Flight power distribution method for hydrogen-lithium hybrid fuel cell of multi-rotor unmanned aerial vehicle

Abstract

The invention discloses a flight power distribution method for a hydrogen-lithium hybrid fuel cell of a multi-rotor unmanned aerial vehicle. The flight power distribution method comprises the following steps: collecting flight mission parameters, environment parameters and load change parameters in real time; initializing a power distribution strategy of the hydrogen fuel cell and the lithium battery based on deep reinforcement learning; then, combining lithium battery charge-discharge cycle data and fuel battery start-stop times to construct a life prediction model, and dynamically adjusting a power distribution proportion according to the residual life; meanwhile, a combined health management module is added to monitor the states of the fuel cell and the lithium battery, so that over-discharge or over-charge is avoided; and finally, according to the real-time data and the feedback of the joint health management module, continuously optimizing a power distribution strategy by using deep reinforcement learning, and realizing dynamic balance among the targets of endurance, efficiency, safety and the like. The flight performance and the energy utilization efficiency of the multi-rotor unmanned aerial vehicle can be improved, the service life of the system is prolonged, and the flight safety is ensured.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) technology, and in particular to a method for distributing flight power of a multi-rotor UAV powered by a hydrogen-lithium hybrid fuel cell. Background Technology

[0002] As multi-rotor drone technology continues to develop, its application in long-endurance missions is limited by energy supply issues. Traditional single-battery power supply methods are insufficient to meet the demands of long-duration flights, while hydrogen fuel cells, although possessing high energy density, suffer from problems in practical applications such as complex start-stop control and poor power distribution coordination when combined with lithium batteries.

[0003] Existing power allocation methods are mostly based on fixed rules or simple algorithms, failing to fully consider the impact of multiple factors such as flight mission characteristics, environmental disturbances, and load changes. This makes it difficult to achieve dynamic optimization among multiple objectives, including range, efficiency, and safety. Furthermore, there is a lack of effective means for managing the lifespan of lithium batteries and fuel cells, making it impossible to accurately predict lifespan and rationally adjust power allocation strategies based on battery charge-discharge cycle data and fuel cell start-stop cycles. Simultaneously, existing discharge protection circuits are insufficient for joint health management of fuel cells and lithium batteries in hydrogen-lithium hybrid systems, easily leading to over-discharge or over-charge phenomena that affect the overall system performance and safety. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the existing technology and propose a method for power distribution of hydrogen-lithium hybrid fuel cells in multi-rotor UAVs.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A method for distributing flight power of a multi-rotor UAV using a hydrogen-lithium hybrid fuel cell includes the following specific steps:

[0007] S1: Data Acquisition: Real-time acquisition of flight mission parameters, environmental parameters, and load change parameters of the multi-rotor UAV using sensors;

[0008] S2: Power allocation strategy initialization based on deep reinforcement learning (DRL): Construct a deep reinforcement learning (DRL) model to initialize the power allocation ratio of hydrogen fuel cells and lithium batteries according to the pre-set flight mission objectives and initial environmental conditions;

[0009] S3: Lifetime prediction model construction and power adjustment: Combining the charge and discharge cycle data of lithium batteries and the start-stop count of fuel cells, a lifetime prediction model is constructed. Based on the working state of lithium batteries and fuel cells under the current power allocation strategy, the remaining lifetime is predicted, and the power allocation ratio is dynamically adjusted according to the remaining lifetime to balance the overall lifetime of the system.

[0010] S4: Joint Health Management Module Monitoring and Protection: Add a joint health management module for fuel cells and lithium batteries to monitor battery parameters in real time. When any battery is detected to be close to the over-discharge or overcharge threshold, adjust the power distribution strategy to limit the charging and discharging power of the corresponding battery, avoid over-discharge or overcharge, and ensure the safe and stable operation of the battery.

[0011] S5: Power Allocation Strategy Optimization and Update: Based on the feedback information from S2-S4, the deep reinforcement learning (DRL) model is used to continuously learn and optimize the power allocation strategy, dynamically balancing the goals of endurance, efficiency, and safety, so that the multi-rotor UAV can achieve optimal power allocation in different flight phases and operating conditions, thereby improving flight performance and energy utilization efficiency.

[0012] As a further technical solution of the present invention, in S1, the flight mission parameters include flight attitude, flight speed, flight altitude and flight path; the environmental parameters include wind speed, air pressure, temperature and altitude; and the load change parameters include changes in the power requirements of the mission equipment and changes in the weight of the UAV itself.

[0013] As a further technical solution of the present invention, in S2, the pre-set flight mission objectives include maximizing endurance, maximizing energy efficiency, and prioritizing flight safety.

[0014] As a further technical solution of the present invention, S2 specifically includes:

[0015] S21: State Space Definition: Quantizing flight mission parameters, environmental parameters, and load variation parameters into state vectors. ,in Represents the quantization value of flight attitude. This represents the normalized value of flight speed. This represents the normalized value of flight altitude. This represents a quantified value indicating the complexity of the flight path. This represents the normalized value of wind speed. This represents the normalized value of air pressure. Represents the normalized value of temperature. Represents the altitude normalized value. This indicates the rate of change in the power requirements of the task equipment. This indicates the rate of change of the drone's own weight;

[0016] S22: Action Space Definition: Action Vector ,in The initial power allocation ratio for the hydrogen fuel cell. The initial power allocation ratio for the lithium battery, with values ​​ranging from [value range missing]. And satisfy ;

[0017] S23: Deep Reinforcement Learning (DRL) Model Structure: Employs a Deep Q-Network (DQN), whose network structure includes an input layer, two hidden layers, and an output layer. The number of neurons in the input layer is... (with state vector) (The dimensions are consistent), the number of hidden layer neurons are 64 and 32 respectively, and the number of output layer neurons is 2 (corresponding to the two components of the action vector A). The activation function is the ReLU function, that is, for input x, output x = 2πf(x) / x ... ;

[0018] S24: Initialization process: Under the pre-set flight mission objectives and initial environmental conditions, the current state vector is... Input a deep Q-network (DQN), and the network calculates the Q value of the output action vector A through forward propagation;

[0019] The specific calculation process is as follows: Linear transformation from the input layer to the first hidden layer: ,in The weight matrix from the input layer to the hidden layer has a dimension of . , Let be the bias vector of the hidden layer, with dimension . The activation function of the first hidden layer is calculated as follows: Linear transformation from the first hidden layer to the second hidden layer: ,in The weight matrix from the first hidden layer to the second hidden layer has a dimension of . Let be the bias vector of the second hidden layer, with dimension . The activation function of the second hidden layer is calculated as follows: Linear transformation from the second hidden layer to the output layer: ,in This is the weight matrix from the second hidden layer to the output layer, with dimension [missing information]. Let be the bias vector of the output layer, with dimension . The final output action vector of Value The corresponding initial power allocation ratios are as follows: Initially, the model parameters (weight matrix) and bias vector The value is determined through random initialization or pre-training.

[0020] As a further technical solution of the present invention, in S3, the charge-discharge cycle data of the lithium battery includes charge-discharge current, charge-discharge voltage, charge-discharge time, and number of charge-discharge cycles; the number of start-stop cycles of the fuel cell is counted by the counting module of the fuel cell controller.

[0021] As a further technical solution of the present invention, S3 specifically includes:

[0022] S31: Construction of Lithium-ion Battery Life Prediction Model: Define the initial capacity of the lithium-ion battery as... (Unit: Ah), current capacity is (Unit: Ah) Based on the charge-discharge cycle data of lithium batteries, the remaining lifespan is predicted using a capacity decay model. (Unit: number of iterations), the formula is: ,in This is the lithium battery life degradation coefficient, with a value ranging from 0 to 1, which can be determined based on the lithium battery's materials, manufacturing process, and usage history data.

[0023] S32: Construction of Fuel Cell Lifetime Prediction Model: Define the initial performance parameters of the fuel cell as follows (Unit: W), current performance parameters are (Unit: W), based on the number of start-stop cycles of the fuel cell. Predicting remaining lifetime using a performance degradation model (Unit: Number of start-stop cycles), the formula is: ,in This is the fuel cell life degradation coefficient, with a value ranging from 0 to 1. It is mainly affected by the degradation characteristics of key components of the fuel cell, such as the catalyst and proton exchange membrane.

[0024] S33: Comprehensive Lifetime Assessment: Calculate the comprehensive weighting coefficients for the lifetime of lithium batteries and fuel cells. and ,satisfy The choice between lithium batteries and fuel cells can be determined based on their importance in multi-rotor drone missions and their impact on overall performance. Typically, if lithium batteries are used to provide short-term high-power support, and fuel cells are used for long-term stable power supply, then... A smaller value can be chosen, and vice versa;

[0025] S34: Power Adjustment Strategy: Calculate the current remaining lifespan of the lithium battery and fuel cell based on their operating states under the current power allocation strategy. and Then, according to the comprehensive life assessment formula Determine the overall remaining life of the hydrogen-lithium hybrid system for multi-rotor UAVs. ;

[0026] To balance system lifespan, the current power allocation ratio of hydrogen fuel cells will be adjusted. , Define the power adjustment factor in relation to remaining lifetime. ,in This represents the desired target value for the remaining system lifetime.

[0027] The dynamically adjusted power allocation ratio is as follows: ,

[0028] Meanwhile, to ensure the stability and continuity of power distribution, a smooth transition factor is introduced. (Values ​​range from 0 to 1), and the final power allocation ratio is determined as follows: .

[0029] As a further technical solution of the present invention, in S4, the joint health management module includes a signal acquisition unit, a data processing unit and a control execution unit. The signal acquisition unit is used to acquire battery status parameters, the data processing unit analyzes and processes the received parameters, and the control execution unit adjusts the power allocation strategy according to the processing results.

[0030] As a further technical solution of the present invention, S4 specifically includes:

[0031] S41: Battery Status Parameter Monitoring: Set the over-discharge protection voltage threshold for the lithium battery to... (Unit: V), Overcharge protection voltage threshold is (Unit: V), its current voltage is (Unit: V); Set the over-discharge protection current threshold for the fuel cell to be... (Unit: A), Overcharge protection current threshold is (Unit: A), its current is (Unit: A); The safe temperature ranges for lithium batteries and fuel cells are also set as follows: (unit: )and (unit: Their current temperatures are respectively (unit: )and (unit: );

[0032] S42: Power Distribution Adjustment Trigger Condition Judgment: When or or or When this is triggered, adjustments to the power distribution related to the lithium battery are made; when or or or At that time, adjustments to the power distribution related to the fuel cell are triggered;

[0033] S43: Adjustment of power allocation strategy algorithm:

[0034] When it is necessary to adjust the power distribution of the lithium battery, calculate the voltage deviation coefficient of the lithium battery. (when (time) or (when (time), and temperature deviation coefficient (when (time) or (when (At that time); the larger of the absolute values ​​of the above deviation coefficients is taken as the comprehensive deviation coefficient of the lithium battery. Then calculate the lithium battery power distribution adjustment amount. ,in This represents the current power allocation ratio for the lithium battery; simultaneously, the power allocation ratio for the fuel cell is adjusted accordingly. ,make sure ;

[0035] When it is necessary to adjust the power distribution of the fuel cell, calculate the current deviation coefficient of the fuel cell. when (time) or (when (time), and temperature deviation coefficient (when (time) or (when (At that time); the larger of the absolute values ​​of the above deviation coefficients is taken as the overall deviation coefficient of the fuel cell. Then calculate the fuel cell power distribution adjustment amount. ,in This represents the current power allocation ratio for the fuel cell; simultaneously, the power allocation ratio for the lithium battery is adjusted accordingly. ,make sure ;

[0036] If the adjustment conditions for both the lithium battery and the fuel cell are triggered simultaneously, the deviation coefficients of both are taken into account, and the adjustment amounts are calculated and adjusted separately according to the above method to finally obtain a new power distribution ratio. and This limits the charging and discharging power of the corresponding battery to avoid over-discharging or over-charging, ensuring that the battery operates within a safe and stable power range.

[0037] As a further technical solution of the present invention, in step S5, a reward function is introduced during the optimization of the power allocation strategy. The construction of the reward function considers multiple factors such as endurance, energy efficiency and flight safety, so as to guide the deep reinforcement learning (DRL) model to converge toward the optimal solution.

[0038] As a further technical solution of the present invention, S5 specifically includes:

[0039] S51: Reward Function The system is designed taking into account three objectives: range, efficiency, and safety. The specific calculation formula is as follows: ;in: The weighting coefficients for the goals of range, efficiency, and safety are respectively, satisfying... The priority of the flight mission can be dynamically adjusted. For example, in long-endurance reconnaissance missions, the priority can be set. ;

[0040] These are respectively: battery life bonus, efficiency bonus, and safety bonus;

[0041] Battery life bonus The formula is calculated as follows: The remaining flight time is calculated based on the ratio of the estimated remaining flight time to the maximum possible remaining flight time under the current power allocation strategy. ,in Calculated using the current remaining energy and power consumption rate. The maximum remaining flight time estimated under an ideal power allocation strategy; efficiency bonus. The energy conversion efficiency is measured by the ratio of the current power allocation strategy to the historical highest energy conversion efficiency, as shown in the formula: ,in It is calculated by the efficiency weighted average of hydrogen fuel cells and lithium batteries at the current power output, with the weights representing their power allocation ratios. The highest energy conversion efficiency under different operating conditions, determined in advance through experiments or simulations;

[0042] Safety Rewards The safety assessment is based on battery state. When the voltage, current, temperature, and other parameters of both the hydrogen fuel cell and the lithium battery are within safe ranges, If any battery parameter exceeds the safe range, the degradation will be calculated exponentially based on the degree of exceedance, using the following formula: ,in This is the attenuation coefficient (value 0.1). The relative extent to which a parameter exceeds the safe range (e.g., the percentage by which the voltage exceeds the upper limit);

[0043] S52: Deep Reinforcement Learning (DRL) Algorithm Update: Introduce a target network with the same structure as the Deep Q Network (DQN), but whose parameters are periodically copied from the Deep Q Network (DQN) (e.g., updated every 1000 steps) to stabilize the training process. The target network is used to generate the target Q value.

[0044] S53: Power Allocation Strategy Optimization Process:

[0045] Obtain a reward signal after implementing the new power allocation strategy. The experience data is stored in a buffer, and the policy network and value network are updated by sampling data from the buffer at preset training intervals (such as every 10 time steps or after flying a certain distance).

[0046] Strategy Smoothing and Stability: To prevent frequent and large fluctuations in the power allocation ratio from affecting flight stability, a smoothing factor is introduced after each update of the power allocation ratio. (Value 0.8), according to the formula , The final power allocation ratio is smoothed.

[0047] S54: Flight Phase and Operating Condition Adaptation: At the beginning of each flight phase, the weighting coefficients in the reward function are adjusted according to the typical characteristics of that phase (e.g., high power output for short periods during takeoff, energy efficiency balance during cruise, and power allocation adjustment based on specific mission actions during mission execution). , , Dynamic adjustments are made to guide the agent to learn power allocation strategies that adapt to the needs of different stages. At the same time, by combining transfer learning technology, some of the experience and knowledge learned in the previous flight stage (such as the efficient power allocation mode mastered in the cruise stage) are transferred to the current stage, which accelerates the strategy optimization process and reduces the exploration time in the early stage of learning.

[0048] The beneficial effects of this invention are as follows:

[0049] 1. By introducing deep reinforcement learning (DRL), and fully considering the impact of multiple factors such as flight mission, environmental disturbances and load changes, a dynamic optimization allocation strategy among objectives such as endurance, efficiency and safety is realized, which effectively improves the overall flight performance and energy utilization efficiency of multi-rotor UAVs, enabling them to better adapt to complex and ever-changing flight environments and mission requirements.

[0050] 2. By combining the charge-discharge cycle data of lithium batteries with the start-stop count of fuel cells to build a life prediction model, the battery health status can be accurately assessed, and the power distribution can be dynamically adjusted accordingly to balance the overall life of the system, extend the service life of the UAV, reduce long-term operating costs, and improve the reliability and economy of the system.

[0051] 3. A joint health management module for fuel cells and lithium batteries has been added, which further enhances the battery management function based on the original discharge protection circuit. This effectively avoids over-discharge or over-charge, enhances the safety and stability of the system, and provides a strong guarantee for the long-term stable operation of multi-rotor UAVs. Attached Figure Description

[0052] Figure 1This is a flowchart of a method for allocating flight power of a multi-rotor UAV using a hydrogen-lithium hybrid fuel cell, as proposed in this invention. Detailed Implementation

[0053] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0054] Please see the appendix Figure 1 A method for distributing flight power of a multi-rotor UAV using a hydrogen-lithium hybrid fuel cell includes the following specific steps:

[0055] S1: Data Acquisition: Real-time acquisition of flight mission parameters, environmental parameters, and load change parameters of the multi-rotor UAV using sensors;

[0056] S2: Power allocation strategy initialization based on deep reinforcement learning (DRL): Construct a deep reinforcement learning (DRL) model to initialize the power allocation ratio of hydrogen fuel cells and lithium batteries according to the pre-set flight mission objectives and initial environmental conditions;

[0057] S3: Lifetime prediction model construction and power adjustment: Combining the charge and discharge cycle data of lithium batteries and the start-stop count of fuel cells, a lifetime prediction model is constructed. Based on the working state of lithium batteries and fuel cells under the current power allocation strategy, the remaining lifetime is predicted, and the power allocation ratio is dynamically adjusted according to the remaining lifetime to balance the overall lifetime of the system.

[0058] S4: Joint Health Management Module Monitoring and Protection: Add a joint health management module for fuel cells and lithium batteries to monitor battery parameters in real time. When any battery is detected to be close to the over-discharge or overcharge threshold, adjust the power distribution strategy to limit the charging and discharging power of the corresponding battery, avoid over-discharge or overcharge, and ensure the safe and stable operation of the battery.

[0059] S5: Power Allocation Strategy Optimization and Update: Based on the feedback information from S2-S4, the deep reinforcement learning (DRL) model is used to continuously learn and optimize the power allocation strategy, dynamically balancing the goals of endurance, efficiency, and safety, so that the multi-rotor UAV can achieve optimal power allocation in different flight phases and operating conditions, thereby improving flight performance and energy utilization efficiency.

[0060] In a preferred embodiment, in S1, the flight mission parameters include flight attitude, flight speed, flight altitude, and flight path; the environmental parameters include wind speed, air pressure, temperature, and altitude; and the load change parameters include changes in the power requirements of the mission equipment and changes in the weight of the UAV itself.

[0061] In a preferred embodiment, in S2, the pre-set flight mission objectives include maximizing endurance, maximizing energy efficiency, and prioritizing flight safety.

[0062] In a preferred embodiment, S2 specifically includes:

[0063] S21: State Space Definition: Quantizing flight mission parameters, environmental parameters, and load variation parameters into state vectors. ,in Represents the quantization value of flight attitude. This represents the normalized value of flight speed. This represents the normalized value of flight altitude. This represents a quantified value indicating the complexity of the flight path. This represents the normalized value of wind speed. This represents the normalized value of air pressure. Represents the normalized value of temperature. Represents the altitude normalized value. This indicates the rate of change in the power requirements of the task equipment. This indicates the rate of change of the drone's own weight;

[0064] S22: Action Space Definition: Action Vector ,in The initial power allocation ratio for the hydrogen fuel cell. The initial power allocation ratio for the lithium battery, with values ​​ranging from [value range missing]. And satisfy ;

[0065] S23: Deep Reinforcement Learning (DRL) Model Structure: Employs a Deep Q-Network (DQN), whose network structure includes an input layer, two hidden layers, and an output layer. The number of neurons in the input layer is... (with state vector) (The dimensions are consistent), the number of hidden layer neurons are 64 and 32 respectively, and the number of output layer neurons is 2 (corresponding to the two components of the action vector A). The activation function is the ReLU function, that is, for input x, output x = 2πf(x) / x ... ;

[0066] S24: Initialization process: Under the pre-set flight mission objectives and initial environmental conditions, the current state vector is... Input a deep Q-network (DQN), and the network calculates the Q value of the output action vector A through forward propagation;

[0067] The specific calculation process is as follows: Linear transformation from the input layer to the first hidden layer: ,in The weight matrix from the input layer to the hidden layer has a dimension of . , Let be the bias vector of the hidden layer, with dimension . The activation function of the first hidden layer is calculated as follows: Linear transformation from the first hidden layer to the second hidden layer: ,in The weight matrix from the first hidden layer to the second hidden layer has a dimension of . Let be the bias vector of the second hidden layer, with dimension . The activation function of the second hidden layer is calculated as follows: Linear transformation from the second hidden layer to the output layer: ,in This is the weight matrix from the second hidden layer to the output layer, with dimension [missing information]. Let be the bias vector of the output layer, with dimension . The final output action vector of Value The corresponding initial power allocation ratios are as follows: Initially, the model parameters (weight matrix) and bias vector The value is determined through random initialization or pre-training.

[0068] In a preferred embodiment, in S3, the charge-discharge cycle data of the lithium battery includes charge-discharge current, charge-discharge voltage, charge-discharge time, and number of charge-discharge cycles; the number of start-stop cycles of the fuel cell is counted by the counting module of the fuel cell controller.

[0069] In a preferred embodiment, S3 specifically includes:

[0070] S31: Construction of Lithium-ion Battery Life Prediction Model: Define the initial capacity of the lithium-ion battery as... (Unit: Ah), current capacity is (Unit: Ah) Based on the charge-discharge cycle data of lithium batteries, the remaining lifespan is predicted using a capacity decay model. (Unit: number of iterations), the formula is: ,in This is the lithium battery life degradation coefficient, with a value ranging from 0 to 1, which can be determined based on the lithium battery's materials, manufacturing process, and usage history data.

[0071] S32: Construction of Fuel Cell Lifetime Prediction Model: Define the initial performance parameters of the fuel cell as follows (Unit: W), current performance parameters are (Unit: W), based on the number of start-stop cycles of the fuel cell. Predicting remaining lifetime using a performance degradation model (Unit: Number of start-stop cycles), the formula is: ,in This is the fuel cell life degradation coefficient, with a value ranging from 0 to 1. It is mainly affected by the degradation characteristics of key components of the fuel cell, such as the catalyst and proton exchange membrane.

[0072] S33: Comprehensive Lifetime Assessment: Calculate the comprehensive weighting coefficients for the lifetime of lithium batteries and fuel cells. and ,satisfy The choice between lithium batteries and fuel cells can be determined based on their importance in multi-rotor drone missions and their impact on overall performance. Typically, if lithium batteries are used to provide short-term high-power support, and fuel cells are used for long-term stable power supply, then... A smaller value can be chosen, and vice versa;

[0073] S34: Power Adjustment Strategy: Calculate the current remaining lifespan of the lithium battery and fuel cell based on their operating states under the current power allocation strategy. and Then, according to the comprehensive life assessment formula Determine the overall remaining life of the hydrogen-lithium hybrid system for multi-rotor UAVs. ;

[0074] To balance system lifespan, the current power allocation ratio of hydrogen fuel cells will be adjusted. , Define the power adjustment factor in relation to remaining lifetime. ,in This represents the desired target value for the remaining system lifetime.

[0075] The dynamically adjusted power allocation ratio is as follows: ,

[0076] Meanwhile, to ensure the stability and continuity of power distribution, a smooth transition factor is introduced. (Values ​​range from 0 to 1), and the final power allocation ratio is determined as follows: .

[0077] In a preferred embodiment, in S4, the joint health management module includes a signal acquisition unit, a data processing unit, and a control execution unit. The signal acquisition unit is used to acquire battery status parameters, the data processing unit analyzes and processes the received parameters, and the control execution unit adjusts the power allocation strategy based on the processing results.

[0078] In a preferred embodiment, S4 specifically includes:

[0079] S41: Battery Status Parameter Monitoring: Set the over-discharge protection voltage threshold for the lithium battery to... (Unit: V), Overcharge protection voltage threshold is (Unit: V), its current voltage is (Unit: V); Set the over-discharge protection current threshold for the fuel cell to be... (Unit: A), Overcharge protection current threshold is (Unit: A), its current is (Unit: A); The safe temperature ranges for lithium batteries and fuel cells are also set as follows: (unit: )and (unit: Their current temperatures are respectively (unit: )and (unit: );

[0080] S42: Power Distribution Adjustment Trigger Condition Judgment: When or or or When this is triggered, adjustments to the power distribution related to the lithium battery are made; when or or or At that time, adjustments to the power distribution related to the fuel cell are triggered;

[0081] S43: Adjustment of power allocation strategy algorithm:

[0082] When it is necessary to adjust the power distribution of the lithium battery, calculate the voltage deviation coefficient of the lithium battery. (when (time) or (when (time), and temperature deviation coefficient (when (time) or (when (At that time); the larger of the absolute values ​​of the above deviation coefficients is taken as the comprehensive deviation coefficient of the lithium battery. Then calculate the lithium battery power distribution adjustment amount. ,in This represents the current power allocation ratio for the lithium battery; simultaneously, the power allocation ratio for the fuel cell is adjusted accordingly. ,make sure ;

[0083] When it is necessary to adjust the power distribution of the fuel cell, calculate the current deviation coefficient of the fuel cell. when (time) or (when (time), and temperature deviation coefficient (when (time) or (when (At that time); the larger of the absolute values ​​of the above deviation coefficients is taken as the overall deviation coefficient of the fuel cell. Then calculate the fuel cell power distribution adjustment amount. ,in This represents the current power allocation ratio for the fuel cell; simultaneously, the power allocation ratio for the lithium battery is adjusted accordingly. ,make sure ;

[0084] If the adjustment conditions for both the lithium battery and the fuel cell are triggered simultaneously, the deviation coefficients of both are taken into account, and the adjustment amounts are calculated and adjusted separately according to the above method to finally obtain a new power distribution ratio. and This limits the charging and discharging power of the corresponding battery to avoid over-discharging or over-charging, ensuring that the battery operates within a safe and stable power range.

[0085] In a preferred embodiment, in S5, a reward function is introduced during the optimization of the power allocation strategy. The construction of the reward function considers multiple factors such as endurance, energy efficiency, and flight safety to guide the deep reinforcement learning (DRL) model to converge toward the optimal solution.

[0086] In a preferred embodiment, S5 specifically includes:

[0087] S51: Reward Function The system is designed taking into account three objectives: range, efficiency, and safety. The specific calculation formula is as follows: ;in: The weighting coefficients for the goals of range, efficiency, and safety are respectively, satisfying... The priority of the flight mission can be dynamically adjusted. For example, in long-endurance reconnaissance missions, the priority can be set. ;

[0088] These are respectively: battery life bonus, efficiency bonus, and safety bonus;

[0089] Battery life bonus The formula is calculated as follows: The remaining flight time is calculated based on the ratio of the estimated remaining flight time to the maximum possible remaining flight time under the current power allocation strategy. ,in Calculated using the current remaining energy and power consumption rate. The maximum remaining flight time estimated under an ideal power allocation strategy; efficiency bonus. The energy conversion efficiency is measured by the ratio of the current power allocation strategy to the historical highest energy conversion efficiency, as shown in the formula: ,in It is calculated by the efficiency weighted average of hydrogen fuel cells and lithium batteries at the current power output, with the weights representing their power allocation ratios. The highest energy conversion efficiency under different operating conditions, determined in advance through experiments or simulations;

[0090] Safety Rewards The safety assessment is based on battery state. When the voltage, current, temperature, and other parameters of both the hydrogen fuel cell and the lithium battery are within safe ranges, If any battery parameter exceeds the safe range, the degradation will be calculated exponentially based on the degree of exceedance, using the following formula: ,in This is the attenuation coefficient (value 0.1). The relative extent to which a parameter exceeds the safe range (e.g., the percentage by which the voltage exceeds the upper limit);

[0091] S52: Deep Reinforcement Learning (DRL) Algorithm Update: Introduce a target network with the same structure as the Deep Q Network (DQN), but whose parameters are periodically copied from the Deep Q Network (DQN) (e.g., updated every 1000 steps) to stabilize the training process. The target network is used to generate the target Q value.

[0092] S53: Power Allocation Strategy Optimization Process:

[0093] Obtain a reward signal after implementing the new power allocation strategy. The experience data is stored in a buffer, and the policy network and value network are updated by sampling data from the buffer at preset training intervals (such as every 10 time steps or after flying a certain distance).

[0094] Strategy Smoothing and Stability: To prevent frequent and large fluctuations in the power allocation ratio from affecting flight stability, a smoothing factor is introduced after each update of the power allocation ratio. (Value 0.8), according to the formula , The final power allocation ratio is smoothed.

[0095] S54: Flight Phase and Operating Condition Adaptation: At the beginning of each flight phase, the weighting coefficients in the reward function are adjusted according to the typical characteristics of that phase (e.g., high power output for short periods during takeoff, energy efficiency balance during cruise, and power allocation adjustment based on specific mission actions during mission execution). , , Dynamic adjustments are made to guide the agent to learn power allocation strategies that adapt to the needs of different stages. At the same time, by combining transfer learning technology, some of the experience and knowledge learned in the previous flight stage (such as the efficient power allocation mode mastered in the cruise stage) are transferred to the current stage, which accelerates the strategy optimization process and reduces the exploration time in the early stage of learning.

[0096] As can be seen from the above description, the above embodiments of the present invention achieve the following technical effects: by introducing deep reinforcement learning (DRL), the influence of multiple factors such as flight mission, environmental disturbance and load change is fully considered, and a dynamic optimization allocation strategy among objectives such as endurance, efficiency and safety is realized, which effectively improves the comprehensive flight performance and energy utilization efficiency of multi-rotor UAVs, enabling them to better adapt to complex and ever-changing flight environments and mission requirements.

[0097] By combining the charge-discharge cycle data of lithium batteries with the start-stop count of fuel cells to build a lifespan prediction model, the battery health status can be accurately assessed, and the power distribution can be dynamically adjusted accordingly. This balances the overall lifespan of the system, extends the service life of the drone, reduces long-term operating costs, and improves the reliability and economy of the system.

[0098] The addition of a joint health management module for fuel cells and lithium batteries further enhances battery management functions based on the original discharge protection circuit, effectively avoiding over-discharge or over-charge phenomena, enhancing the safety and stability of the system, and providing strong support for the long-term stable operation of multi-rotor drones.

[0099] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of the invention (including the claims) is limited to these examples; within the framework of the invention, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of the different aspects of the invention as described above, which are not provided in the details for the sake of brevity.

[0100] This invention is intended to cover all such substitutions, modifications, and variations falling within the broad scope of the claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A multi-rotor unmanned aerial vehicle hydrogen-lithium hybrid fuel cell flight power distribution method, characterized in that, The method comprises the following specific steps: S1: data acquisition: real-time acquisition of flight task parameters, environmental parameters and load change parameters of the multi-rotor unmanned aerial vehicle by using sensors; S2: power distribution strategy initialization based on deep reinforcement learning: constructing a deep reinforcement learning model, initializing the power distribution ratio of the hydrogen fuel cell and the lithium battery according to the pre-set flight task target and initial environmental condition; S3: life prediction model construction and power adjustment: combining the charge-discharge cycle data of the lithium battery and the start-stop number of the fuel cell, constructing a life prediction model, predicting the remaining life of the lithium battery and the fuel cell according to the working state of the lithium battery and the fuel cell under the current power distribution strategy, and dynamically adjusting the power distribution ratio according to the remaining life; S4: joint health management module monitoring and protection: adding a joint health management module of the fuel cell and the lithium battery to monitor the battery parameters in real time, and adjusting the power distribution strategy when any battery state is close to the over-discharge or over-charge threshold; S5: power distribution strategy optimization update: according to the feedback information of S2-S4, using the deep reinforcement learning model to continuously learn and optimize the power distribution strategy, dynamically balancing the endurance, efficiency and safety targets, and realizing the optimal power distribution.

2. The multi-copter unmanned aerial vehicle hydrogen-lithium hybrid fuel cell flight power distribution method of claim 1, wherein, In the S1, the flight task parameters include flight attitude, flight speed, flight height and flight path; the environmental parameters include wind speed, air pressure, temperature and altitude; and the load change parameters include power demand change of the task equipment and weight change of the unmanned aerial vehicle itself.

3. The method of claim 1, wherein the hydrogen lithium hybrid fuel cell flight power distribution method of a multi-rotor unmanned aerial vehicle is characterized by, In the S2, the pre-set flight task target includes maximum endurance time, maximum energy efficiency and flight safety priority.

4. The multi-copter unmanned aerial vehicle hydrogen-lithium hybrid fuel cell flight power distribution method of claim 3, wherein, The S2 specifically comprises: S21 : State space definition: Quantify the flight mission parameters, environmental parameters, and payload variation parameters into a state vector ; S22: Action space definition: action vector wherein is the initial power distribution ratio for the hydrogen fuel cell, is the initial power distribution ratio for the lithium battery; S23: deep Q network structure: adopting a deep Q network, the network structure of which comprises an input layer, two hidden layers and an output layer; S24: initialization process: under the pre-set flight task target and initial environmental conditions, the current state vector The input deep Q network calculates the Q value of the output action vector A through forward propagation.

5. The method of claim 1, wherein, In the S3, the charge-discharge cycle data of the lithium battery includes charge-discharge current, charge-discharge voltage, charge-discharge time and charge-discharge cycle number; and the start-stop number of the fuel cell is counted by a counting module of the fuel cell controller.

6. The multi-copter unmanned aerial vehicle hydrogen-lithium hybrid fuel cell flight power distribution method of claim 5, wherein, The S3 specifically comprises: S31: Lithium battery life prediction model construction: define the initial capacity of the lithium battery as , the current capacity as , and according to the charge and discharge cycle data of the lithium battery, predict the remaining life by using the capacity attenuation model , the formula is , wherein is the lithium battery life attenuation coefficient; S32: Fuel cell life prediction model construction: define the initial performance parameters of the fuel cell as , the current performance parameters as , the start-stop number of the fuel cell as , and predict the remaining life by using the performance attenuation model , the formula being , wherein is the fuel cell life attenuation coefficient; S33: Comprehensive life evaluation: calculate the comprehensive weight coefficient of the life of lithium battery and fuel cell and , meet ; S34: power adjustment strategy: according to the working state of lithium battery and fuel cell under the current power distribution strategy, the current remaining life is calculated and , the comprehensive remaining life of the hydrogen-lithium hybrid system of the multi-rotor unmanned aerial vehicle is determined according to the comprehensive life evaluation formula ;​ hydrogen fuel cell power allocation proportion of the current power allocation proportion , associated with the remaining life, defining a power adjustment factor , wherein is a desired system remaining life target value; The power allocation ratio after dynamic adjustment is as follows: , At the same time, a smooth transition factor is introduced The final power distribution ratio is determined as: .

7. The method of claim 1, wherein, In the S4, the joint health management module comprises a signal acquisition unit, a data processing unit and a control execution unit, the signal acquisition unit is used to acquire battery state parameters, the data processing unit analyzes and processes the received parameters, and the control execution unit adjusts the power distribution strategy according to the processing result.

8. The multi-copter unmanned aerial vehicle hydrogen-lithium hybrid fuel cell flight power distribution method of claim 7, wherein, The S4 specifically comprises: S41: battery state parameter monitoring: set the over-discharge protection voltage threshold of the lithium battery as , the over-charge protection voltage threshold as , the current voltage as ; set the over-discharge protection current threshold of the fuel cell as , the over-charge protection current threshold as , the current current as ; set the temperature safety range of the lithium battery and the fuel cell as and respectively, and the current temperature as and respectively; S42: Power distribution adjustment trigger condition judgment: when or or or , the adjustment of the lithium battery related power distribution is triggered; when or or or , the adjustment of the fuel cell related power distribution is triggered; S43: adjusting power distribution strategy algorithm: When the lithium battery power distribution needs to be adjusted, the voltage deviation coefficient of the lithium battery is calculated or , and the temperature deviation coefficient or ; take the absolute value of the larger one of the above deviation coefficients as the comprehensive deviation coefficient of the lithium battery , and then calculate the lithium battery power distribution adjustment amount , wherein is the current power distribution ratio of the lithium battery; at the same time, the power distribution ratio of the fuel cell is adjusted as , to ensure ; When the fuel cell power distribution needs to be adjusted, the current deviation coefficient of the fuel cell is calculated or , and the temperature deviation coefficient or ; take the absolute value of the larger one of the above deviation coefficients as the comprehensive deviation coefficient of the fuel cell , then calculate the fuel cell power distribution adjustment amount , wherein is the current fuel cell power distribution ratio; at the same time, the power distribution ratio of the lithium battery is adjusted as , to ensure ; If the adjustment conditions of the lithium battery and the fuel cell are triggered at the same time, the deviation coefficients of both are comprehensively considered, the adjustment amounts are calculated and adjusted according to the above method respectively, and finally the new power distribution ratio is obtained and .

9. The method of claim 1, wherein, In the S5, a reward function is introduced in the process of optimizing the power distribution strategy, and the reward function is constructed by considering multiple factors such as endurance time, energy efficiency and flight safety to guide the deep reinforcement learning model to converge to the optimal solution.

10. A method for distributing flight power of a multi-rotor UAV using a hydrogen-lithium hybrid fuel cell according to claim 9, characterized in that, The S5 specifically comprises: S51: reward function : considering the three targets of endurance, efficiency and safety, the specific calculation formula is: ; wherein: are the weight coefficients of endurance, efficiency and safety targets, are the endurance reward, efficiency reward and safety reward, respectively; S52: deep reinforcement learning algorithm update: introducing a target network, which has the same structure as the deep Q network, but the parameters are periodically copied from the deep Q network; S53: power distribution strategy optimization process: After the new power allocation strategy is executed, a reward signal is obtained and the experience data is stored to a buffer, and the policy network and the value network are updated according to preset training intervals by sampling data from the buffer. Strategy smoothing and stabilization: introduce a smoothing factor after each update of the power allocation ratio , the final power allocation ratio is smoothed according to the formula 、 ​ S54: Flight phase and working condition adaptation: At the beginning of each flight phase, the weight coefficients in the reward function are dynamically adjusted according to the typical characteristics of the phase, guiding the agent to learn the power distribution strategy that adapts to the requirements of different phases. , , ​

Images