A method for controlling a fuel cell system of a drone at high altitude

Abstract

The application relates to the technical field of battery control, and discloses a high-altitude adaptive fuel cell system control method for an unmanned aerial vehicle, which comprises the following steps: collecting environment parameters in time synchronization and mapping the environment parameters into dimensionless parameters to establish a unified dynamic time base; generating a feedforward working condition reference instruction and a real reachable gas supply envelope affected by a Reynolds number according to load demand; monitoring the time sequence dislocation of gas path pressure building and oxygen supply, quantitatively extracting the pseudo-steady oxygen debt window strength and oxygen debt accumulation amount induced by the phenomenon of pressure first and flow later; synthesizing a closed-loop risk error based on the oxygen debt window strength and the oxygen debt accumulation amount, adaptively outputting an air inlet instruction correction compensation, and distributing the online identification matrix to each actuator; finally, dynamically limiting the current rise rate according to the gas supply margin, and updating the mapping parameters in real time through residual error self-learning. The application eliminates the oxygen supply pseudo-steady state misjudgment blind area under the high-altitude thin air condition, and realizes the anti-power-off safety endurance in the whole airspace.

Description

Technical Field

[0001] This invention relates to the field of battery control technology, and more specifically, to a control method for a high-altitude adaptive fuel cell system for unmanned aerial vehicles. Background Technology

[0002] With the increasing application of drones in high-altitude areas such as plateaus, fuel cells have become a highly advantageous power solution. To reduce the load on the fuselage, airborne air supply pipelines often adopt a series arrangement of miniature centrifugal air compressors, electronic throttle valves, and short flow channels.

[0003] However, atmospheric pressure and air density are significantly reduced at high altitudes. When an aircraft performs high-maneuver climbs or encounters sudden load increases, the air system must provide a large amount of oxygen in a very short time. Under the combined constraints of low Reynolds numbers and thin atmospheres, the working envelope of the micro compressor suffers severe degradation. Due to the compressibility of the gas itself and the sluggish transmission of airflow through the obstructed channels, the average pressure at the system inlet is often raised by the compressor first, but the effective oxygen molecule flow rate that actually reaches the fuel cell stack to participate in the electrochemical reaction arrives far later.

[0004] Existing conventional control methods generally use the rise in gas inlet pressure as the criterion for determining whether oxygen supply meets standards, failing to recognize the timing misalignment between pressure build-up and actual oxygen supply. When the system front-end pressure appears sufficient, the controller is prone to misjudging that gas supply is ready, thus allowing the fuel cell stack to significantly increase its output power within a blind spot where effective oxygen supply is severely delayed. This inevitably leads to a hidden oxygen-deficient operating state in the core reaction area of ​​the battery, which not only causes polarity reversal in individual cells and accelerated oxidation of the carbon carrier, but also triggers a severe accident of power supply collapse during periods of intense load increase. Summary of the Invention

[0005] This invention provides a control method for a high-altitude adaptive fuel cell system for unmanned aerial vehicles (UAVs), which solves the technical problems mentioned in the background art.

[0006] This invention provides a control method for a high-altitude adaptive fuel cell system for unmanned aerial vehicles (UAVs), applied to a series air circuit system comprising a micro centrifugal air compressor, an electronic throttle, a humidification buffer chamber, and a fuel cell stack cathode channel, including: Collect environmental information, gas path status information, and fuel cell stack electrical status information, convert them into dimensionless environmental physical property parameters, obtain normalized fuel cell stack current and normalized current change rate, and obtain normalized gas residence time. The feedforward operating condition reference command is obtained based on the normalized stack current and the normalized current change rate, the Reynolds number characterization term is calculated, and the true achievable gas supply envelope is obtained by correcting the Reynolds number characterization term. Obtain the pressure recovery degree and the oxygen excess recovery degree, and construct the pseudo steady-state oxygen debt window intensity and oxygen debt accumulation based on the time-series recovery difference between the two. The dynamically tuned baseline control gain is obtained, and the pseudo-steady-state oxygen debt window intensity and the accumulated oxygen debt are introduced into the closed-loop regulation system. The flow target correction and pressure target correction are generated by combining the dynamically tuned baseline control gain. The local Jacobian matrix of the series air circuit system is identified online, and the flow target correction and the pressure target correction are calculated into the speed command of the micro centrifugal air compressor and the opening command of the electronic throttle using the local Jacobian matrix. The current normalized stack voltage and the current load request current are obtained. Based on the true reachable gas supply envelope, the current normalized stack voltage and the pseudo steady-state oxygen debt window intensity, the allowable current rise rate is generated. The allowable current rise rate is used to limit the ramp-up of the output current and to obtain the protected power command. The model prediction error for this control cycle is calculated, and the mapping parameters of the feedforward operating condition baseline command and the actual reachable gas supply envelope are updated using a recursive algorithm that includes data quality weights.

[0007] The beneficial effects of this invention include: reconstructing the hidden misalignment evolution process of pressure rise precedes effective oxygen supply lag in lightweight series pipelines. By quantitatively extracting the pseudo-steady-state oxygen debt state and deeply integrating it into a fully closed-loop control chain of error risk weighting, dynamic gain compensation, separation command calculation, and power ramp-up boundary interception, this invention can strictly suppress the blind grabbing of fuel cell power within the high-risk window of false pressure compliance. Load commands are then smoothly released only after the actual oxygen supply capacity is fully realized. This control mechanism not only eradicates the hidden hypoxia damage induced by the collapse of environmental air supply capacity and the lag in long, narrow channels, but also, through its quality-weighted self-learning capability, ensures the extreme endurance stability and long battery life of the UAV under conditions of extreme cold, thin oxygen, and high-frequency maneuvering. Attached Figure Description

[0008] Figure 1 This is a flowchart of a control method for a high-altitude adaptive fuel cell system for unmanned aerial vehicles (UAVs) according to the present invention. Detailed Implementation

[0009] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0010] like Figure 1 As shown, a control method for a high-altitude adaptive fuel cell system for unmanned aerial vehicles (UAVs) is applied to a series air circuit system comprising a micro centrifugal air compressor, an electronic throttle, a humidification buffer chamber, and a fuel cell stack cathode channel, including: Collect environmental information, gas path status information, and fuel cell stack electrical status information, convert them into dimensionless environmental physical property parameters, obtain normalized fuel cell stack current and normalized current change rate, and obtain normalized gas residence time. The feedforward operating condition reference command is obtained based on the normalized stack current and the normalized current change rate, the Reynolds number characterization term is calculated, and the true achievable gas supply envelope is obtained by correcting the Reynolds number characterization term. Obtain the pressure recovery degree and the oxygen excess recovery degree, and construct the pseudo steady-state oxygen debt window intensity and oxygen debt accumulation based on the time-series recovery difference between the two. The dynamically tuned baseline control gain is obtained, and the pseudo-steady-state oxygen debt window intensity and the accumulated oxygen debt are introduced into the closed-loop regulation system. The flow target correction and pressure target correction are generated by combining the dynamically tuned baseline control gain. The local Jacobian matrix of the series air circuit system is identified online, and the flow target correction and the pressure target correction are calculated into the speed command of the micro centrifugal air compressor and the opening command of the electronic throttle using the local Jacobian matrix. The current normalized stack voltage and the current load request current are obtained. Based on the true reachable gas supply envelope, the current normalized stack voltage and the pseudo steady-state oxygen debt window intensity, the allowable current rise rate is generated. The allowable current rise rate is used to limit the ramp-up of the output current and to obtain the protected power command. The model prediction error for this control cycle is calculated, and the mapping parameters of the feedforward operating condition baseline command and the actual reachable gas supply envelope are updated using a recursive algorithm that includes data quality weights.

[0011] Preferably, the collected environmental information, gas path status information, and fuel cell stack electrical status information are converted into dimensionless environmental physical property parameters, and the normalized fuel cell stack current and normalized current change rate are obtained, along with the normalized gas residence time, including: Simultaneously acquire altitude, ambient temperature, ambient pressure, cathode inlet pressure, cathode gas temperature, humidification status information, air molar flow rate, and equivalent cathode gas volume; divide the altitude by the characteristic height of the international standard atmosphere to obtain the normalized altitude. Subtract the normalized altitude from the value 1 Normalized ambient temperature : The normalized ambient temperature The preset adiabatic index The power as normalized environmental pressure : The normalized ambient temperature The preset adiabatic index The normalized environmental density is obtained by subtracting the difference from 1 to the power of the difference. : The water vapor partial pressure ratio is obtained based on the humidification status information. Subtract the water vapor partial pressure ratio from the value 1 Determine the effective oxygen availability : The normalized equivalent cathode gas volume calculated from the equivalent cathode gas volume. The cathode feed pressure ratio calculated from the cathode inlet pressure With the normalized environmental pressure The continuous multiplication is used as the numerator, and the normalized cathode gas temperature calculated from the cathode gas temperature is... The normalized air molar flow rate calculated from the air molar flow rate The product plus the preset numerical stability term The normalized gas residence time is obtained by dividing the numerator by the denominator. : Altitude is the height information of the drone's current flight position relative to the mean sea level.

[0012] Ambient temperature refers to the temperature information of the incoming airflow to the drone.

[0013] Environmental pressure is the absolute pressure information of the external atmosphere in which the drone is located.

[0014] The cathode inlet pressure is the absolute pressure information at the inlet before air enters the cathode of the fuel cell stack. It can be obtained by a pressure sensor installed at the cathode inlet manifold.

[0015] The cathode gas temperature is the actual temperature of the gas at the cathode inlet. It can be obtained using a temperature sensor installed at the outlet of the humidification buffer chamber or in the cathode inlet pipe section.

[0016] Humidification status information is information that characterizes the humidity level of the cathode-supply gas. It can be obtained through a relative humidity sensor, a dew point sensor, and a humidifier operating status acquisition circuit.

[0017] Air molar flow rate is the amount of air entering the cathode gas path per unit time.

[0018] The equivalent cathode gas volume is the equivalent total gas volume after converting the outlet chamber of the micro centrifugal air compressor, the humidification buffer chamber, the cathode inlet manifold, and the cathode flow channel dead space. It is preferably between 80 ml and 180 ml, a range that reflects the true hysteresis without amplifying structural dead space errors, thus conforming to the effective retention volume characteristics of the series short gas path structure of a micro UAV fuel cell.

[0019] Normalized fuel cell current is the dimensionless ratio of the actual operating current of the fuel cell to the rated continuous output current.

[0020] The normalized rate of change of current is the dimensionless measure of how fast the normalized stack current changes with control time.

[0021] The characteristic altitude of the international standard atmosphere is a reference scale altitude that scales altitude to a normalized altitude. The engineering value is 44,330 meters.

[0022] Normalized altitude is a dimensionless representation of altitude relative to the characteristic height of the international standard atmosphere.

[0023] Normalized ambient temperature is the ratio of ambient temperatures calculated according to the altitude scale.

[0024] The adiabatic index is a power exponent used to derive normalized ambient pressure and normalized ambient density from normalized ambient temperature. A value of 5.256 is preferred, as this value meets the engineering application requirements of the international standard for atmospheric tropospheric pressure-temperature relationship and allows the normalized ambient density to naturally correspond to a density exponent on the order of 4.256.

[0025] Normalized environmental pressure is the dimensionless ratio of absolute environmental pressure to standard reference pressure.

[0026] Normalized ambient density is the dimensionless ratio of ambient air density to standard reference density.

[0027] The water vapor partial pressure ratio is the ratio of the water vapor partial pressure in the cathode mixture to the total pressure.

[0028] Effective oxygen availability is the percentage of available oxygen that can be used for electrochemical reactions after deducting the effect of water vapor dilution.

[0029] The normalized equivalent cathode gas volume is the dimensionless ratio of the equivalent cathode gas volume to the reference structural volume.

[0030] The cathode inlet pressure ratio is the ratio of the absolute pressure at the cathode inlet to the ambient pressure.

[0031] The normalized cathode gas temperature is the dimensionless ratio of the absolute temperature of the cathode gas to the standard reference temperature.

[0032] Normalized air molar flow rate is the dimensionless ratio of air molar flow rate to the rated reference air molar flow rate.

[0033] The numerical stabilization term is a protective constant used to prevent numerical divergence caused by denominators that are too small or close to 0. It is preferably between 0.001 and 0.01, a range that can suppress the amplification effect of small denominators without significantly changing the proportional relationship between the dominant physical quantities.

[0034] Normalized gas residence time is a dimensionless gas residence time scale obtained by combining equivalent volume, pressure, temperature and flow rate.

[0035] In practical implementation, to obtain the normalized stack current, the original value of the stack bus current needs to be collected first through a Hall current sensor, and then the rated continuous output current is used as the normalization reference for conversion. To avoid spike noise caused by sudden load changes, the original current should be subjected to first-order low-pass filtering before normalization. The normalization result is preferably limited to the range of 0 to 1.5, where 0 to 1 corresponds to the normal continuous output region and 1 to 1.5 corresponds to the short-term overload description region. When the sampled value is abnormally lost, the previous valid value is maintained for 1 control cycle. If there are abnormalities for 3 consecutive control cycles, the load reduction protection state is entered.

[0036] In practice, the calculation of the normalized current change rate requires dividing the difference between the normalized stack currents of two adjacent control cycles by the time step to obtain the discrete change rate, and then performing a first-order low-pass filter on the change rate. The normalized current change rate is preferably limited in a positive-negative symmetrical manner to avoid jumps in the feedforward interpolation table caused by single noise. The normalized current change rate can be set to 0 directly in the first 3 control cycles to prevent differential anomalies at the moment of power-on.

[0037] In practical implementation, for time alignment of synchronous acquisition, all acquired signals need to be driven by the same control clock issued by the main controller, preferably with a control frequency of 50 Hz to 100 Hz; if the refresh frequency of a certain sensor is lower than the main control frequency, the most recent moment is kept and timestamp is added for resampling; for sensors with fixed transmission delay, a delay compensation table should be set in the software so that the altitude, pressure, temperature and flow data in the same control cycle correspond to the same physical moment.

[0038] In practical implementation, for the setting of the characteristic altitude and adiabatic index of the international standard atmosphere, the characteristic altitude of the international standard atmosphere should be preferably taken as 44,330 meters, and the adiabatic index should be preferably taken as 5.256. The former is used to quickly convert the altitude to the normalized altitude, and the latter is used to establish the power function relationship between the normalized ambient temperature and the normalized ambient pressure and normalized ambient density. If the deviation between the sea area, season or airport ground reference and the standard atmosphere is large, it is permissible to make fine adjustments to the above constants during the ground calibration stage, but the adjustment amount should preferably not exceed 0.10.

[0039] In practice, for the correction of ambient temperature and ambient pressure, it is necessary to first estimate the normalized ambient temperature and normalized ambient pressure according to the standard atmospheric relationship, and then use the real-time collected ambient temperature and ambient pressure as online consistency verification and deviation correction quantities. When the deviation between the measured ambient pressure and the standard atmospheric estimated pressure exceeds 0.08, the measured ambient pressure should replace the estimated value in subsequent calculations. When the deviation between the measured ambient temperature and the temperature calculated from altitude continues to exceed 10 degrees Celsius, the normalized ambient temperature should be updated using a fusion correction method.

[0040] In practical implementation, regarding the conversion of humidification status information to water vapor partial pressure ratio, when the system directly provides dew point temperature, the saturated water vapor pressure can be found from the dew point temperature first, and then divided by the total pressure at the cathode inlet to obtain the water vapor partial pressure ratio; when the system provides relative humidity, the saturated water vapor pressure at that temperature can be found from the cathode gas temperature first, and then the relative humidity can be multiplied by the saturated water vapor pressure to obtain the actual water vapor partial pressure, and finally divided by the total pressure at the cathode inlet; if the relative humidity sensor fails, the humidifier setpoint and the average temperature of the last 3 seconds can be used for backup estimation.

[0041] In practical implementation, for the equivalent cathode gas volume, it is necessary to summarize the total actual cavity volume between the outlet of the micro centrifugal air compressor and the inlet of the cathode reaction zone of the fuel cell stack, and then multiply it by a reduction factor that takes into account non-uniform flow and local dead zones to obtain the equivalent cathode gas volume; the reduction factor is preferably between 1.05 and 1.30; in engineering, it can be determined through three-dimensional structural volume calculation, gas replacement test or step flow response identification; when normalizing, the reference volume of the rated design structure should be used as the denominator.

[0042] In specific implementation, for the benchmark settings of cathode feed pressure ratio, normalized cathode gas temperature and normalized air molar flow rate, the cathode feed pressure ratio should be preferably defined as the absolute pressure at the cathode inlet divided by the ambient pressure, the normalized cathode gas temperature should be preferably defined as the absolute temperature of the cathode gas divided by 288.15 Kelvin, and the normalized air molar flow rate should be preferably defined as the actual air molar flow rate divided by the reference air molar flow rate corresponding to the rated continuous power point.

[0043] In practical implementation, for numerical stability terms and boundary limits, the numerical stability term should preferably be uniformly set to 0.005, or it can be set in segments from 0.001 to 0.01 according to different denominator sensitivity. The effective oxygen availability should be set with a lower limit of not less than 0.05, the normalized air molar flow rate should be set with a lower limit of not less than 0, and the normalized gas residence time should be set with a reasonable upper limit to prevent the value from being infinitely amplified under the state of flow stoppage.

[0044] Preferably, the step of obtaining the feedforward operating condition reference command based on the normalized stack current and the normalized current change rate, calculating the Reynolds number characterization term, and using the Reynolds number characterization term to correct and obtain the true achievable gas supply envelope includes: The preset target oxygen excess coefficient With the normalized stack current Multiply by and divide by the effective oxygen availability. With the preset numerical stability term The sum of these values ​​yields the required airflow ratio. : Using the normalized stack current The normalized rate of change of current The normalized environmental pressure With the normalized environmental density Through a preset multidimensional interpolation function Interpolation extracts the feedforward operating condition reference command, which includes the feedforward airflow ratio. Feedforward reactor pressure ratio Feedforward air compressor speed ratio and feedforward throttle opening ratio : The normalized environmental density The speed ratio of the feedforward air compressor Multiply and divide by the preset normalized aerodynamic viscosity. With the preset numerical stability term The sum of these yields the normalized Reynolds number characterization. : According to the feedforward air compressor speed ratio and the feedforward reactor pressure ratio Obtain initial reachable traffic Subtracting the normalized Reynolds number from the value 1 represents The difference multiplied by the preset flow reduction factor The first correction term is derived by subtracting the first correction term from the value 1 and then multiplying the result by the initial achievable flow rate. The reachable airflow ratio in the actual reachable air supply envelope is obtained. : According to the feedforward air compressor speed ratio and the required airflow ratio Obtaining initial reachable pressure Subtracting the normalized Reynolds number from the value 1 represents The difference multiplied by the preset pressure reduction factor The second correction term is derived by subtracting the second correction term from the value 1 and then multiplying the result by the initial achievable pressure. The achievable infeed pressure ratio in the actual achievable gas supply envelope was obtained. : The target oxygen excess coefficient is the target oxygen supply margin reserved on the cathode side relative to the theoretical stoichiometric oxygen demand. It is preferably between 2.0 and 2.4, a range that can retain sufficient oxygen redundancy under conditions of low oxygen at high altitudes and transient high loads, while taking into account voltage stability and system efficiency.

[0045] The demand air flow ratio is the air supply ratio required to meet the current stack current and target oxygen excess coefficient requirements.

[0046] The feedforward reference command is a set of gas supply reference commands given in advance based on the current load and environmental conditions.

[0047] The feedforward airflow ratio is the target airflow ratio corresponding to the feedforward operating condition baseline command.

[0048] The feedforward pressure ratio is the target ratio of cathode feedforward pressure in the feedforward operating condition reference command.

[0049] The feedforward air compressor speed ratio is the relative ratio of the target speed of the micro centrifugal air compressor in the feedforward operating condition reference command.

[0050] The feedforward throttle opening ratio is the relative ratio of the electronic throttle target opening in the feedforward operating condition reference command.

[0051] Normalized aerodynamic viscosity is a dimensionless viscosity parameter used to characterize the current level of air viscous drag. It is preferably 1.0, normalized to 1 using the aerodynamic viscosity at the standard reference temperature to maintain a consistent Reynolds number characterization scale; in temperature correction implementation, a lookup value in the range of 0.90 to 1.10 is preferred.

[0052] Normalized Reynolds number is a dimensionless index that uses ambient density, air compressor speed and aerodynamic viscosity to characterize the current aerodynamic working capacity.

[0053] The initial achievable flow rate is the basic air flow capacity that can be achieved under the current feedforward compressor speed ratio and feedforward infeed pressure ratio, without low Reynolds number correction.

[0054] The flow reduction factor is a correction factor describing the intensity of airflow capacity attenuation under low Reynolds number conditions. It is preferably between 0.25 and 0.55, which aligns with the characteristic that the flow capacity degradation of high-altitude micro centrifugal air compressors is more significant than the pressure capacity degradation, requiring a stronger reduction margin.

[0055] The first correction term is the flow reduction obtained based on the normalized Reynolds number and the flow reduction factor.

[0056] The true reachable gas supply envelope is the set of gas supply boundaries that the system can actually reach under the current operating conditions after considering low Reynolds number decay.

[0057] The reachable airflow ratio is the boundary ratio of the achievable airflow within the true reachable air supply envelope.

[0058] The initial achievable pressure is the basic pressure capability that can be achieved at the current feedforward compressor speed ratio and demand air flow ratio before low Reynolds number correction.

[0059] The pressure reduction factor is a correction factor that describes the intensity of the decrease in reactor pressure capacity under low Reynolds number conditions. It is preferably between 0.20 and 0.45, which aligns with the characteristic that the decrease in pressure capacity at high altitudes is generally slightly less than the decrease in flow capacity, and is therefore slightly lower than the flow reduction factor.

[0060] The second correction term is the pressure reduction obtained based on the normalized Reynolds number and the pressure reduction factor.

[0061] The achievable feed pressure ratio is the ratio of the achievable feed pressure boundary within the true achievable gas supply envelope.

[0062] In practice, the target excess oxygen coefficient should be set to 2.0 under stable low-altitude cruise conditions, increased to 2.2 under conditions of continuous altitude increase and decreased environmental pressure, and further increased to 2.4 under transient conditions of significant load increase and positive normalized current change rate. When the system detects long-term stable low-load operation, it can also be appropriately reduced to 1.9 to 2.0 to reduce the parasitic work of the air compressor.

[0063] In specific implementation, the construction of a multidimensional interpolation table for the feedforward operating condition reference command requires that the interpolation table take at least four input axes: normalized stack current, normalized current change rate, normalized ambient pressure, and normalized ambient density, and four outputs: feedforward airflow ratio, feedforward in-reactor pressure ratio, feedforward air compressor speed ratio, and feedforward throttle opening ratio. The preferred interpolation method is four-dimensional linear interpolation, and the nearest boundary point clamping method is used when the boundary is exceeded.

[0064] In practical implementation, the normalized aerodynamic viscosity should be set using the aerodynamic viscosity corresponding to the standard reference temperature as the normalization benchmark of 1.0, and a maintenance positive table should be established based on the cathode gas temperature. When the cathode gas temperature is lower than the standard reference temperature, the normalized aerodynamic viscosity should be appropriately reduced, and when the cathode gas temperature is higher than the standard reference temperature, the normalized aerodynamic viscosity should be appropriately increased. To simplify airborne calculations, a constant of 1.0 can be directly used, but its error should be absorbed synchronously during calibration.

[0065] In practice, to establish the initial reachable flow mapping relationship, a steady-state scan of the micro centrifugal air compressor and throttle valve series system is performed on a ground test bench, using the feedforward air compressor speed ratio and feedforward reactor pressure ratio as two-dimensional input axes. The maximum stable air flow ratio when there is no surge, no overheating, and no overspeed is recorded, and the recorded results are stored as an initial reachable flow mapping table. For uncalibrated points, bilinear interpolation is used to obtain the initial reachable flow.

[0066] In practice, to establish the initial achievable pressure mapping relationship, it is necessary to use a ground test bench with the feedforward air compressor speed ratio and the demand air flow ratio as two-dimensional input axes to record the maximum cathode inlet pressure ratio that the system can achieve under stable flow conditions, and store the recorded results as an initial achievable pressure mapping table; this mapping should cover typical operating conditions such as UAV takeoff, climb, cruise and recovery.

[0067] In practice, the flow reduction factor and pressure reduction factor should be selected with a preferred value of 0.25 to 0.55 and a preferred value of 0.20 to 0.45. The calibration should be based on the high-altitude low-pressure chamber test so that the calculated achievable air flow ratio and achievable in-reactor pressure ratio are not higher than the measured capacity boundary.

[0068] In practice, the normalized Reynolds number (NRN) should be limited to the range of 0 to 1.20 for the first correction term, the second correction term, and the limit of the normalized Reynolds number characterization. When the NRN is greater than 1, it can be considered that no further attenuation is added. When the NRN is too low, at least a lower limit of 0 should be retained. The first and second correction terms should be limited to the range of 0 to 0.80 to prevent the capability boundary from being corrected to a negative value under extreme conditions.

[0069] In practical implementation, for the construction of the real achievable gas supply envelope, the achievable air flow ratio and the achievable reactor pressure ratio need to be used as the two main boundaries of the current control cycle. At the same time, the two, together with the demand air flow ratio and the feedforward reactor pressure ratio, constitute the feasible gas supply region. Subsequent current rise rate protection, risk closed-loop correction and online learning update should all use this region as the constraint boundary.

[0070] Preferably, the acquisition of pressure recovery degree and oxygen excess recovery degree, and the construction of pseudo-steady-state oxygen debt window intensity and oxygen debt accumulation based on the time-series recovery difference between the two, includes: Set the preset time step Divided by the normalized gas residence time With the preset numerical stability term The sum of these values ​​yields the integration step size. : The normalized air molar flow rate With the aforementioned effective oxygen availability Multiply, then divide by the normalized stack current. With the preset numerical stability term The sum of these values ​​yields the actual excess oxygen coefficient. : The feedforward airflow ratio With the aforementioned effective oxygen availability Multiply, then divide by the normalized stack current. With the preset numerical stability term The sum of these values ​​yields the feedforward target oxygen excess baseline. : The cathode feed pressure ratio The difference between subtracting the value 1 and dividing by the feedforward reactor pressure ratio Subtract the value 1 and the preset value stability term The sum of these values ​​yields the pressure recovery rate. : The actual oxygen excess coefficient The difference between subtracting the value 1 and dividing by the feedforward target oxygen excess benchmark Subtract the value 1 and the preset value stability term The sum of these values ​​yields the oxygen excess recovery rate. : Extract the pressure recovery degree Greater than the oxygen excess recovery rate Positive deviation at time The positive deviation and the integration step size are combined. The product of these factors, plus the pseudo-steady-state oxygen debt window intensity from the previous control cycle. With the preset natural exponential decay term The product of these factors yields the pseudo-steady-state oxygen debt window intensity for this control cycle. : The actual oxygen excess coefficient Divided by the aforementioned feedforward target oxygen excess benchmark With the preset numerical stability term The sum, obtained by subtracting the quotient from the stated value of 1, yields the hypoxia depth, and the accumulated oxygen debt from the previous control cycle is calculated. Add the hypoxia depth and the integration step size The product of these, minus the accumulated oxygen debt from the previous control cycle. With the integration step size The product of these factors yields the accumulated oxygen debt for this control cycle. : The time step is the discrete control time interval between two adjacent control cycles. It is preferably 0.01 seconds to 0.02 seconds. This range meets the effective control frequency requirement of 50 Hz to 100 Hz for UAV fuel cell air systems, which can both track rapid load changes and control noise amplification.

[0071] The integral step size is a dimensionless integral propagation scale obtained by converting the time step size relative to the normalized gas residence time.

[0072] The actual oxygen excess coefficient is the ratio of the effective oxygen currently supplied to the cathode to the actual oxygen demand of the fuel cell stack.

[0073] The feedforward target oxygen excess benchmark is the target oxygen supply margin benchmark value corresponding to the feedforward air flow ratio.

[0074] Pressure recovery is the degree to which the current cathode feed pressure recovers relative to the target feed pressure.

[0075] Oxygen excess recovery rate is the degree of recovery of the current actual oxygen excess coefficient relative to the feedforward target oxygen excess baseline.

[0076] Positive deviation is the positive difference between pressure recovery and oxygen excess recovery when the pressure recovery is greater than the oxygen excess recovery.

[0077] The natural exponential decay term is a decay factor used to describe the natural decay trend of the pseudo-steady-state oxygen debt window intensity over time.

[0078] The pseudo-steady-state oxygen debt window strength is a state quantity that characterizes the strength of the danger window where pressure rise precedes effective oxygen molecule replenishment. It corresponds to the historical value in the previous control period and the current value in the current control period.

[0079] The hypoxia depth is the degree of deficiency of the actual oxygen excess coefficient relative to the feedforward target oxygen excess baseline.

[0080] The cumulative oxygen debt is a risk status quantity obtained by dynamically accumulating the hypoxia depth over multiple consecutive control cycles. It corresponds to the historical value in the previous control cycle and the current value in the current control cycle.

[0081] In practice, regarding the setting of the time step, the controller should preferably run at a fixed period, with a preferred time step of 0.02 seconds, or 0.01 seconds depending on the processor's capabilities. When scheduling jitter occurs in the control task, the time step of the current period should be recalculated based on the actual timestamp, rather than using a fixed nominal value for an extended period.

[0082] In practice, for the boundary treatment of pressure recovery degree and oxygen excess recovery degree, if the value of the target quantity in the denominator of both is too small after subtracting 1, a numerical stability term should be added first and an effective denominator lower limit of not less than 0.05 should be set; the calculation results of both should preferably be limited to the range of 0 to 1.50.

[0083] In practice, for the extraction of positive deviation, a positive deviation should only be considered to exist when the pressure recovery rate is continuously greater than the oxygen excess recovery rate, and a dead zone threshold of not less than 0.02 and a continuous triggering criterion of not less than 2 control cycles should be set.

[0084] In practice, for updating the natural exponential decay term and the pseudo-steady-state oxygen debt window strength, the pseudo-steady-state oxygen debt window strength needs to be initialized to 0 upon power-on, and can be gradually cleared to zero when the system is shut down, fault is reset, or the recovery rate of continuous oxygen excess is higher than the pressure recovery rate. The natural exponential decay term is preferably calculated directly based on the current integration step size, so that the decay is faster when the normalized gas residence time is shorter and slower when the normalized gas residence time is longer.

[0085] In practice, regarding the limitation of hypoxia depth, when the actual oxygen excess coefficient is higher than the feedforward target oxygen excess benchmark, the hypoxia depth should be forcibly set to 0; when the actual oxygen excess coefficient is significantly lower than the target, the upper limit of hypoxia depth should preferably be set to 1.

[0086] In practice, the initialization and release of oxygen debt accumulation should be initiated by setting the oxygen debt accumulation to 0 upon power-on, preferably within an effective range of 0 to 3. When hypoxia depth persists, the accumulation should be gradually increased according to the integral step size. When hypoxia depth disappears, the accumulation should be gradually released according to the product of itself and the integral step size. When the system is shut down, the fault is cleared, or oxygen supply is stable for a sufficient period of time, the oxygen debt accumulation should be reset to 0 to prevent historical risks from having an unreasonable residual impact on subsequent flight phases.

[0087] Preferably, the step of obtaining the dynamically tuned reference control gain, introducing the pseudo-steady-state oxygen debt window intensity and the accumulated oxygen debt into the closed-loop regulation system, and combining the dynamically tuned reference control gain to generate the flow target correction and pressure target correction includes: Subtract the pressure recovery rate from the value 1. The pressure closed-loop error was obtained. : The pseudo-steady-state oxygen debt window intensity of this control cycle Multiply by the preset first risk weighting coefficient And the accumulated oxygen debt during this control cycle Multiply by the preset second risk weighting coefficient Add the sum of the products of the two to the stated value of 1, and then subtract the stated oxygen excess recovery rate. The closed-loop error of excess oxygen was obtained. : Utilizing the absolute value of the local sensitivity of the flow channel With the normalized gas residence time Automatic calculation is performed to generate the basic proportional gain corresponding to the target flow correction amount. Basic integral gain With fundamental differential gain And utilize the absolute value of the local sensitivity of the pressure channel. Similarly, the base gain corresponding to the pressure target correction is generated: Mapping function constructed based on fuzzy reasoning logic According to the oxygen excess closed-loop error and its discrete rate of change The intensity of the pseudo-steady-state oxygen debt window during this control period The cumulative oxygen debt during this control period Output dynamic scaling factor : The oxygen excess closed-loop error and its integral term With differential terms Multiply by the corresponding base proportional gain respectively The basic integral gain With the aforementioned fundamental differential gain Then sum them up, and multiply by the dynamic scaling factor mentioned above. The target flow correction amount is obtained. And using the same operational logic based on the pressure closed-loop error. and the corresponding dynamic scaling factor The pressure target correction amount is obtained. : Pressure closed-loop error is the remaining error amount between the current pressure recovery degree and the fully recovered state.

[0088] The first risk weighting coefficient is used to amplify the impact of the pseudo-steady-state oxygen debt window intensity on the oxygen excess closed-loop error. It is preferably between 0.8 and 1.2. The pseudo-steady-state oxygen debt window intensity reflects the immediate danger window and should have a high priority for flow closed-loop correction.

[0089] The second risk weighting coefficient is a weighting coefficient used to amplify the impact of accumulated oxygen debt on the closed-loop error of oxygen excess. It is preferably between 0.3 and 0.8. Accumulated oxygen debt reflects cumulative risk, and its impact should be significant but typically lower than the immediate danger window.

[0090] The oxygen excess closed-loop error is the oxygen supply risk closed-loop error obtained by combining the oxygen excess recovery degree, the pseudo-steady-state oxygen debt window intensity, and the cumulative amount of oxygen debt.

[0091] The absolute value of the local sensitivity of the flow channel is the absolute value of the local gain that reflects the strength of the response of the flow output to changes in the actuator under the current operating conditions.

[0092] The base proportional gain is an automatically calculated base gain based on the absolute value of the local sensitivity of the flow channel.

[0093] The basic integral gain is an automatically calculated flow integral gain based on the absolute value of the local sensitivity of the flow channel and the normalized gas residence time.

[0094] The basic differential gain is the flow differential adjustment basic gain, which is automatically calculated based on the absolute value of the local sensitivity of the flow channel and the normalized gas residence time.

[0095] The absolute value of the local sensitivity of the pressure channel is the absolute value of the local gain of the response of the pressure output to changes in the actuator under the current operating conditions.

[0096] The proportional term in the base gain corresponding to the pressure target correction is the pressure proportional base gain, which is automatically calculated based on the absolute value of the local sensitivity of the pressure channel.

[0097] The integral term in the base gain corresponding to the pressure target correction is the pressure integral base gain, which is automatically calculated based on the absolute value of the local sensitivity of the pressure channel and the normalized gas residence time.

[0098] The differential term in the base gain corresponding to the pressure target correction is the pressure differential base gain, which is automatically calculated based on the absolute value of the local sensitivity of the pressure channel and the normalized gas residence time.

[0099] The rate of change of oxygen excess closed-loop error is the speed at which the oxygen excess closed-loop error changes between adjacent control cycles.

[0100] The dynamic scaling factor is the flow closed-loop dynamic scaling factor obtained by fuzzy inference based on the oxygen excess closed-loop error, the rate of change of oxygen excess closed-loop error, the pseudo-steady-state oxygen debt window strength, and the cumulative amount of oxygen debt.

[0101] The integral term of the oxygen excess closed-loop error is an integral state quantity obtained by accumulating the oxygen excess closed-loop error over multiple control cycles.

[0102] The differential term of the oxygen excess closed-loop error is a differential state quantity that characterizes the changing trend of the oxygen excess closed-loop error.

[0103] The airflow target correction is the closed-loop compensation amount of airflow calculated based on the oxygen excess closed-loop error and its integral and differential terms.

[0104] The dynamic scaling factor corresponding to the pressure channel is the pressure closed-loop dynamic amplification factor obtained by adjusting the pressure closed-loop error, its changing trend, and the risk status.

[0105] The rate of change of pressure closed-loop error is the speed at which the pressure closed-loop error changes between adjacent control cycles.

[0106] The integral term of the pressure closed-loop error is an integral state quantity obtained by accumulating the pressure closed-loop errors over multiple control cycles.

[0107] The differential term of the pressure closed-loop error is a differential state quantity that characterizes the changing trend of the pressure closed-loop error.

[0108] The pressure target correction is the pressure closed-loop compensation calculated based on the pressure closed-loop error and its integral and differential terms.

[0109] In specific implementation, for the coordinated setting of the first risk weighting coefficient and the second risk weighting coefficient, the first risk weighting coefficient should preferably be around 1.0, and the second risk weighting coefficient should preferably be around 0.5. During calibration, the lowest voltage of the stack unit, the transient power response, and the number of oxygen-deficient alarms during the transition from normal flight to ramp-up should be used as comprehensive indicators, and the two should be jointly adjusted according to the principle that the immediate risk is higher than the cumulative risk.

[0110] In practice, for obtaining the absolute values ​​of the local sensitivity of the flow channel and the pressure channel, it is preferable to directly take the absolute values ​​of the diagonal elements of the local Jacobian matrix obtained by online identification as the main sensitivity. When the online identification is not yet stable, the ground test calibration value can be used as the initial value. In order to prevent the basic gain from being abnormally amplified due to the sensitivity being too small, a minimum sensitivity threshold of not less than 0.05 should be set.

[0111] In practical implementation, for the generation of the proportional, integral, and differential terms in the basic gain corresponding to the pressure target correction, the proportional term should preferably be determined by adding the absolute value of the local sensitivity of the pressure channel to the reciprocal of the numerical stability term; the integral term should preferably be determined by adding the normalized gas residence time to the absolute value of the local sensitivity of the pressure channel and the numerical stability term to the reciprocal of the product of the two; and the differential term should preferably be determined by dividing the normalized gas residence time by the sum of the absolute value of the local sensitivity of the pressure channel and the numerical stability term.

[0112] In practical implementation, for the fuzzy inference of the dynamic scaling factor and the dynamic scaling factor corresponding to the pressure channel, the oxygen excess closed-loop error, the rate of change of the oxygen excess closed-loop error, the pseudo-steady-state oxygen debt window strength, and the oxygen debt accumulation should be divided into five membership intervals: very small, small, medium, large, and very large. The output scaling factor range should be set to 0.6 to 2.0. When the oxygen excess closed-loop error is large and the pseudo-steady-state oxygen debt window strength is high, a larger dynamic scaling factor should be output. When the oxygen excess closed-loop error is small and the risk state is low, a dynamic scaling factor close to 1 or slightly less than 1 should be output. The corresponding dynamic scaling factor of the pressure channel adopts the same logic, but the input is changed to the pressure closed-loop error and its rate of change, and the sensitivity to the oxygen debt accumulation can be appropriately reduced to avoid overly aggressive pressure regulation.

[0113] In practical implementation, for the calculation of the error change rate of oxygen excess closed loop and the error change rate of pressure closed loop, it is preferable to divide the error difference between two adjacent control cycles by the time step to obtain the discrete change rate, and then pass it through a low-pass filter to obtain the final usable value; the low-pass filter time constant is preferably between 0.05 seconds and 0.10 seconds.

[0114] In practical implementation, to prevent integral saturation of the integral terms of oxygen excess closed-loop error and pressure closed-loop error, integration should only be allowed if the actuator has not reached the saturation boundary or if the current error direction helps the actuator to escape saturation. The integral state should be set with upper and lower limits, preferably within the range of -0.5 to 0.5. The integral terms should be cleared to zero when the system fails, stops, or the operating condition changes abruptly.

[0115] In practice, for the output of the flow target correction amount and the pressure target correction amount, both need to be set with single-cycle variation limit and total limit, and a smoothing process should be added before output; the single-cycle variation limit is preferably no more than 0.05, and the total limit is preferably no more than the range of -0.30 to +0.30.

[0116] Preferably, the step of online identification of the local Jacobian matrix of the series air circuit system, and using the local Jacobian matrix to calculate the flow target correction and the pressure target correction into the speed command of the micro centrifugal air compressor and the opening command of the electronic throttle valve, includes: Calculate the actuator change vector consisting of the changes in the speed command and the opening command. The L2 norm is used to divide the value by itself and the sum of the values ​​1 to obtain the adaptive update weights. : Using the local Jacobian matrix of the previous control cycle Multiply by the value 1 and the adaptive update weight The difference, plus the adaptive update weight The local Jacobian matrix for this control cycle is recursively derived by multiplying it with the system output difference quotient matrix based on the current measurement data. : (in (This is the system output change vector composed of the changes in actual flow rate and actual pressure); calculate the local Jacobian matrix for this control cycle. itself and its transpose trace of product and divide by the determinant of the product. With the preset numerical stability term The sum of these factors yields the damping factor. : By the damping factor The local Jacobian matrix of this control cycle Constructing a pseudo-inverse matrix ,in For identity matrix: The traffic target correction amount With the pressure target correction amount The target compensation vector is left-multiplied by the pseudo-inverse matrix. The decoupled actuator correction vector is obtained. : The decoupled actuator correction vector The corresponding components are respectively added to the feedforward air compressor speed ratio. Compared with the feedforward throttle opening ratio After passing through the preset physical boundary saturation limiting function After processing, the speed command is output and sent to the underlying layer. With the opening command : The local Jacobian matrix is ​​an identification matrix that characterizes the local linear mapping relationship between actuator changes and system flow and pressure output changes.

[0117] The actuator change vector is an actuator increment vector composed of the speed command change and the opening command change.

[0118] The adaptive update weights are weights that fuse the old and new Jacobian information, which are adaptively determined based on the size of the executor change vector.

[0119] The system output difference quotient matrix is ​​a local sensitivity approximation matrix constructed by the current measurement output change and the actuator change.

[0120] The system output change vector is an output increment vector composed of the changes in actual flow rate and actual pressure between adjacent control cycles.

[0121] The actual flow rate is the cathode gas supply flow rate that the system actually measures during the current control cycle.

[0122] The actual pressure is the cathode inlet pressure that the system actually measures during the current control cycle.

[0123] The damping factor is a regularization coefficient introduced to prevent matrix singularity or ill-conditionedness during the pseudo-inverse solution process.

[0124] The pseudo-inverse matrix is ​​a decoupled solution matrix constructed by the local Jacobian matrix and the damping factor.

[0125] The target compensation vector is a compensation target vector composed of the flow target correction amount and the pressure target correction amount.

[0126] The decoupled actuator correction vector is the actuator increment vector obtained by solving the target compensation vector through a pseudo-inverse matrix.

[0127] The speed command is the target speed command issued to the underlying driver of the micro centrifugal air compressor.

[0128] The opening command is the target opening command issued to the underlying actuator of the electronic throttle.

[0129] In practical implementation, the local Jacobian matrix should be preferably defined as a 2x2 matrix. The first row represents the sensitivity of the flow output to the actuator, the second row represents the sensitivity of the pressure output to the actuator, the first column represents the sensitivity to the air compressor speed, and the second column represents the sensitivity to the throttle opening. The sign direction should be uniformly defined as positive for increasing air compressor speed, increasing throttle opening, increasing actual flow rate, and increasing actual pressure.

[0130] In practical implementation, for the actuator change vector and the system output change vector, the actuator change vector needs to be composed of the speed command of the current control cycle minus the speed command of the previous control cycle, and the opening command of the current control cycle minus the opening command of the previous control cycle; the system output change vector needs to be composed of the actual flow rate of the current control cycle minus the actual flow rate of the previous control cycle, and the actual pressure of the current control cycle minus the actual pressure of the previous control cycle; both types of vectors must be calculated based on data aligned with the same timestamp, and a low-pass filter must be performed before construction.

[0131] In practical implementation, for the system output difference quotient matrix, it is preferable to update the system output difference quotient matrix only when the magnitude of the actuator change vector exceeds the minimum excitation threshold, which can be taken as 0.01; if the actuator change is too small, the local Jacobian matrix of the previous control cycle is maintained to avoid sensitivity divergence caused by dividing noise by small excitation.

[0132] In practice, for the initial value of the local Jacobian matrix and the power-on strategy, when power-on is required, the initial local Jacobian matrix obtained from ground calibration should be loaded, and then gradually corrected by online identification; in the first second after the system starts, a smaller adaptive update weight can be used to make the matrix smoothly transition from offline value to online identification value.

[0133] In practical implementation, for numerical protection of the damping factor and pseudo-inverse matrix, the damping factor should be set with upper and lower limits, preferably in the range of 0.01 to 10; when the local Jacobian matrix is ​​close to singular, the damping factor should be actively increased to improve the stability of the pseudo-inverse solution; after the pseudo-inverse solution, outlier checks should be performed on the correction vector. If it exceeds the physically acceptable range, the result should be rolled back to the previous control cycle.

[0134] In practical implementation, for speed commands and opening commands, the normalized speed commands should be converted into actual speed commands according to the allowable speed range of the micro centrifugal air compressor, and the normalized opening commands should be converted into actual opening commands according to the full stroke of the throttle valve. Before output, the output should be subject to physical boundary saturation limiting, and a single-cycle slope limit should be applied. It is preferred that the air compressor speed ratio changes no more than 0.03 per cycle, and the throttle valve opening ratio changes no more than 0.05 per cycle.

[0135] Preferably, the step of obtaining the current normalized stack voltage and the current load requested current, generating an allowable current rise rate based on the true achievable gas supply envelope, the current normalized stack voltage, and the pseudo-steady-state oxygen debt window intensity, and using the allowable current rise rate to limit the output current ramp-up and derive a protected power command includes: Subtract the speed command from the value 1 Determine the available margin of the air compressor : The achievable airflow ratio Divide by the required airflow ratio With the preset numerical stability term The sum of these yields the achievable gas supply capacity ratio. : The available margin of the air compressor The achievable gas supply capacity ratio and the current normalized stack voltage The numerator is obtained by continuous multiplication, and the value 1 and the pseudo-steady-state oxygen debt window intensity of this control period are then used. The cumulative oxygen debt during this control period Adding the three together gives the denominator; dividing the numerator by the denominator gives the allowable rate of rise of the current. : Compare the current requested current of the current load Compared with the protected output current command of the previous control cycle The difference between the two is taken and combined with the integration step size. Perform asymmetric limiting to generate the current protected output current command. : The current protected output current command Multiplied by the current normalized stack voltage The protected power command is calculated. : The current normalized stack voltage is the dimensionless ratio of the current stack operating voltage to the rated reference voltage.

[0136] The current load requested current is the target output current requested by the current flight mission or the onboard power management system.

[0137] The available capacity of an air compressor is the remaining capacity margin between the current speed command and the maximum permissible speed limit of the air compressor.

[0138] The achievable air supply capacity ratio is an indicator of the degree to which the achievable air flow ratio meets the demand air flow ratio.

[0139] The permissible current rise rate is the permissible load ramp-up rate calculated based on the available margin of the air compressor, the achievable air supply capacity ratio, the current normalized stack voltage, and the oxygen debt risk status.

[0140] The protected output current command is a safe output current command formed after gas supply capacity constraints and asymmetrical limiting. It corresponds to the historical value in the previous control cycle and the current value in the current control cycle.

[0141] The protected power command is the protected output power target obtained by combining the current protected output current command and the current normalized stack voltage.

[0142] In practical implementation, to obtain the current normalized stack voltage, it is necessary to first collect the original value of the total stack voltage, and then use the rated steady-state output voltage as the normalization reference for conversion; low-pass filtering should be performed before the voltage signal enters the controller, and the filtering time constant is preferably 0.05 seconds to 0.10 seconds; at the same time, a low-voltage protection threshold should be set, preferably when the normalized voltage is lower than 0.75, the allowable current rise rate should be further tightened to prevent individual unit voltage collapse.

[0143] In practice, for the acquisition of the current load request current, it is necessary that the current load request current can be issued by the flight controller, propulsion system controller or airborne energy management unit, and preferably the refresh cycle is consistent with the main control cycle; for external requests, validity verification and abnormal mutation screening should be implemented.

[0144] In practice, for the conversion of the allowable current rise rate and the current protected output current command, the allowable current rise rate should preferably be calculated first as the normalized rise rate per second relative to the rated continuous output current, then multiplied by the rated continuous output current to restore the actual current rise rate, and finally multiplied by the time step to obtain the actual current increase allowed in this cycle. Only after this conversion can it be compared with the current load request current and the protected output current command of the previous control cycle in the same dimension.

[0145] In practice, for the initialization and limiting of the protected output current command, when power-on startup is required, the protected output current command should preferably be initialized to 0 or the minimum safe current allowed by the system; its upper limit should not exceed the rated maximum output current, and its lower limit should not be less than 0; under normal operating conditions, the downward direction can quickly follow the requested current, and the upward direction must be constrained by the allowable current rise rate; when the system is stopped, fault reset or restarted, the protected output current command should be reinitialized.

[0146] In practical implementation, for the restoration of the protected power command, if the current normalized stack voltage adopts a normalized value relative to the rated steady-state output voltage, then the protected power command in engineering implementation should be equal to the current protected output current command multiplied by the current normalized stack voltage and then multiplied by the rated steady-state output voltage; if the system only uses the normalized power quantity, it should be restored to the actual power unit before being sent to the external power management unit.

[0147] Preferably, the calculation of the model prediction error for this control cycle, by updating the mapping parameters of the feedforward operating condition baseline command and the actual achievable gas supply envelope through a recursive algorithm including data quality weights, includes: The normalized stack current The normalized rate of change of current The normalized environmental pressure The normalized environmental density The effective oxygen availability The intensity of the pseudo-steady-state oxygen debt window during this control period The cumulative oxygen debt during this control period The speed command and the opening command The regression vector is formed by concatenating the value 1 with the given value. : Get the current model parameter matrix stored in the system The parameter covariance matrix of the previous control cycle ; using the current model parameter matrix With the regression vector Calculate the predicted output It will output the actual system measurements representing the actual flow rate and actual pressure. Subtract the predicted output The model prediction error for this control cycle is then obtained. : The model prediction error used in this control cycle The negative quadratic form is used to construct an exponential decay function to derive the data quality weights. : Based on the model prediction error of this control cycle The adaptive forgetting factor is derived from the L2 norm. : The data quality weight The parameter covariance matrix of the previous control cycle and the regression vector Multiply to obtain the intermediate vector, then divide by the aforementioned adaptive forgetting factor. With the regression vector The sum of quadratic forms yields the recursive update gain. : Using the recursive update gain The model prediction error of this control cycle Correct the current model parameter matrix The updated model parameter matrix is ​​obtained. and the updated model parameter matrix The corresponding components within are written back to the feedforward operating condition reference command and the mapping parameters of the actual reachable gas supply envelope: The regression vector is a model input feature vector that is formed by concatenating the current operating conditions, environment, oxygen debt risk status, and actuator commands in a fixed order.

[0148] The current model parameter matrix is ​​the parameter matrix used in the current control period to map the regression vector to the predicted output.

[0149] The parameter covariance matrix of the previous control cycle is the parameter uncertainty and update step weight matrix retained from the previous control cycle.

[0150] The predicted output is the system output prediction value calculated using the current model parameter matrix and regression vector.

[0151] The system's true measurement output is a true output vector composed of the flow rate and pressure actually measured by the system.

[0152] Model prediction error is the vector difference between the actual measured output and the predicted output of the system.

[0153] Data quality weights are weight coefficients generated based on the magnitude of model prediction errors to characterize the reliability of the current sample.

[0154] The adaptive forgetting factor is a coefficient that adaptively adjusts the retention rate of historical parameters based on the magnitude of the model's prediction error.

[0155] The intermediate vector is a recursive transition vector formed by multiplying the data quality weights, the parameter covariance matrix of the previous control period, and the regression vector.

[0156] The recursive update gain is a parameter-corrected gain vector calculated based on the intermediate vector, adaptive forgetting factor, and regression vector.

[0157] The updated model parameter matrix is ​​a new model parameter matrix obtained by recursively updating the gain and correcting the model prediction error.

[0158] The updated parameter covariance matrix is ​​a new covariance matrix obtained by recursively updating the gain, the adaptive forgetting factor, and the parameter covariance matrix of the previous control cycle.

[0159] The mapping parameters of the feedforward operating condition reference command are a set of calibration parameters used to determine the mapping results of the feedforward air flow ratio, feedforward reactor pressure ratio, feedforward air compressor speed ratio, and feedforward throttle opening ratio.

[0160] The mapping parameters of the true reachable gas supply envelope are a set of calibration parameters used to determine the initial reachable flow rate, initial reachable pressure, and boundary results after capacity reduction.

[0161] In specific implementation, for the current model parameter matrix, the regression vector needs to contain a total of 10 input components. Therefore, the current model parameter matrix is ​​preferably set to 10 rows and 2 columns, where the first column is used to predict the actual flow rate and the second column is used to predict the actual pressure. The first row of the matrix corresponds to the constant bias term, and the remaining rows correspond to the normalized stack current, normalized current change rate, normalized ambient pressure, normalized ambient density, effective oxygen availability, pseudo-steady-state oxygen debt window intensity, oxygen debt accumulation, speed command, and opening command, respectively. The initial power-on value can be obtained by regression from a large-sample ground test. If there is no sufficient sample, it can be set to 0 first and then gradually learned in the first few dozen stable periods.

[0162] In practice, for the parameter covariance matrix of the previous control cycle, the parameter covariance matrix should preferably be set as a 10-row, 10-column diagonal matrix. The initial diagonal elements can be uniformly set to 100 to 500 to indicate that there is a large uncertainty in the initial model parameters. During the recursive update process, its symmetry and positive definiteness should always be maintained. If numerical drift occurs, it should be corrected by diagonal reset or eigenvalue lower limit correction.

[0163] In practical implementation, for the actual system measurement, the actual system measurement output should preferably consist of actual flow rate and actual pressure in a fixed order, where the first component is actual flow rate and the second component is actual pressure; both should use the same units as the predicted output, preferably normalized units; at the same time, the two measurement values ​​must be aligned to the same control cycle timestamp.

[0164] In practice, for data quality weights and adaptive forgetting factors, the data quality weights need to be generated by the magnitude of the model prediction error. The larger the error, the smaller the data quality weight. Similarly, the adaptive forgetting factor decreases as the model prediction error increases. To prevent extreme abnormal samples from causing the update to freeze completely or become excessively divergent, the data quality weights are preferably limited to between 0.05 and 1, and the adaptive forgetting factor is preferably limited to between 0.10 and 0.99.

[0165] In practice, for intermediate vectors and recursive update gains, when the system detects sensor failure, actuator saturation lasting too long, communication interruption, or model prediction error exceeding a preset threshold, the recursive update should be paused, and the model parameter matrix and parameter covariance matrix of the previous control cycle should be directly maintained. During normal updates, the recursive update gain should be set with an upper limit on the magnitude to prevent a single abnormal sample from pulling the model parameter matrix out of the normal range.

[0166] In specific implementation, for the mapping parameters of the feedforward operating condition baseline command and the mapping parameters of the actual achievable air supply envelope, the components related to the feedforward air flow ratio, feedforward in-reactor pressure ratio, feedforward air compressor speed ratio, and feedforward throttle opening ratio in the updated model parameter matrix should be written back to the mapping parameters of the feedforward operating condition baseline command at a small ratio; the components related to the initial achievable flow rate and initial achievable pressure should be written back to the mapping parameters of the actual achievable air supply envelope at a small ratio; the write-back ratio is preferably between 0.01 and 0.05 to ensure a smooth learning process; when persisting parameters, they should only be written to non-volatile memory after the system has been running stably for a period of time to prevent transient abnormal samples from being permanently solidified into the model during the next startup.

[0167] It should be noted that the input and output parameters in the calculation formulas of this application are all dimensionless calculations performed through normalization processing. The formulas are all derived from software simulations based on a large amount of collected data, and the preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0168] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A control method for a high-altitude adaptive fuel cell system for unmanned aerial vehicles, applied to a series air circuit system comprising a micro centrifugal air compressor, an electronic throttle, a humidification buffer chamber, and a fuel cell stack cathode channel, characterized in that, include: Collect environmental information, gas path status information, and fuel cell stack electrical status information, convert them into dimensionless environmental physical property parameters, obtain normalized fuel cell stack current and normalized current change rate, and obtain normalized gas residence time. The feedforward operating condition reference command is obtained based on the normalized stack current and the normalized current change rate, the Reynolds number characterization term is calculated, and the true achievable gas supply envelope is obtained by correcting the Reynolds number characterization term. Obtain the pressure recovery degree and the oxygen excess recovery degree, and construct the pseudo steady-state oxygen debt window intensity and oxygen debt accumulation based on the time-series recovery difference between the two. The dynamically tuned baseline control gain is obtained, and the pseudo-steady-state oxygen debt window intensity and the accumulated oxygen debt are introduced into the closed-loop regulation system. The flow target correction and pressure target correction are generated by combining the dynamically tuned baseline control gain. The local Jacobian matrix of the series air circuit system is identified online, and the flow target correction and the pressure target correction are calculated into the speed command of the micro centrifugal air compressor and the opening command of the electronic throttle using the local Jacobian matrix. The current normalized stack voltage and the current load request current are obtained. Based on the true reachable gas supply envelope, the current normalized stack voltage and the pseudo steady-state oxygen debt window intensity, the allowable current rise rate is generated. The allowable current rise rate is used to limit the ramp-up of the output current and to obtain the protected power command. The model prediction error for this control cycle is calculated, and the mapping parameters of the feedforward operating condition baseline command and the actual reachable gas supply envelope are updated using a recursive algorithm that includes data quality weights.

2. The control method for a high-altitude adaptive fuel cell system for unmanned aerial vehicles according to claim 1, characterized in that, The collected environmental information, gas path status information, and fuel cell stack electrical status information are converted into dimensionless environmental physical property parameters. Normalized fuel cell stack current and normalized current change rate are obtained, along with normalized gas residence time, including: Simultaneously acquire altitude, ambient temperature, ambient pressure, cathode inlet pressure, cathode gas temperature, humidification status information, air molar flow rate, and equivalent cathode gas volume; The normalized altitude is obtained by dividing the altitude by the characteristic height of the international standard atmosphere, and the normalized ambient temperature is obtained by subtracting the normalized altitude from the value 1. The normalized ambient temperature is used as the normalized ambient pressure by raising the preset adiabatic index power. The normalized ambient density is the power of the difference between the preset adiabatic index of the normalized ambient temperature and the value of 1. The water vapor partial pressure ratio is obtained based on the humidification status information, and the effective oxygen availability is obtained by subtracting the water vapor partial pressure ratio from the value 1. The normalized equivalent cathode gas volume calculated from the equivalent cathode gas volume, the cathode feed pressure ratio calculated from the cathode inlet pressure, and the normalized ambient pressure are continuously multiplied together to obtain the numerator. The product of the normalized cathode gas temperature calculated from the cathode gas temperature and the normalized air molar flow rate calculated from the air molar flow rate is added to a preset numerical stability term to obtain the denominator. The numerator is divided by the denominator to obtain the normalized gas residence time.

3. The control method for a high-altitude adaptive fuel cell system for unmanned aerial vehicles according to claim 2, characterized in that, The process of obtaining the feedforward operating condition reference command based on the normalized stack current and the normalized current change rate, calculating the Reynolds number characterization term, and using the Reynolds number characterization term to correct and obtain the true achievable gas supply envelope includes: The required airflow ratio is obtained by multiplying the preset target oxygen excess coefficient by the normalized stack current and dividing by the sum of the effective oxygen availability and the preset numerical stability term. Using the normalized stack current, the normalized current change rate, the normalized ambient pressure, and the normalized ambient density, the feedforward operating condition reference command is extracted by interpolation. The feedforward operating condition reference command includes the feedforward air flow ratio, the feedforward in-stack pressure ratio, the feedforward air compressor speed ratio, and the feedforward throttle opening ratio. Multiply the normalized ambient density by the feedforward air compressor speed ratio and divide by the sum of the preset normalized aerodynamic viscosity and the preset numerical stability term to obtain the normalized Reynolds number characterization. The initial achievable flow rate is obtained based on the feedforward air compressor speed ratio and the feedforward infeed pressure ratio. The difference represented by the normalized Reynolds number is subtracted from the value 1 and multiplied by a preset flow rate reduction factor to obtain the first correction term. The first correction term is subtracted from the value 1 and multiplied by the initial achievable flow rate to obtain the achievable air flow rate ratio in the true achievable air supply envelope. The initial achievable pressure is obtained based on the feedforward air compressor speed ratio and the required air flow ratio. The difference represented by the normalized Reynolds number is subtracted from the value 1 and multiplied by a preset pressure reduction factor to obtain the second correction term. The value 1 is then subtracted from the second correction term and multiplied by the initial achievable pressure to obtain the achievable infeed pressure ratio in the actual achievable air supply envelope.

4. The control method for a high-altitude adaptive fuel cell system for unmanned aerial vehicles according to claim 3, characterized in that, The acquisition of pressure recovery degree and oxygen excess recovery degree, and the construction of pseudo-steady-state oxygen debt window intensity and oxygen debt accumulation based on the time-series recovery difference between the two, include: The integration step size is obtained by dividing the preset time step by the sum of the normalized gas residence time and the preset numerical stability term. Multiply the normalized air molar flow rate by the effective oxygen availability, and divide by the sum of the normalized fuel cell current and the preset numerical stability term to obtain the actual oxygen excess coefficient. Multiply the feedforward airflow ratio by the effective oxygen availability, and divide by the sum of the normalized stack current and the preset numerical stability term to obtain the feedforward target oxygen excess benchmark. The pressure recovery degree is obtained by dividing the difference between the cathode feed pressure ratio and the value 1 by the sum of the feedforward feed pressure ratio minus the value 1 and the preset numerical stability term. The difference between the actual oxygen excess coefficient and the value 1 is divided by the sum of the feedforward target oxygen excess benchmark minus the value 1 and the preset numerical stability term to obtain the oxygen excess recovery degree. Extract the positive deviation when the pressure recovery degree is greater than the oxygen excess recovery degree, and add the product of the positive deviation and the integral step size to the product of the pseudo steady-state oxygen debt window intensity of the previous control cycle and the preset natural exponential decay term to obtain the pseudo steady-state oxygen debt window intensity of the current control cycle. Divide the actual oxygen excess coefficient by the sum of the feedforward target oxygen excess benchmark and the preset numerical stability term, subtract the quotient from the value 1 to obtain the hypoxia depth, add the accumulated oxygen debt of the previous control cycle to the product of the hypoxia depth and the integral step size, and then subtract the product of the accumulated oxygen debt of the previous control cycle and the integral step size to obtain the accumulated oxygen debt of the current control cycle.

5. The control method for a high-altitude adaptive fuel cell system for unmanned aerial vehicles according to claim 4, characterized in that, The process of obtaining the dynamically tuned baseline control gain, introducing the pseudo-steady-state oxygen debt window intensity and the accumulated oxygen debt into the closed-loop regulation system, and combining the dynamically tuned baseline control gain to generate flow target correction and pressure target correction includes: Subtracting the pressure recovery degree from the value 1 yields the pressure closed-loop error. Multiply the pseudo steady-state oxygen debt window intensity of this control period by a preset first risk weighting coefficient, and multiply the cumulative amount of oxygen debt of this control period by a preset second risk weighting coefficient. Add the sum of the products of the two to the value 1 and then subtract the oxygen excess recovery degree to obtain the oxygen excess closed-loop error. The basic proportional gain, basic integral gain, and basic differential gain corresponding to the flow target correction amount are automatically calculated by using the absolute value of the local sensitivity of the flow channel and the normalized gas residence time. Similarly, the basic gain corresponding to the pressure target correction amount is generated by using the absolute value of the local sensitivity of the pressure channel. Based on fuzzy reasoning logic, a dynamic scaling factor is output according to the oxygen excess closed-loop error and its rate of change, the pseudo steady-state oxygen debt window strength and the cumulative amount of oxygen debt in the current control cycle. The oxygen excess closed-loop error and its integral and differential terms are multiplied by the corresponding basic proportional gain, basic integral gain and basic differential gain, respectively, and then summed and multiplied by the dynamic scaling factor to obtain the flow target correction amount; and the same calculation logic is used to obtain the pressure target correction amount based on the pressure closed-loop error.

6. The control method for a high-altitude adaptive fuel cell system for unmanned aerial vehicles according to claim 5, characterized in that, The online identification of the local Jacobian matrix of the series air circuit system, and the calculation of the flow target correction and the pressure target correction into the speed command of the micro centrifugal air compressor and the opening command of the electronic throttle using the local Jacobian matrix, includes: Calculate the L2 norm of the actuator change vector composed of the changes in the speed command and the opening command, and divide it by the sum of itself and the value 1 to obtain the adaptive update weight; The local Jacobian matrix of the current control cycle is obtained by multiplying the local Jacobian matrix of the previous control cycle by the difference between the value 1 and the adaptive update weight, and adding the product of the adaptive update weight and the system output difference quotient matrix based on the current measurement data. Calculate the trace of the product of the local Jacobian matrix itself and its transpose matrix for this control cycle, and divide it by the sum of the determinant of the product and the preset numerical stability term to obtain the damping factor. A pseudo-inverse matrix is ​​constructed from the damping factor and the local Jacobian matrix of the current control cycle; Multiply the target compensation vector composed of the flow target correction and the pressure target correction by the pseudo-inverse matrix to obtain the decoupled actuator correction vector; The components of the decoupled actuator correction vector are superimposed on the feedforward air compressor speed ratio and the feedforward throttle opening ratio, respectively. After saturation limiting, the speed command and the opening command are output and sent to the bottom layer.

7. The control method for a high-altitude adaptive fuel cell system for unmanned aerial vehicles according to claim 6, characterized in that, The process of acquiring the current normalized stack voltage and the current load requested current, generating an allowable current rise rate based on the true achievable gas supply envelope, the current normalized stack voltage, and the pseudo-steady-state oxygen debt window intensity, and using the allowable current rise rate to limit the output current ramp-up and derive a protected power command includes: Subtracting the speed command from the value 1 yields the available margin of the air compressor. The achievable air flow ratio is obtained by dividing the required air flow ratio by the sum of the preset numerical stability terms. The numerator is obtained by continuously multiplying the available margin of the air compressor, the achievable air supply capacity ratio, and the current normalized stack voltage. The denominator is obtained by adding the value 1, the pseudo-steady-state oxygen debt window intensity of the current control cycle, and the cumulative amount of oxygen debt of the current control cycle. The allowable current rise rate is obtained by dividing the numerator by the denominator. The current load request current is compared with the protected output current command of the previous control cycle. The difference between the two is combined with the integral step size to perform asymmetric limiting and generate the current protected output current command. The protected power command is calculated by multiplying the current protected output current command by the current normalized stack voltage.

8. The control method for a high-altitude adaptive fuel cell system for unmanned aerial vehicles according to claim 7, characterized in that, The calculation of the model prediction error for this control cycle, and the updating of the mapping parameters of the feedforward operating condition baseline command and the actual achievable gas supply envelope using a recursive algorithm that includes data quality weights, include: The normalized stack current, the normalized current change rate, the normalized ambient pressure, the normalized ambient density, the effective oxygen availability, the pseudo-steady-state oxygen debt window intensity of the current control cycle, the cumulative oxygen debt of the current control cycle, the speed command, and the opening command are concatenated with the value 1 to form a regression vector. Obtain the current model parameter matrix and the parameter covariance matrix of the previous control cycle stored in the system; The predicted output is calculated using the current model parameter matrix and the regression vector, and the predicted output is subtracted from the actual system measurement output representing the actual flow and actual pressure to obtain the model prediction error for this control cycle. The data quality weights are derived by constructing an exponential decay function using the negative quadratic form of the model prediction error in this control cycle. The adaptive forgetting factor is derived from the L2 norm of the model prediction error in this control cycle. The intermediate vector is obtained by multiplying the data quality weight, the parameter covariance matrix of the previous control cycle, and the regression vector, and then divided by the sum of the adaptive forgetting factor and the quadratic form of the regression vector to obtain the recursive update gain. The current model parameter matrix is ​​corrected by using the recursive update gain and the model prediction error of the current control cycle to obtain the updated model parameter matrix. The corresponding components in the updated model parameter matrix are then written back to the feedforward operating condition reference command and the mapping parameters of the actual reachable gas supply envelope.

Images