speed calibration values. In actual operation, the system... Figure 4 The intelligent adjustment algorithm shown achieves dynamic control. This control strategy, through multi-parameter fusion judgment (valve opening, absolute temperature deviation, and dynamic change trend), ensures heat dissipation safety under extreme operating conditions, optimizes energy by avoiding unnecessary gear switching, keeps the thermostat angle always operating within the flow linear range, and ensures the radiator gear operates within the optimal efficiency range, thus achieving efficient and precise control of the system's thermal management.

[0114] Furthermore, the conditions for upgrading are: Temperature control valve opening > upper limit of linear range (insufficient large circulation flow regulation capability). And simultaneously meet any of the following over-temperature conditions: When the upper temperature difference threshold 1 is reached, the temperature continues to rise (As); when the upper temperature difference threshold 2 is reached, the temperature continues to rise (Bs); when the upper temperature difference threshold 3 is reached, the temperature continues to rise (Cs); and the temperature continues to rise. (Confirmed by differential temperature detection at Xms intervals, when the temperature rise count > temperature rise judgment threshold) Furthermore, the conditions for downgrading are: Temperature control valve opening degree < lower limit of linear range (excessive small circulation flow regulation capacity) And simultaneously meet any of the following under-temperature conditions: The temperature continues to decrease after reaching the lower temperature difference threshold 1 for Ds; after reaching the lower temperature difference threshold 2 for Es; after reaching the lower temperature difference threshold 3 for Fs; and the temperature continues to decrease. (Confirmed when the temperature rise count is less than the temperature drop judgment threshold by differential temperature detection at Xms intervals) Temperature trend determination employs an intelligent counting algorithm: using 50 as a baseline, the current temperature is compared with the temperature of the previous sampling point every X ms. If the difference is greater than Y℃, the count is incremented by 1; if it is less than -Y℃, the count is decremented by 1, remaining unchanged within the range of ±Y℃. When the deviation between the actual temperature and the target temperature returns to within ±Z℃, the count is reset to the baseline value of 50.

[0115] Furthermore, Figure 5 This is a schematic diagram of the online learning control process for thermal management of a fuel cell system based on a genetic algorithm. It mainly consists of five parts: thermostat linear range confirmation, stack flow rate and thermostat control range matching, initial fan speed table formulation, fan speed adaptive online learning, and speed parameter storage. By executing the above five parts in sequence, the online learning control for thermal management of the fuel cell system based on the genetic algorithm can be completed.

[0116] In one embodiment of the present invention, the linear range of the relationship between the thermostat angle and the large circulation flow rate is confirmed; the required cooling flow rate and water pump speed are calculated based on the heat generation of the fuel cell stack at different operating points, and it is ensured that each flow rate is within the linear range of the thermostat angle; the fan speed is evenly divided into N speeds according to the heat dissipation capacity; a two-dimensional fan speed lookup table of ambient temperature and system operating power is compiled (the initial speed can be arbitrarily configured, at least ensuring the minimum heat dissipation capacity of the radiator); an adaptive online learning algorithm for fan speed is imported, and steady-state load tests are performed on each of the full power points of the two-dimensional lookup table of ambient temperature and system operating power (the next operating point test can be carried out once the cooling temperature and fan speed are stable, i.e., the online self-learning process of fan speed); after all operating point tests are completed, the system is shut down. During the shutdown process, the self-learning parameters of the fan speed are iteratively stored through the software and hardware functions of the control device as the initial fan speed for the next system operation.

[0117] Example 3 To achieve the above embodiments, such as Figure 6 As shown, this embodiment also provides a fuel cell system thermal management adaptive control device 10 based on an online learning algorithm, including: The linear interval determination module 100 is used to determine the linear interval range of the relationship between the thermostat angle and the large circulation flow rate; The operating condition control module 200 is used to calculate the cooling flow rate and water pump speed based on the heat generation at different operating points of the fuel cell stack, and to control the thermostat angle within the linear range. The fan speed configuration module 300 is used to evenly divide the heat sink fan speed into N speeds according to the heat dissipation capacity, and to establish a two-dimensional fan speed lookup table of ambient temperature and system operating power. The online learning adjustment module 400 is used to perform steady-state load tests on the full power points in the two-dimensional lookup table through an online learning algorithm, and dynamically adjust the fan speed parameters according to the temperature stability state. The parameter storage and update module 500 is used to store the current optimal fan speed parameters to the control device hardware when the system is shut down, and iteratively update the initial calibration parameters to form a closed-loop learning.

[0118] Furthermore, the linear interval determination module 100 is also used for: The nonlinear relationship curve between the thermostat angle and the large circulation flow rate was calculated using multiphysics simulation. The nonlinear relationship curve is fitted based on experimental test data, and the upper and lower limits of the linear interval are determined based on the fitting results.

[0119] Furthermore, the operating condition control module 200 is also used for: A thermodynamic model was used to calculate the real-time heat generation of the fuel cell stack at the target temperature, and the pump speed was dynamically adjusted based on the heat generation. By analyzing the real-time feedback of the coolant temperature deviation from the target temperature, the thermostat angle is adjusted to maintain it within the linear range.

[0120] Furthermore, the fan speed configuration module 300 is also used for: Based on the fan speed-heat dissipation power characteristic curve, the fan speed is divided into N+1 levels, where level 0 corresponds to the fan stopping and level N+1 corresponds to the maximum heat dissipation power. Based on the combined operating conditions of ambient temperature and system operating power, the configuration parameters of the initial gear table are optimized using a genetic algorithm.

[0121] Furthermore, the online learning adjustment module 400 is also used for: The differential temperature detection method is used to determine the temperature change trend. When the temperature continues to rise and the thermostat opening exceeds the upper limit of the linear range, the upshift operation is triggered. The temperature trend is quantitatively evaluated by an intelligent counting algorithm. When the count value exceeds the preset threshold, the fan speed is switched and steady-state operating data is recorded.

[0122] The adaptive control device for thermal management of fuel cell systems based on online learning algorithms in this invention can improve the thermal management response speed and control stability of fuel cell systems under complex operating conditions. It achieves adaptive adjustment of fan speed through online learning algorithms, reduces redundant heat dissipation configuration, and enhances system economy and robustness.

[0123] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0124] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

Claims

1. An adaptive control method for thermal management of a fuel cell system based on an online learning algorithm, characterized in that, include: S1, determine the linear range of the relationship between the thermostat angle and the large circulation flow rate; S2, calculate the cooling flow rate and water pump speed based on the heat generation of the fuel cell stack at different operating conditions, and control the thermostat angle within the linear range; S3, divide the radiator fan speed into N speeds evenly according to the heat dissipation capacity, and establish a two-dimensional fan speed lookup table for ambient temperature and system operating power; S4 uses an online learning algorithm to perform steady-state load tests on the full-power points in the two-dimensional lookup table, dynamically adjusting the fan speed parameters based on the stable temperature state. The criteria for judging the steady-state load test are: the deviation between the stack outlet temperature and the target temperature is less than ±Z℃, and the fan speed remains stable under the current operating conditions without frequent switching. When the steady-state conditions are met, the system enters the next operating point test, forming a complete self-learning process. A differential temperature detection method is used to determine the temperature change trend. When the temperature continues to rise and the thermostat opening exceeds the upper limit of the linear range, a speed increase operation is triggered. An intelligent counting algorithm is used to quantitatively evaluate the temperature trend. When the count value exceeds the preset threshold, the fan speed is switched and the steady-state operating data is recorded. S5 stores the current optimal fan speed parameters to the control device hardware when the system is shut down, and iteratively updates the initial calibration parameters to form a closed-loop learning.

2. The method as described in claim 1, characterized in that, The S1 further includes: S11, calculates the nonlinear relationship curve between thermostat angle and large circulation flow rate through multiphysics simulation; S12, Fit the nonlinear relationship curve based on the experimental test data, and determine the upper and lower limits of the linear interval based on the fitting results.

3. The method as described in claim 1, characterized in that, The S2 further includes: S21 uses a thermodynamic model to calculate the real-time heat generation of the fuel cell stack at the target temperature and dynamically adjusts the pump speed based on the heat generation. S22 corrects the thermostat angle by real-time feedback of the coolant temperature deviation from the target temperature to maintain it within the linear range.

4. The method as described in claim 1, characterized in that, The S3 further includes: S31, based on the fan speed-heat dissipation power characteristic curve, divides the fan speed into N+1 levels, where level 0 corresponds to the fan stopping and level N+1 corresponds to the maximum heat dissipation power; S32 optimizes the configuration parameters of the initial gear table based on the combined operating conditions of ambient temperature and system operating power using a genetic algorithm.

5. The method as described in claim 1, characterized in that, The S5 also includes: S51, based on the genetic algorithm, performs crossover and mutation operations on the fan speed parameters of this run to generate the next generation of optimized parameter set; S52 compares the optimized parameter set with the initial calibration parameters and selects the parameter with the highest fitness value as the initial value for the next run.

6. An adaptive control device for thermal management of a fuel cell system based on an online learning algorithm, characterized in that, include: The linear interval determination module is used to determine the linear interval range of the relationship between the thermostat angle and the large circulation flow rate; The operating condition control module is used to calculate the cooling flow rate and water pump speed based on the heat generation of the fuel cell stack at different operating conditions, and to control the thermostat angle within the linear range. The fan speed configuration module is used to evenly divide the heat sink fan speed into N speeds according to the heat dissipation capacity, and to establish a two-dimensional fan speed lookup table of ambient temperature and system operating power. The online learning and adjustment module is used to perform steady-state load tests on the full-power points in the two-dimensional lookup table using an online learning algorithm, and dynamically adjusts the fan speed parameters according to the temperature stability. The criteria for judging the steady-state load test are: the deviation between the stack outlet temperature and the target temperature is less than ±Z℃, and the fan speed remains stable under the current operating conditions without frequent switching. When the steady-state conditions are met, the system enters the next operating point test, forming a complete self-learning process. The differential temperature detection method is used to judge the temperature change trend. When the temperature continues to rise and the thermostat opening exceeds the upper limit of the linear range, the speed increase operation is triggered. The intelligent counting algorithm is used to quantitatively evaluate the temperature trend. When the count value exceeds the preset threshold, the fan speed is switched and the steady-state operating data is recorded. The parameter storage and update module is used to store the current optimal fan speed parameters to the control device hardware when the system is shut down, and iteratively update the initial calibration parameters to form a closed-loop learning.

7. The apparatus as claimed in claim 6, characterized in that, The linear interval determination module is also used for: The nonlinear relationship curve between the thermostat angle and the large circulation flow rate was calculated using multiphysics simulation. The nonlinear relationship curve is fitted based on experimental test data, and the upper and lower limits of the linear interval are determined based on the fitting results.

8. The apparatus as claimed in claim 6, characterized in that, The operating condition control module is also used for: A thermodynamic model was used to calculate the real-time heat generation of the fuel cell stack at the target temperature, and the pump speed was dynamically adjusted based on the heat generation. By analyzing the real-time feedback of the coolant temperature deviation from the target temperature, the thermostat angle is adjusted to maintain it within the linear range.

9. The apparatus as claimed in claim 6, characterized in that, The fan speed configuration module is also used for: Based on the fan speed-heat dissipation power characteristic curve, the fan speed is divided into N+1 levels, where level 0 corresponds to the fan stopping and level N+1 corresponds to the maximum heat dissipation power. Based on the combined operating conditions of ambient temperature and system operating power, the configuration parameters of the initial gear table are optimized using a genetic algorithm.

Images