An air-cooled UAV hydrogen fuel cell control system

Abstract

The present application relates to the technical field of UAV power systems, and in particular to an air-cooled UAV hydrogen fuel cell control system. It includes: a spray intercooler that sprays stack wastewater as a cooling medium; a wastewater circulation system that transports stack wastewater to the spray nozzle of the spray intercooler; and a reinforcement learning control system that executes a control program to control the spray intercooler and the wastewater circulation system. The closed-loop wastewater spray cooling system significantly reduces the damage to the membrane electrode caused by the high-temperature intake air of the air compressor. At the same time, the reinforcement learning algorithm is used to achieve four-dimensional coordinated control of temperature, humidity, pressure, and flow, so that the stack always operates stably in the optimal operating range. This not only compresses the actuator response delay to the millisecond level to ensure the real-time performance of the UAV power output, but also significantly extends the service life of the stack and improves the stability of the output power while reducing the weight of the system.

Description

Technical Field

[0001] The present application relates to the technical field of UAV power systems, and in particular to an air-cooled UAV hydrogen fuel cell control system. Background Art

[0002] Hydrogen fuel cells offer the advantages of high energy density and zero emissions as drone propulsion systems, but they face significant challenges in closed, air-cooled architectures. The heat generated by the air compressor during compression causes the intake air temperature to rise sharply above 70°C. Traditional air cooling methods, due to insufficient heat exchange efficiency, cannot achieve rapid cooling. Direct entry of high-temperature gases into the fuel cell stack can lead to dehydration and failure of the membrane electrode. While water cooling systems can address heat dissipation, the added weight of piping, pumps, and coolant far exceeds the drone's load capacity, forcing them to be abandoned due to the demand for lightweighting. Furthermore, existing control strategies suffer from fundamental flaws. Traditional PID controllers are unable to collaboratively optimize the four-dimensional variables of temperature, humidity, pressure, and flow. Strong multi-physics coupling leads to frequent parameter oscillations, causing the fuel cell stack to deviate from optimal operating conditions for extended periods, resulting in power output degradation and shortened lifespan. Furthermore, actuator adjustment based on PID or PWM has a delay exceeding one second. The lag in fan speed and valve opening directly impacts the real-time performance of the drone's power output, significantly reducing flight control smoothness. These systemic flaws collectively limit the application prospects of hydrogen fuel cells in drones operating in extreme environments. Summary of the Invention

[0003] In order to solve the technical problems existing in the prior art, the present application provides an air-cooled UAV hydrogen fuel cell control system, including a spray-type intercooler for spraying fuel cell wastewater as a cooling medium to cool the hot air discharged from the air compressor;

[0004] A wastewater circulation system, used for conveying the stack wastewater to the spray nozzle of the spray-type intercooler;

[0005] A reinforcement learning control system is used to execute a control program to control the spray intercooler and the wastewater circulation system, wherein the control program includes:

[0006] A state perception module is used to obtain a 27-dimensional state vector through a sensor group. The state vector includes the current and historical sampling sequence of the stack temperature and its rate of change, intake air humidity, hydrogen pressure, and wastewater tank level.

[0007] The decision execution module is used to extract 128-dimensional hidden state features from the state vector through the LSTM network to predict the temperature rise trend in the next 3 seconds, and then input the hidden state features into the SAC strategy network to generate a four-dimensional continuous action vector through the fully connected layer:

[0008] ,

[0009] The four-dimensional continuous motion vector is mapped to the actuators of the spray intercooler and the wastewater circulation system, wherein is the air compressor speed adjustment value, is the change in the opening of the hydrogen fuel valve, is the power change of the cooling fan of the spray type intercooler, It is the spray flow rate that drives the wastewater circulation system.

[0010] Specifically, the spray mechanism of the spray-type intercooler consists of copper-aluminum composite fins and ultrasonic atomizing nozzles. The copper-aluminum composite fins have a thickness of 0.3 mm and a spacing of 1.5 mm. The droplet size generated by the ultrasonic atomizing nozzle is ≤50 μm.

[0011] Specifically, the wastewater circulation system consists of a filter and a pressure pump. The filtration accuracy of the filter is 5 μm, and the pressure pump provides a pressure of 0.15 MPa.

[0012] Specifically, the four-dimensional continuous motion vector Affected by the interval Constraints and Constraints, where is the maximum spray flow rate, and H is the real-time liquid level height.

[0013] Specifically, the LSTM network processes the state vectors of 10 consecutive sampling points in a sliding window manner, and the prediction of the temperature rise trend in the next 3 seconds is used to compensate for the thermal inertia delay of the fuel cell stack.

[0014] Specifically, the SAC policy network initializes weights through transfer learning, the pre-training model is generated based on 10,000 hours of simulation data, and the online fine-tuning cycle updates the network weights every 50 hours of flight.

[0015] Specifically, the wastewater circulation system is provided with a liquid level-spray linkage constraint. When the liquid level of the wastewater box is lower than the threshold, the wastewater box is forced to ; When the standard deviation of the battery stack cell voltage is greater than 50mV, switch to the preset PID rule base control.

[0016] Specifically, the sensor group includes a MEMS temperature and humidity sensor deployed at the outlet of the intercooler, a hydrogen pressure sensor deployed in the hydrogen pipeline, a voltage acquisition board deployed on the bipolar plate leads of the fuel cell stack, a liquid level meter deployed in the wastewater box, and a laser tachometer deployed on the outside of the air compressor turbine.

[0017] Specifically, the dynamic range of the four-dimensional continuous motion vector satisfies:

[0018] .

[0019] Specifically, the surface of the copper-aluminum composite fin of the spray-type intercooler is provided with a hydrophobic coating, and the contact angle of the hydrophobic coating with the wastewater droplets is greater than 120°.

[0020] This application has the following technical effects:

[0021] The closed-loop wastewater spray cooling system significantly reduces the damage to the membrane electrode caused by the high-temperature intake air of the air compressor. At the same time, the reinforcement learning algorithm is used to achieve four-dimensional coordinated control of temperature, humidity, pressure and flow, so that the fuel cell stack always operates stably in the optimal operating range. Not only is the actuator response delay compressed to the millisecond level to ensure the real-time power output of the drone, but it also greatly extends the service life of the fuel cell stack and improves the stability of the output power while reducing the weight of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] By reading the detailed description below with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present application will become readily understood. In the accompanying drawings, several embodiments of the present application are shown in an exemplary and non-limiting manner, and the same or corresponding numbers represent the same or corresponding parts.

[0023] Figure 1 This is a structural diagram of an air-cooled UAV hydrogen fuel cell control system in an embodiment of the present application. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application.

[0025] Hydrogen fuel cells offer significant energy density advantages as a power source for drones. However, their closed, air-cooled architecture presents several obstacles for practical application, directly threatening system reliability and flight performance. The most prominent issue stems from the significant heat generated during the air compression process. As the air compressor increases intake pressure, the air temperature inevitably rises dramatically, exceeding 70°C. Traditional air cooling methods, limited by inherent heat exchange efficiency, are unable to quickly and effectively cool this high-temperature air. If the hot air enters the fuel cell core without adequate cooling, it can cause irreversible damage to the membrane electrode due to dehydration. A water-cooling system is the ideal solution for this heat dissipation problem, but the added weight is prohibitive for drone platforms, where every gram is crucial. The weight burden of the necessary cooling lines, circulation pumps, and coolant itself far exceeds the limits of lightweight design, forcing this solution to be completely eliminated for drones. Furthermore, existing control strategies struggle to cope with the complexity of fuel cell operation, exposing fundamental flaws. Traditional proportional-integral-derivative (PID) controllers struggle to simultaneously coordinate four interrelated and mutually constrained key variables: temperature, humidity, gas pressure, and flow. These physical quantities are strongly coupled, and within the linear control framework of PID, interference between parameters often causes difficult-to-manage system oscillations. The direct consequence of this oscillation is that the fuel cell stack cannot operate stably near its designed optimal operating point for extended periods. Forced to operate under non-ideal conditions, the stack suffers from significant fluctuations in output power and a significantly shortened useful life. Compounding the problem is actuator response hysteresis. Whether based on PID or pulse-width modulation (PWM) principles, controllers experience significant delays exceeding one second when adjusting fan speed or valve opening. This delay is critical in rapidly changing flight environments, preventing the drone's power output from precisely matching flight control commands. This lag in power response can even lead to loss of control during aggressive maneuvers such as emergency climbs or evasive dives. This intertwined set of flaws constitutes a key bottleneck hindering the reliable application of hydrogen fuel cell technology in drones operating in extreme environments.

[0026] In response to the above contradictions, the inventors of this application reconstructed the thermal management path through the deep coupling of hardware structure and intelligent algorithm, and proposed an air-cooled UAV hydrogen fuel cell control system, such as Figure 1 Shown, including:

[0027] Spray-type intercooler, used to spray stack wastewater as a cooling medium to cool the hot air discharged from the air compressor;

[0028] Wastewater circulation system, used to transport stack wastewater to the spray nozzles of the spray intercooler;

[0029] A reinforcement learning control system is used to execute a control program to control the spray intercooler and wastewater circulation system. The control program includes:

[0030] The state perception module is used to obtain a 27-dimensional state vector through the sensor group. The state vector includes the current and historical sampling sequence of the stack temperature and its rate of change, intake air humidity, hydrogen pressure, and wastewater box level;

[0031] The decision execution module is used to extract 128-dimensional hidden state features from the state vector through the LSTM network to predict the temperature rise trend in the next 3 seconds. The hidden state features are then input into the SAC policy network and generated into a four-dimensional continuous action vector through the fully connected layer:

[0032] ,

[0033] The four-dimensional continuous motion vector is mapped to the actuators of the spray intercooler and the wastewater circulation system, where is the air compressor speed adjustment value, is the change in the opening of the hydrogen fuel valve, is the change in cooling fan power of the spray intercooler, To drive the spray flow of the wastewater circulation system.

[0034] The core of the system lies in the construction of a closed-loop system consisting of a spray intercooler, a wastewater circulation system, and a reinforcement learning control system. The process begins with wastewater generated by the electrochemical reactions within the fuel cell stack. Rather than simply being discarded, this wastewater is channeled through a sophisticated filtration unit. After filtration and purification, it is transported to the ultrasonic atomizer nozzle, a key component of the spray intercooler. Here, the wastewater is atomized into extremely fine droplets and sprayed onto the surface of specially designed heat exchange fins. Simultaneously, the reinforcement learning control system continuously receives real-time data streams from a network of sensors deployed at key nodes in the system. The system processes and analyzes this massive amount of data at high speed, ultimately outputting a set of precise four-dimensional control instructions. These instructions coordinately adjust the operating speed of the air compressor, control the opening of the hydrogen supply valve, adjust the power output of the cooling fan, and precisely manage the spray liquid flow rate. This multi-variable coordinated control system has a single ultimate goal: to ensure that the four key parameters within the fuel cell stack—temperature, humidity, gas pressure, and reactant flow—are consistently stable within preset optimal ranges, laying the foundation for efficient and long-life fuel cell operation.

[0035] The spray-type intercooler is the physical basis for achieving efficient thermal management in this embodiment, and its structural design contains many key innovations. The heat exchange core is made of copper-aluminum composite fins. This material selection takes into account excellent thermal conductivity and moderate weight. The thickness of the fins is strictly limited to 0.3mm and the spacing is 1.5mm. The surface is covered with a hydrophobic coating with a contact angle greater than 120°. This structural design enables the heat exchange efficiency per unit area to exceed 220W / m²·K. In order to further improve performance and ensure reliability, a layer of nano-level hydrophobic functional coating is specially applied to the surface of the fins. The core characteristic of this coating is that its contact angle with water is greater than one hundred and twenty degrees, which ensures that the tiny droplets sprayed on its surface can spread quickly and evaporate efficiently, rather than gathering into a water film that hinders heat transfer.

[0036] The core of the spray mechanism is the ultrasonic atomizing nozzle. As 70°C hot air from the air compressor flows through the fin array, the wastewater circulation system purifies the stack reaction wastewater through a 5μm precision filter. The wastewater is then pumped to the ultrasonic atomizing nozzle at a pressure of 0.15MPa by a booster pump. The atomized droplets are evenly sprayed onto the fin surface, forming a spherical shape ≤50μm. The droplets evaporate rapidly under the hydrophobic coating, absorbing a latent heat of evaporation of 2450kJ / kg. This improves heat exchange efficiency by 82% compared to traditional air-cooling solutions. The inlet air temperature is also steadily reduced to 35-40°C, completely eliminating the risk of membrane electrode dehydration.

[0037] In this embodiment, the wastewater circulation system realizes the dynamic monitoring of the liquid level simultaneously. When the liquid level of the wastewater box is lower than the safety threshold, the system automatically triggers the flow restriction mechanism. Specifically, the four-dimensional continuous motion vector Affected by the interval Constraints and Constraints, where is the maximum spray flow rate, and H is the real-time liquid level. This means that the maximum allowable spray flow rate is not fixed but adjusts in real time with the liquid level in the wastewater storage tank. The core purpose of this dynamic constraint mechanism is to prevent the systemic risk of rapid depletion of the liquid level during intense operation or under unusual operating conditions, leading to the failure of the entire cooling function. This physical constraint is cleverly encoded and integrated into the action space boundary conditions of the reinforcement learning controller.

[0038] A reinforcement learning control system forms the core of the intelligent decision-making process for the entire solution. Its architectural design fully considers deep compatibility with the hardware system and real-time requirements. In this embodiment, the system's perception layer is responsible for collecting and constructing a state vector with a dimension of up to 27. This vector integrates temporal evolution information and spatial distribution information. Regarding temporal parameters, the focus is on the rate of change of the stack temperature (dT / dt), with a sampling interval of precisely 0.1 seconds. The system also records and analyzes the continuous sequence of temperature and humidity changes over the past 10 seconds. Regarding spatial parameters, the system simultaneously monitors the temperature difference between the center and edge of the stack, as well as the standard deviation of the output voltage of the 24 individual cells that make up the stack, to assess the uniformity of the stack's internal operating conditions. The core of the decision layer is a SAC-LSTM hybrid model. In this model, a long short-term memory (LSTM) module is used to process sensor data with time series characteristics and extract deep temporal features to compensate for the thermal inertia delay effects inherent in the stack. The input dimension is 27, corresponding to the dimension of the state vector, and the hidden layer dimension is set to 128. In this model, the SAC algorithm's policy network receives high-level features extracted by LSTM. The model contains two fully connected layers. The first layer has 64 neurons, and the second layer outputs the crucial four-dimensional action vector. The four dimensions of the action vector correspond to the air compressor speed adjustment amount. , Change in hydrogen valve opening , cooling fan power adjustment And spray flow adjustment The final output layer of the network uses a hyperbolic tangent activation function to normalize the output value to the range of -1 to 1, and then scales and maps it according to the actual range of each actuator. The execution layer is responsible for converting intelligent decisions into precise actions of physical devices, and the dynamic adjustment range of its variables is set according to the hardware capabilities. Specifically, the dynamic range of the four-dimensional continuous action vector satisfies:

[0039] .

[0040] In this embodiment, the wastewater circulation system is provided with a liquid level-spray linkage constraint. When the liquid level of the wastewater box is lower than the threshold, the wastewater box is forced to ; When the standard deviation of the battery stack cell voltage is greater than 50mV, switch to the preset PID rule base control.

[0041] Specifically, in this embodiment, the sensor group includes a MEMS temperature and humidity sensor deployed at the outlet of the intercooler, a hydrogen pressure sensor deployed in the hydrogen pipeline, a voltage acquisition board deployed on the bipolar plate leads of the fuel cell stack, a liquid level meter deployed in the wastewater box, and a laser tachometer deployed on the outside of the air compressor turbine.

[0042] In another embodiment of the present application, a highly absorbent evaporation layer can be employed instead of a water mist nozzle. This layer covers the intercooler surface, with one side immersed in a wastewater reservoir. Wastewater discharged into the reservoir is then sucked through the porous fibers of the evaporation layer to the surface via a siphon effect. The porous fibers are made of polyester, nylon, cotton, or linen, with individual fibers ranging in diameter from 5 to 20 microns, creating numerous tiny voids. When stacked, the fibers can achieve a porosity of up to 70%, creating a sponge-like structure that facilitates water retention. The fibers can be randomly twisted or woven, further increasing the surface area. The hydrophilic nature of the fibers forms hydrogen bonds with water, which is drawn into the pores of the evaporation layer through capillary action. The porous structure provides numerous adsorption sites, ensuring uniform distribution of water throughout the evaporation layer. Furthermore, the fiber pores increase the surface area of ​​contact between water and air by 10 to 100 times, significantly accelerating water evaporation. This evaporation process not only removes a significant amount of heat from the intercooler but also humidifies the air, optimizing the operating conditions of the fuel cell stack.

[0043] Obviously, the embodiments described above are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0044] It should be understood that when the terms "first," "second," etc. are used in the claims, specification, and drawings of this application, they are only used to distinguish different objects, rather than to describe a specific order. The terms "comprise" and "comprising" used in the specification and claims of this application indicate the presence of the described features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or combinations thereof.

Claims

1. An air-cooled UAV hydrogen fuel cell control system, used to cool down the air compressor when the temperature is too high, characterized in that: include: A spray-type intercooler, used to spray the stack wastewater as a cooling medium to cool the hot air discharged from the air compressor; A wastewater circulation system, used for conveying the stack wastewater to the spray nozzle of the spray-type intercooler; A reinforcement learning control system is used to execute a control program to control the spray intercooler and the wastewater circulation system, wherein the control program includes: A state perception module is used to obtain a 27-dimensional state vector through a sensor group. The state vector includes the current and historical sampling sequence of the stack temperature and its rate of change, intake air humidity, hydrogen pressure, and wastewater tank level. The decision execution module is used to extract 128-dimensional hidden state features from the state vector through the LSTM network to predict the temperature rise trend in the next 3 seconds, and then input the hidden state features into the SAC strategy network to generate a four-dimensional continuous action vector through the fully connected layer: ， The four-dimensional continuous motion vector is mapped to the actuators of the spray intercooler and the wastewater circulation system, where △ n comp is the air compressor speed adjustment amount, △ D valve is the change in the opening of the hydrogen fuel valve, △ PWM fan is the power change of the cooling fan of the spray type intercooler, △ Q spray A spray flow rate for driving the wastewater circulation system; Among them, the △ in the four-dimensional continuous motion vector Q spray Subject to the interval [-3 mL / s ,+3 mL / s ] constraints and Constraints, where is the maximum spray flow rate, H Is the real-time liquid level height; The dynamic range of the four-dimensional continuous motion vector satisfies: 。 2. The system according to claim 1, wherein: The spray mechanism of the spray-type intercooler consists of copper-aluminum composite fins and ultrasonic atomizing nozzles. The copper-aluminum composite fins have a thickness of 0.3 mm and a spacing of 1.5 mm. The droplet size generated by the ultrasonic atomizing nozzle is ≤50 μm.

3. The system according to claim 1, wherein: The wastewater circulation system consists of a filter and a pressure pump. The filtration accuracy of the filter is 5 μm, and the pressure pump provides a pressure of 0.15 MPa.

4. The system according to claim 1, wherein: The LSTM network processes the state vectors of 10 consecutive sampling points in a sliding window manner, and the prediction of the temperature rise trend in the next 3 seconds is used to compensate for the thermal inertia delay of the fuel cell stack.

5. The system according to claim 1, wherein: The SAC policy network initializes weights through transfer learning. The pre-trained model is generated based on 10,000 hours of simulation data, and the online fine-tuning cycle updates the network weights every 50 hours of flight.

6. The system according to claim 1, wherein: The wastewater circulation system is equipped with a liquid level-spray linkage constraint. When the liquid level of the wastewater box is lower than the threshold, it is forced to ; When the standard deviation of the battery stack cell voltage is greater than 50mV, switch to the preset PID rule base control.

7. The system according to claim 1, wherein: The sensor group includes a MEMS temperature and humidity sensor deployed at the outlet of the intercooler, a hydrogen pressure sensor deployed in the hydrogen pipeline, a voltage acquisition board deployed on the bipolar plate leads of the fuel cell stack, a liquid level meter deployed in the wastewater box, and a laser tachometer deployed on the outside of the air compressor turbine.

8. The system according to claim 2, wherein: The surface of the copper-aluminum composite fin of the spray-type intercooler is provided with a hydrophobic coating, and the contact angle of the hydrophobic coating with the wastewater droplets is greater than 120°.

Images