Hybrid propulsion system of hovercar and energy management method of hybrid propulsion system

Abstract

The invention discloses an aerocar hybrid propulsion system and an energy management method thereof. The method comprises the following steps: establishing a power coupling model of the aerocar hybrid propulsion system; designing a state variable, an action space, a reward function and a performance constraint of the improved SAC algorithm; according to the method, the hyper-parameters of the improved SAC algorithm are optimized, the power coupling model of the hybrid propulsion system of the hovercar is solved, off-line iterative training is carried out under representative working conditions, actions can be generated according to input states, corresponding rewards are obtained, and the multi-objective optimization effect is obtained. According to the hydrogen fuel cell-lithium ion battery-super capacitor three-source hybrid propulsion system, energy is rapidly compensated in the instantaneous high-power flight stage through the super capacitor, and the effects of rapidly responding to the instantaneous high-power requirement of the system and effectively protecting the hydrogen fuel cell and the lithium ion battery are achieved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of hybrid energy management of hybrid air vehicles, and particularly relates to a hybrid propulsion system of a hybrid air vehicle and an energy management method thereof. BACKGROUND

[0002] The hybrid air vehicle operates in multiple working conditions such as take-off, climbing, hovering and cruising, and therefore has much higher requirements for the power density and dynamic response capability of the propulsion system than ground vehicles, especially in the take-off and climbing stages, the propulsion system needs to have the ability of instantaneous high-power output and rapid power regulation. The traditional hydrogen fuel cell-lithium ion battery dual-source hybrid propulsion system has been widely used in ground vehicles, and the dual-source system is provided with continuous power by the fuel cell and power compensation by the lithium ion battery. However, the fuel cell has a slow response speed and the lithium ion battery has a limited instantaneous output capacity, and when responding to the instantaneous and variable power demand changes of the hybrid air vehicle, the power response of the dual-source system is lagged and the peak power cannot be met. In view of the above situation, a super capacitor is introduced into the traditional dual-source system to form a hydrogen fuel cell-lithium ion battery-super capacitor three-source hybrid propulsion system, the super capacitor has high power density and fast charging and discharging characteristics, can provide power compensation in the instantaneous high-power stage, absorb instantaneous energy, rapidly respond to power demand, and effectively protect the hydrogen fuel cell and the lithium ion battery. Energy management of the three energy sources can meet the power demand of the hybrid air vehicle in multiple working conditions and improve the dynamic response capability of the system.

[0003] Existing researches generally use deep reinforcement learning methods to manage the energy of the hybrid propulsion system with hydrogen fuel cell as the main power source. Among them, the traditional soft actor-critic (SAC) algorithm, as an improved policy gradient algorithm in deep reinforcement learning, is widely used due to its high stability, fast convergence speed and excellent strategy exploration capability in continuous action space. However, it is difficult to solve the energy management problem of the three-source hybrid propulsion system only by using the traditional SAC algorithm: the energy density, power response speed and working characteristics of each energy source of the three-source hybrid propulsion system are significantly different, the number of state variables and action variables to be observed and controlled is doubled, the dimensions of the state space and the action space are doubled when using the traditional SAC algorithm for solution, and there are significant differences between the variables, which all lead to a doubling of the time for action exploration by the existing strategy in the traditional SAC algorithm, and the exploration of invalid actions that violate the safety constraints of the three-source hybrid propulsion system, resulting in interrupted exploration, a doubling of the convergence speed, a doubling of the training time and a poor energy management effect.

[0004] Therefore, how to overcome the above problems, improve the solution speed and reliability of the energy management method while meeting the safety constraints of the three-source hybrid propulsion system, is a technical problem to be solved in the field. SUMMARY

[0005] To address the shortcomings of the existing technologies, the present invention aims to provide a hybrid propulsion system for flying cars and its energy management method. This method solves the problems of power response lag and inability to meet peak power requirements when the existing hydrogen fuel cell-lithium-ion battery dual-source hybrid propulsion system meets the high dynamic power demands of flying cars. It also addresses the problems of the traditional SAC algorithm for energy management of three-source hybrid propulsion systems, such as actions that easily violate system safety constraints, slow convergence speed, and deterioration of energy management effect. The invention aims to improve the solution speed and reliability of energy management of three-source hybrid propulsion systems in complex multimodal environments between air and ground.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] The present invention provides a hybrid propulsion system for a flying car, comprising: a power output module, an energy source module, and a power distribution module;

[0008] The power output module includes: a flight propulsion unit and a ground drive unit;

[0009] The flight propulsion unit includes: a rotor motor, a flight propulsion mode gearbox, and a coaxial reversing unit. The input end of the rotor motor is electrically connected to the energy source module, and the output end is mechanically connected to the input end of the flight mode gearbox. The coaxial reversing unit is mechanically connected to the output end of the flight propulsion mode gearbox.

[0010] The ground drive unit includes: a hub motor, a ground driving mode gearbox, and wheels. The input end of the hub motor is electrically connected to the energy source module, and the output end is mechanically connected to the input end of the ground driving mode gearbox. The wheels are mechanically connected to the output end of the ground driving mode gearbox.

[0011] The energy source module includes: a hydrogen fuel cell stack, a hydrogen storage tank, a lithium-ion battery pack, and a supercapacitor;

[0012] The input end of the hydrogen fuel cell stack is connected to the hydrogen storage tank, and the output end is electrically connected to the energy source module, serving as the main energy source for the flying car.

[0013] The output of the lithium-ion battery pack is electrically connected to the energy source module and is connected in parallel with the supercapacitor group to form a composite power source, serving as the auxiliary energy source for the flying car.

[0014] The output of the supercapacitor is electrically connected to the energy source module;

[0015] The power distribution module includes: an energy management unit, a DC / AC converter, a DC / DC converter, and a power bus;

[0016] The energy management unit dynamically calculates and decomposes the power output and charging / discharging strategies of each energy source based on the total power demand for land and air travel, the real-time status of each energy source, and the current land and air travel mode.

[0017] DC / DC converters, including: fuel cell-side DC / DC converters, lithium-ion battery-side DC / DC converters, and supercapacitor-side DC / DC converters;

[0018] One end of the DC / DC converter on the fuel cell side is electrically connected to the output of the hydrogen fuel cell stack, and the other end is electrically connected to the power bus; one end of the DC / DC converter on the lithium-ion battery side is electrically connected to the output of the lithium-ion battery pack, and the other end is electrically connected to the power bus; one end of the DC / DC converter on the supercapacitor side is electrically connected to the output of the supercapacitor, and the other end is electrically connected to the power bus.

[0019] DC / AC converters, including: rotor motor-side DC / AC converters and hub motor-side DC / AC converters;

[0020] One end of the DC / AC converter on the rotor motor side is electrically connected to the power bus, and the other end is electrically connected to the rotor motor; one end of the DC / AC converter on the hub motor side is electrically connected to the power bus, and the other end is electrically connected to the hub motor.

[0021] Furthermore, the number of coaxial reversing units is eight, each coaxial reversing unit has a pair of counter-rotating propellers, and each propeller is connected to a rotor motor through a flight propulsion mode gearbox.

[0022] Furthermore, the number of wheels is four, and each wheel is connected to a hub motor via a ground driving mode gearbox.

[0023] In this invention, the three-source hybrid propulsion system, consisting of a hydrogen fuel cell, a lithium-ion battery, and a supercapacitor, allows each energy source to leverage its respective advantages under different operating conditions. The hydrogen fuel cell provides stable and continuous power under various operating conditions, ensuring the system's long-term efficient operation. The lithium-ion battery regulates power during stable flight conditions, balancing energy changes in the system. The supercapacitor provides power compensation during instantaneous high-power conditions, absorbing instantaneous energy, rapidly responding to the system's high-power demands, and effectively protecting the hydrogen fuel cell and lithium-ion battery. Energy management of the three energy sources can meet the multi-condition power requirements of the flying car and improve the system's dynamic response capability.

[0024] This invention also provides an energy management method for a hybrid propulsion system for flying cars, based on the above system, with the following steps:

[0025] Step 1): Establish a power coupling model for the hybrid propulsion system of the flying car to reflect the multi-source coupling characteristics among the hybrid propulsion systems of the flying car and characterize the relationship between the power demand of the flying car for air-to-ground operation and the output power of the energy source;

[0026] Step 2): Design the state variables, action space, reward function, and performance constraints of the improved SAC algorithm;

[0027] Step 3): Optimize and improve the hyperparameters of the SAC algorithm, solve the power coupling model of the hybrid propulsion system of the flying car, and conduct offline iterative training under representative working conditions so that it can generate actions and obtain corresponding rewards according to the input state, thereby achieving multi-objective optimization results.

[0028] Further, step 1) specifically includes: establishing a power balance model, a ground longitudinal dynamics model, an aerodynamic model, a hydrogen fuel cell model, a lithium-ion battery model, and a supercapacitor model for the flying car's hybrid propulsion system; as follows:

[0029] 11) Establish a power balance model;

[0030] The relationship between total power demand and the power of each component is as follows:

[0031] ;

[0032] In the formula, Total power demand; and These are the required power outputs for the rotor motor and the hub motor, respectively. , , These are the output powers of the hydrogen fuel cell stack, lithium-ion battery pack, and supercapacitor, respectively. and The operating efficiencies of the DC / DC converter and the DC / AC converter are respectively.

[0033] 12) Establish a longitudinal dynamic model of the ground;

[0034] Mechanical power requirements of a flying car traveling longitudinally on the ground The formula is:

[0035] ;

[0036] ;

[0037] In the formula, To improve the working efficiency of the wheel hub motor; The speed at which the flying car travels underground; For the mechanical efficiency of the transmission system, The full load mass of the flying car; It is the acceleration due to gravity; This refers to the wheel rolling resistance coefficient. The road slope angle; This refers to the air drag coefficient; The area of ​​the windward zone for the flying car to travel on the ground; air density; This is the rotational mass conversion factor; Acceleration for the flying car on the ground;

[0038] 13) Establish an aerodynamic model;

[0039] Mechanical power requirements of flying cars when driving in the air The formula is:

[0040] ;

[0041] ;

[0042] In the formula, The operating efficiency of the propeller rotor motor; and These are the induced power and the profile power, respectively, which can be obtained based on blade element theory and momentum theory. The power of the flying car during its vertical motion phase; The power of the flying car during its horizontal flight phase;

[0043] 14) Establish a hydrogen fuel cell model;

[0044] The relationship between the output power of a hydrogen fuel cell stack and its voltage and current is as follows:

[0045] ;

[0046] In the formula, For the output power of the hydrogen fuel cell stack, and These represent the output voltage and current of the hydrogen fuel cell stack, respectively. For the power of the auxiliary system;

[0047] 15) Establish a lithium-ion battery model;

[0048] The relationship between the output power, voltage, current, and state of charge of a lithium-ion battery pack during charge-discharge cycles is as follows:

[0049] ;

[0050] In the formula, The output power for lithium-ion battery charge-discharge cycles, and These are the output voltage and current of the lithium-ion battery, respectively. This refers to the state of charge (SOC) of a lithium-ion battery. and These are the initial and nominal battery capacities of the lithium-ion battery, respectively. For battery coulomb efficiency, Open circuit voltage, This refers to the battery's internal resistance. and These are the battery internal resistances under discharge and charging states, respectively.

[0051] 16) Establish a supercapacitor model;

[0052] The relationship between the output power, voltage, current, and state of charge of a supercapacitor is as follows:

[0053] ;

[0054] In the formula, This refers to the output power of the supercapacitor. and These are the output voltage and current of the supercapacitor, respectively. This refers to the state of charge of a supercapacitor. and These are the maximum safe operating voltage and the minimum safe operating voltage, respectively. For battery coulomb efficiency; Ideal voltage; This is the rated capacitor; This is the equivalent series resistance.

[0055] Furthermore, the formulas for calculating the various sub-powers of the mechanical power demand during aerial flight are as follows:

[0056] ;

[0057] In the formula, The radius of the propeller rotor is... The rotational speed of the propeller rotor disc. To induce rotational speed in the propeller, The propeller angular velocity, This is the cross-sectional area coefficient of the propeller rotor. The mean drag coefficient of the rotor airfoil for the flying car. For the rotor thrust ratio of flying cars, and These represent the vertical flight speed and acceleration of the flying car, respectively. and These represent the horizontal flight speed and acceleration of the flying car, respectively. and These represent the air drag coefficient and front end area of ​​the flying car in the vertical direction, respectively. and These represent the horizontal drag coefficient and the front end area of ​​the flying car, respectively. This is the rotational mass conversion factor for the flying car.

[0058] Furthermore, step 14) takes into account the aging effect of the hydrogen fuel cell stack, resulting in a decline in its health status. The calculation formula is:

[0059] ;

[0060] In the formula, This is a driving correction factor; , , and These are the low-load attenuation coefficient, high-load attenuation coefficient, variable-load attenuation coefficient, and start-stop attenuation coefficient, respectively. , , and These are respectively low-load operating time, high-load operating time, variable-load operating time, and complete start-stop time; This is the natural aging coefficient; This represents the total running time.

[0061] Furthermore, the formula for calculating the capacity loss caused by the aging effect of the lithium-ion battery in step 15) is as follows:

[0062] ;

[0063] In the formula, This represents the percentage of capacity loss in a lithium-ion battery. A coefficient used to measure the speed of battery charging and discharging; It is a forward factor; It is the ideal gas constant; Battery temperature; Activation energy is the energy required for the aging reaction to occur. For discharge ampere-hour throughput; It is a power-law factor.

[0064] Furthermore, step 2) specifically includes: designing the state variables, action space, reward function, and performance constraints of the improved SAC algorithm;

[0065] 21) Define the state space and action space;

[0066] The state variables of the improved SAC algorithm are defined as follows:

[0067] ;

[0068] In the formula, To improve the state variables of the SAC algorithm, This represents the percentage of remaining capacity in the lithium-ion battery pack. For the health status of hydrogen fuel cell stacks, This refers to the state of charge of a supercapacitor.

[0069] The action space of the improved SAC algorithm is defined as follows:

[0070] ;

[0071] In the formula, To improve the action variables of the SAC algorithm, This refers to the discharge power of the lithium-ion battery pack. This refers to the discharge power of the lithium-ion battery pack, where ;

[0072] 22) Define the reward function;

[0073] The reward function of the improved SAC algorithm is defined as:

[0074] ;

[0075] In the formula, , and These are the prices of hydrogen, the replacement price of hydrogen fuel cell stacks, and the replacement price of lithium-ion battery packs; , , and The weighting coefficients for relevant quantities reflect the relationship between monetary cost and battery. The relative importance of values; The reference state of charge for the lithium-ion battery pack, i.e., the initial state of charge. ; As a penalty for violating the constraint, a large negative constant is given when the constraint is violated;

[0076] 23) Define performance constraints;

[0077] The performance constraints of the improved SAC algorithm are:

[0078] ;

[0079] In the formula, and These are the minimum and maximum values ​​of the state of charge of a lithium-ion battery pack, respectively. and These are the minimum and maximum values ​​of the state of charge of a lithium-ion battery pack, respectively. and These are the minimum and maximum output power of the lithium-ion battery pack, respectively. and These represent the minimum and maximum output power of the hydrogen fuel cell stack, respectively. and These are the minimum and maximum output power of the supercapacitor, respectively. and These are the minimum and maximum values ​​of the bus voltage, respectively. and These are the minimum and maximum values ​​of the capacity loss ratio of lithium-ion battery packs, respectively. and These represent the minimum and maximum values ​​of the percentage decrease in the health status of the hydrogen fuel cell stack, respectively.

[0080] Furthermore, step 3) specifically includes:

[0081] 31) Initialize parameters;

[0082] 32) Collect environmental data;

[0083] 33) Critics' Network Update;

[0084] 34) The improved gray wolf algorithm is used to optimize the hyperparameters of the SAC algorithm, the log-normal distribution is used to improve the action generation strategy, the power output and charging and discharging strategies of each energy source are dynamically calculated and decomposed, and the actor network and temperature factor are updated.

[0085] 35) Cycles and convergence.

[0086] Further, step 31) specifically includes:

[0087] 311) Initialize the critic network and its parameters , Initialize the target critic network and its parameters , and order , Initialize actor network and its parameters Actor Network The output is the two latent normal distribution parameters of the log-normal distribution strategy. and Initialize temperature factor and experience replay pool ;

[0088] 312) Initialize the parameters of the improved gray wolf optimization algorithm;

[0089] An improved gray wolf optimization algorithm is used to adaptively optimize the key hyperparameters of the improved SAC algorithm, ensuring the convergence speed of training and the global optimization performance.

[0090] Further, step 312) specifically includes:

[0091] 3121) Determine the optimization objective hyperparameters;

[0092] The hyperparameters of the improved SAC algorithm are used as optimization variables for the improved Grey Wolf optimization algorithm; the set of target hyperparameters for optimization. Defined as:

[0093] ;

[0094] In the formula, For soft update step size, As a discount factor, Temperature factor For actors' online learning rate, For the critics' network learning rate;

[0095] 3122) Initialize the gray wolf population;

[0096] Define a by A population of individual gray wolves, where each gray wolf is a single individual gray wolf. Each represents a combination of hyperparameters; the position vector of each gray wolf. With sets The parameters in the table correspond one-to-one, specifically as follows:

[0097] ;

[0098] In the formula, Index for individual gray wolves, population size Set to 50 individuals;

[0099] 3123) Randomly generate the initial position;

[0100] Before starting the optimization iteration, an initial position is randomly generated for each individual gray wolf within a preset range of hyperparameter values. The initial random generation range for each parameter is as follows:

[0101] .

[0102] Further, step 32) specifically includes:

[0103] 321) Current state Below, actor network The action generation strategy is improved based on the log-normal distribution, and the latent normal distribution parameters are output. and ;

[0104] 322) Sample output actions from a log-normal distribution ;

[0105] 323) Change the current action The input is fed into the simulation environment, which calculates the instantaneous reward based on the reward function and the power coupling model of the flying car hybrid propulsion system. And the next state ;

[0106] 324) Store the experience tuple containing the current state, action, reward, and next state into the experience replay pool. .

[0107] Further, step 33) specifically includes:

[0108] 331) From the experience replay pool Randomly select multiple experience tuples ;

[0109] 332) Calculate the critic network according to the soft Bellman iteration formula. The values ​​are as follows:

[0110] ;

[0111] In the formula, The expected Q-value of the commentator network for the next state and the next action; , and These are the current state, action, and reward, respectively. and These are the action and reward for the next state, respectively. Discount factor; Temperature factor; For the actor network strategy function, by Parameterization;

[0112] 333) By minimizing the loss function To update the commentator network parameters ,as follows:

[0113] ;

[0114] In the formula, For experience replay pool, To replay experience pool A randomly selected empirical tuple. The expected value for updating the parameters of randomly selected experience tuples in the experience replay pool; For the target critic network;

[0115] 334) Use a soft update mechanism to update the target commentator network. parameters :

[0116] ;

[0117] In the formula, Use a soft update step size to smooth and stabilize the training process.

[0118] Furthermore, step 34) specifically includes:

[0119] 341) The improved Grey Wolf algorithm is used to optimize the hyperparameters of the improved SAC algorithm;

[0120] 342) Improve the action generation strategy by adopting a log-normal distribution;

[0121] The improved action generation formula based on the log-normal distribution is:

[0122] ;

[0123] In the formula, For the current action; The input noise vector; This is element-wise multiplication; Current state The mean of the potential normal distribution of the output action; Current state The standard deviation of the potential normal distribution of the output action; ;

[0124] 343) By minimizing the loss function To update the actor network parameters ,as follows:

[0125] ;

[0126] In the formula, For experience replay pool Random sampling status Expectations; According to the actor network In state Take action below The average value; This is the entropy term, i.e., the exploration reward term; For state Take action below The probability of;

[0127] 344) Minimize the loss function The optimization method is equivalent to through The gradient descent method for divergence is used to project information, and its formula is as follows:

[0128] ;

[0129] In the formula, for Divergence is an indicator that measures the degree of similarity between two probability distributions; Indicates that it was found To minimize the KL divergence between itself and the target distribution; Let be the probability distribution of the current policy of the actor network, and let represent the state. The probability of taking each possible action; The log-partition function is the normalized distribution. For the commentator network on the state Value assessment of all possible actions; To determine the sign of the exponent; Temperature factor;

[0130] 345) By minimizing the loss function Adjusting temperature factor The size, to balance exploration and exploitation, is as follows:

[0131] ;

[0132] In the formula, According to the actor network In state Take action below The average value, The target entropy value.

[0133] Further, step 341) specifically includes:

[0134] 3411) Design the fitness function;

[0135] Design a fitness function to minimize the total operating cost of the flying car; for each individual gray wolf... The fitness value, representing the hyperparameter combination, is calculated by running a complete reinforcement learning training cycle in the training environment; the fitness value is defined as:

[0136] ;

[0137] In the formula, Let be the fitness value of the gray wolf's position vector. The total number of steps in a single training cycle. To reinforce learning algorithms in the first The reward obtained per step; the negative sign transforms the goal of maximizing the cumulative reward into a problem of minimizing the fitness value;

[0138] 3412) Update and improve the Gray Wolf algorithm;

[0139] In each iteration cycle, the position is updated based on the fitness value of the gray wolves in the population, thereby gradually converging to the optimal hyperparameter combination;

[0140] Based on fitness values, the top three gray wolves in the current population are selected as leaders, namely... Wolf, wolves and Wolves, namely, gray wolves with the lowest fitness value, gray wolves with the second lowest fitness value, and gray wolves with the third lowest fitness value;

[0141] All other gray wolf individuals are updated based on the positions of these three leaders, using the following formula:

[0142] ;

[0143] In the formula, This is the position vector for the next generation of gray wolves. , and Contemporary Leaders Wolf, wolves and The wolf's position vector; The convergence factor is It is a random vector. It is a distance vector;

[0144] 3413) Iterative Looping and Output of Optimal Solution;

[0145] Repeat the fitness assessment and location update process described above until the preset termination condition is met. The wolf's position is used as the optimal hyperparameter of the improved SAC algorithm found by the improved gray wolf algorithm, and is used to guide subsequent reinforcement learning training.

[0146] Further, step 35) specifically includes: repeating steps 32) to 34), updating the parameters of the critic network and the actor network using gradient descent based on the calculated loss function, until the preset number of training iterations is reached or the convergence condition is met.

[0147] The beneficial effects of this invention are:

[0148] (1) This invention proposes a three-source hybrid propulsion system of hydrogen fuel cell-lithium-ion battery-supercapacitor. The supercapacitor can quickly compensate for energy during the instantaneous high-power flight phase, thereby achieving the effect of quickly responding to the instantaneous high-power demand of the system and effectively protecting the hydrogen fuel cell and lithium-ion battery.

[0149] (2) This invention proposes a log-normal distribution improved action generation strategy. By naturally limiting the action value range during the action generation process, it generates effective actions that meet the safety constraints of the three-source hybrid propulsion system, thereby achieving continuous output of effective actions and improving the reliability of the energy management method.

[0150] (3) This invention proposes an improved gray wolf algorithm, which improves the hyperparameter combination of the SAC algorithm through global search optimization, so as to quickly adapt to the significant differences between the energy sources of the three-source hybrid propulsion system and thus accelerate the convergence speed of the algorithm. Attached Figure Description

[0151] Figure 1 This is a schematic diagram of the hybrid propulsion system for flying cars in this invention.

[0152] Figure 2 This is a flowchart illustrating the energy management method of the present invention. Detailed Implementation

[0153] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to embodiments and accompanying drawings. The content mentioned in the embodiments is not intended to limit the present invention.

[0154] Reference Figure 1 As shown, a hybrid propulsion system for a flying car according to the present invention includes: a power output module, an energy source module, and a power distribution module;

[0155] The power output module includes: a flight propulsion unit and a ground drive unit;

[0156] The flight propulsion unit includes: a rotor motor, a flight propulsion mode gearbox, and a coaxial reversing unit. The input end of the rotor motor is electrically connected to the energy source module, and the output end is mechanically connected to the input end of the flight mode gearbox. The coaxial reversing unit is mechanically connected to the output end of the flight propulsion mode gearbox.

[0157] The ground drive unit includes: a hub motor, a ground driving mode gearbox, and wheels. The input end of the hub motor is electrically connected to the energy source module, and the output end is mechanically connected to the input end of the ground driving mode gearbox. The wheels are mechanically connected to the output end of the ground driving mode gearbox.

[0158] The energy source module includes: a hydrogen fuel cell stack, a hydrogen storage tank, a lithium-ion battery pack, and a supercapacitor;

[0159] The input end of the hydrogen fuel cell stack is connected to the hydrogen storage tank, and the output end is electrically connected to the energy source module, serving as the main energy source for the flying car.

[0160] The output of the lithium-ion battery pack is electrically connected to the energy source module and is connected in parallel with the supercapacitor group to form a composite power source, serving as the auxiliary energy source for the flying car.

[0161] The output of the supercapacitor is electrically connected to the energy source module;

[0162] The power distribution module includes: an energy management unit, a DC / AC converter, a DC / DC converter, and a power bus;

[0163] The energy management unit dynamically calculates and decomposes the power output and charging / discharging strategies of each energy source based on the total power demand for land and air travel, the real-time status of each energy source, and the current land and air travel mode.

[0164] DC / DC converters, including: fuel cell-side DC / DC converters, lithium-ion battery-side DC / DC converters, and supercapacitor-side DC / DC converters;

[0165] One end of the DC / DC converter on the fuel cell side is electrically connected to the output of the hydrogen fuel cell stack, and the other end is electrically connected to the power bus; one end of the DC / DC converter on the lithium-ion battery side is electrically connected to the output of the lithium-ion battery pack, and the other end is electrically connected to the power bus; one end of the DC / DC converter on the supercapacitor side is electrically connected to the output of the supercapacitor, and the other end is electrically connected to the power bus.

[0166] DC / AC converters, including: rotor motor-side DC / AC converters and hub motor-side DC / AC converters;

[0167] One end of the DC / AC converter on the rotor motor side is electrically connected to the power bus, and the other end is electrically connected to the rotor motor; one end of the DC / AC converter on the hub motor side is electrically connected to the power bus, and the other end is electrically connected to the hub motor.

[0168] Specifically, there are eight coaxial reversing units, each with a pair of counter-rotating propellers, and each propeller is connected to a rotor motor via a flight propulsion mode gearbox.

[0169] Specifically, the number of wheels is four, and each wheel is connected to a hub motor via a ground driving mode gearbox.

[0170] In this invention, the three-source hybrid propulsion system, consisting of a hydrogen fuel cell, a lithium-ion battery, and a supercapacitor, allows each energy source to leverage its respective advantages under different operating conditions. The hydrogen fuel cell provides stable and continuous power under various operating conditions, ensuring the system's long-term efficient operation. The lithium-ion battery regulates power during stable flight conditions, balancing energy changes in the system. The supercapacitor provides power compensation during instantaneous high-power conditions, absorbing instantaneous energy, rapidly responding to the system's high-power demands, and effectively protecting the hydrogen fuel cell and lithium-ion battery. Energy management of the three energy sources can meet the multi-condition power requirements of the flying car and improve the system's dynamic response capability.

[0171] Reference Figure 2 As shown, the present invention also provides an energy management method for a hybrid propulsion system for flying cars. Based on the above system, the steps are as follows:

[0172] Step 1): Establish a power coupling model for the hybrid propulsion system of the flying car to reflect the multi-source coupling characteristics among the systems and characterize the relationship between the power demand for air-to-ground operation and the output power of the energy sources. Specifically, this includes establishing a power balance model, a ground longitudinal dynamics model, an aerodynamic model, a hydrogen fuel cell model, a lithium-ion battery model, and a supercapacitor model for the hybrid propulsion system. See below:

[0173] 11) Establish a power balance model;

[0174] The relationship between total power demand and the power of each component is as follows:

[0175] ;

[0176] In the formula, Total power demand; and These are the required power outputs for the rotor motor and the hub motor, respectively. , , These are the output powers of the hydrogen fuel cell stack, lithium-ion battery pack, and supercapacitor, respectively. and The operating efficiencies of the DC / DC converter and the DC / AC converter are respectively.

[0177] 12) Establish a longitudinal dynamic model of the ground;

[0178] Mechanical power requirements of a flying car traveling longitudinally on the ground The formula is:

[0179] ;

[0180] ;

[0181] In the formula, To improve the working efficiency of the wheel hub motor; The speed at which the flying car travels underground; For the mechanical efficiency of the transmission system, The full load mass of the flying car; It is the acceleration due to gravity; This refers to the wheel rolling resistance coefficient. The road slope angle; This refers to the air drag coefficient; The area of ​​the windward zone for the flying car to travel on the ground; air density; This is the rotational mass conversion factor; Acceleration for the flying car on the ground;

[0182] 13) Establish an aerodynamic model;

[0183] Mechanical power requirements of flying cars when driving in the air The formula is:

[0184] ;

[0185] ;

[0186] In the formula, The operating efficiency of the propeller rotor motor; and These are the induced power and the profile power, respectively, which can be obtained based on blade element theory and momentum theory. The power of the flying car during its vertical motion phase; The power of the flying car during its horizontal flight phase;

[0187] 14) Establish a hydrogen fuel cell model;

[0188] The relationship between the output power of a hydrogen fuel cell stack and its voltage and current is as follows:

[0189] ;

[0190] In the formula, For the output power of the hydrogen fuel cell stack, and These represent the output voltage and current of the hydrogen fuel cell stack, respectively. For the power of the auxiliary system;

[0191] 15) Establish a lithium-ion battery model;

[0192] The relationship between the output power, voltage, current, and state of charge of a lithium-ion battery pack during charge-discharge cycles is as follows:

[0193] ;

[0194] In the formula, The output power for lithium-ion battery charge-discharge cycles, and These are the output voltage and current of the lithium-ion battery, respectively. This refers to the state of charge (SOC) of a lithium-ion battery. and These are the initial and nominal battery capacities of the lithium-ion battery, respectively. For battery coulomb efficiency, Open circuit voltage, This refers to the battery's internal resistance. and These are the battery internal resistances under discharge and charging states, respectively.

[0195] 16) Establish a supercapacitor model;

[0196] The relationship between the output power, voltage, current, and state of charge of a supercapacitor is as follows:

[0197] ;

[0198] In the formula, This refers to the output power of the supercapacitor. and These are the output voltage and current of the supercapacitor, respectively. This refers to the state of charge of a supercapacitor. and These are the maximum safe operating voltage and the minimum safe operating voltage, respectively. For battery coulomb efficiency; Ideal voltage; This is the rated capacitor; This is the equivalent series resistance.

[0199] The formulas for calculating the various sub-powers of the mechanical power demand during aerial flight are as follows:

[0200] ;

[0201] In the formula, The radius of the propeller rotor is... The rotational speed of the propeller rotor disc. To induce rotational speed in the propeller, The propeller angular velocity, This is the cross-sectional area coefficient of the propeller rotor. The mean drag coefficient of the rotor airfoil for the flying car. For the rotor thrust ratio of flying cars, and These represent the vertical flight speed and acceleration of the flying car, respectively. and These represent the horizontal flight speed and acceleration of the flying car, respectively. and These represent the air drag coefficient and front end area of ​​the flying car in the vertical direction, respectively. and These represent the horizontal drag coefficient and the front end area of ​​the flying car, respectively. This is the rotational mass conversion factor for the flying car.

[0202] In step 14), the aging effect of the hydrogen fuel cell stack and the decline in its health status are taken into account. The calculation formula is:

[0203] ;

[0204] In the formula, This is a driving correction factor; , , and These are the low-load attenuation coefficient, high-load attenuation coefficient, variable-load attenuation coefficient, and start-stop attenuation coefficient, respectively. , , and These are respectively low-load operating time, high-load operating time, variable-load operating time, and complete start-stop time; This is the natural aging coefficient; This represents the total running time.

[0205] The formula for calculating the capacity loss caused by the aging effect of lithium-ion batteries in step 15) is as follows:

[0206] ;

[0207] In the formula, This represents the percentage of capacity loss in a lithium-ion battery (initial capacity is 100%). A coefficient used to measure the speed of battery charging and discharging; It is a forward factor; It is the ideal gas constant; Battery temperature; Activation energy is the energy required for the aging reaction to occur. For discharge ampere-hour throughput; It is a power-law factor.

[0208] Step 2): Design the state variables, action space, reward function, and performance constraints of the improved SAC algorithm; specifically, this includes: designing the state variables, action space, reward function, and performance constraints of the improved SAC algorithm;

[0209] 21) Define the state space and action space;

[0210] The state variables of the improved SAC algorithm are defined as follows:

[0211] ;

[0212] In the formula, To improve the state variables of the SAC algorithm, This represents the percentage of remaining capacity in the lithium-ion battery pack. For the health status of hydrogen fuel cell stacks, This refers to the state of charge of a supercapacitor.

[0213] The action space of the improved SAC algorithm is defined as follows:

[0214] ;

[0215] In the formula, To improve the action variables of the SAC algorithm, This refers to the discharge power of the lithium-ion battery pack. This refers to the discharge power of the lithium-ion battery pack, where ;

[0216] 22) Define the reward function;

[0217] The reward function of the improved SAC algorithm is defined as:

[0218] ;

[0219] In the formula, , and These are the prices of hydrogen, the replacement price of hydrogen fuel cell stacks, and the replacement price of lithium-ion battery packs; , , and The weighting coefficients for relevant quantities reflect the relationship between monetary cost and battery. The relative importance of values; The reference state of charge for the lithium-ion battery pack, i.e., the initial state of charge. ; As a penalty for violating the constraint, a large negative constant is given when the constraint is violated;

[0220] 23) Define performance constraints;

[0221] The performance constraints of the improved SAC algorithm are:

[0222] ;

[0223] In the formula, and These are the minimum and maximum values ​​of the state of charge of a lithium-ion battery pack, respectively. and These are the minimum and maximum values ​​of the state of charge of a lithium-ion battery pack, respectively. and These are the minimum and maximum output power of the lithium-ion battery pack, respectively. and These represent the minimum and maximum output power of the hydrogen fuel cell stack, respectively. and These are the minimum and maximum output power of the supercapacitor, respectively. and These are the minimum and maximum values ​​of the bus voltage, respectively. and These are the minimum and maximum values ​​of the capacity loss ratio of lithium-ion battery packs, respectively. and These represent the minimum and maximum values ​​of the percentage decrease in the health status of the hydrogen fuel cell stack, respectively.

[0224] Step 3): Optimize and improve the hyperparameters of the SAC algorithm, solve the power coupling model of the flying car hybrid propulsion system, and conduct offline iterative training under representative operating conditions to enable it to generate actions and obtain corresponding rewards based on the input state, thus achieving multi-objective optimization results; specifically including:

[0225] 31) Initialize parameters;

[0226] 32) Collect environmental data;

[0227] 33) Critics' Network Update;

[0228] 34) The improved SAC algorithm hyperparameters are optimized by using the improved gray wolf (IGWO) algorithm, the action generation strategy is improved by using the log-normal distribution, the power output and charging and discharging strategies of each energy source are dynamically calculated and decomposed, and the actor network and temperature factor are updated.

[0229] 35) Cycles and convergence.

[0230] Specifically, step 31) includes:

[0231] 311) Initialize the critic network and its parameters , Initialize the target critic network and its parameters , and order , Initialize actor network and its parameters Actor Network The output is the two latent normal distribution parameters of the log-normal distribution strategy. and Initialize temperature factor and experience replay pool ;

[0232] 312) Initialize the parameters of the Improved Gray Wolf Optimization (IGWO) algorithm;

[0233] An improved gray wolf optimization algorithm is used to adaptively optimize the key hyperparameters of the improved SAC algorithm, ensuring the convergence speed of training and the global optimization performance.

[0234] Specifically, step 312) includes:

[0235] 3121) Determine the optimization objective hyperparameters;

[0236] The hyperparameters of the improved SAC algorithm are used as optimization variables for the improved Grey Wolf optimization algorithm; the set of target hyperparameters for optimization. Defined as:

[0237] ;

[0238] In the formula, For soft update step size, As a discount factor, Temperature factor For actors' online learning rate, For the critics' network learning rate;

[0239] 3122) Initialize the gray wolf population;

[0240] Define a by A population of individual gray wolves, where each gray wolf is a single individual gray wolf. Each represents a combination of hyperparameters; the position vector of each gray wolf. With sets The parameters in the table correspond one-to-one, specifically as follows:

[0241] ;

[0242] In the formula, Index for individual gray wolves, population size Set to 50 individuals;

[0243] 3123) Randomly generate the initial position;

[0244] Before starting the optimization iteration, an initial position is randomly generated for each individual gray wolf within a preset range of hyperparameter values ​​(this process ensures the diversity of the initial population, which is beneficial for the IGWO algorithm to perform a more comprehensive global search and avoid getting trapped in local optima). The initial random generation range of each parameter is as follows:

[0245] .

[0246] Specifically, step 32) includes:

[0247] Step 321) Current State Below, actor network The action generation strategy is improved based on the log-normal distribution, and the latent normal distribution parameters are output. and ;

[0248] Step 322) Sample the output action from the log-normal distribution ;

[0249] Step 323) Set the current action The input is fed into the simulation environment, which calculates the instantaneous reward based on the reward function and the power coupling model of the flying car hybrid propulsion system. And the next state ;

[0250] Step 324) Store the experience tuple containing the current state, action, reward, and next state into the experience replay pool. .

[0251] Specifically, step 33) includes:

[0252] 331) From the experience replay pool Randomly select multiple experience tuples ;

[0253] 332) Calculate the critic network according to the soft Bellman iteration formula. The values ​​are as follows:

[0254] ;

[0255] In the formula, The expected Q-value of the commentator network for the next state and the next action; , and These are the current state, action, and reward, respectively. and These are the action and reward for the next state, respectively. Discount factor; Temperature factor; For the actor network strategy function, by Parameterization;

[0256] 333) By minimizing the loss function To update the commentator network parameters ,as follows:

[0257] ;

[0258] In the formula, For experience replay pool, To replay experience pool A randomly selected empirical tuple. The expected value for updating the parameters of randomly selected experience tuples in the experience replay pool; For the target critic network;

[0259] 334) Use a soft update mechanism to update the target commentator network. parameters :

[0260] ;

[0261] In the formula, Use a soft update step size to smooth and stabilize the training process.

[0262] Specifically, step 34) includes:

[0263] 341) The improved Grey Wolf algorithm is used to optimize the hyperparameters of the improved SAC algorithm;

[0264] 342) Improve the action generation strategy by adopting a log-normal distribution;

[0265] The improved action generation formula based on the log-normal distribution is:

[0266] ;

[0267] In the formula, For the current action; The input noise vector; This is element-wise multiplication; Current state The mean of the potential normal distribution of the output action; Current state The standard deviation of the potential normal distribution of the output action; ;

[0268] 343) By minimizing the loss function To update the actor network parameters ,as follows:

[0269] ;

[0270] In the formula, For experience replay pool Random sampling status Expectations; According to the actor network In state Take action below The average value; This is the entropy term, i.e., the exploration reward term; For state Take action below The probability of;

[0271] 344) Minimize the loss function The optimization method is equivalent to through The gradient descent method for divergence is used to project information, and its formula is as follows:

[0272] ;

[0273] In the formula, for Divergence is an indicator that measures the degree of similarity between two probability distributions; Indicates that it was found To minimize the KL divergence between itself and the target distribution; Let be the probability distribution of the current policy of the actor network, and let represent the state. The probability of taking each possible action; The log-partition function is the normalized distribution. For the commentator network on the state Value assessment of all possible actions; To determine the sign of the exponent; Temperature factor;

[0274] 345) By minimizing the loss function Adjusting temperature factor The size, to balance exploration and exploitation, is as follows:

[0275] ;

[0276] In the formula, According to the actor network In state Take action below The average value, The target entropy value.

[0277] Specifically, step 341) includes:

[0278] 3411) Design the fitness function;

[0279] Design a fitness function to minimize the total operating cost of the flying car; for each individual gray wolf... The fitness value, representing the hyperparameter combination, is calculated by running a complete reinforcement learning training cycle in the training environment; the fitness value is defined as:

[0280] ;

[0281] In the formula, Let be the fitness value of the gray wolf's position vector. The total number of steps in a single training cycle. To reinforce learning algorithms in the first The reward obtained per step; the negative sign transforms the goal of maximizing the cumulative reward into a problem of minimizing the fitness value;

[0282] 3412) Update and improve the Gray Wolf algorithm;

[0283] In each iteration cycle, the position is updated based on the fitness value of the gray wolves in the population, thereby gradually converging to the optimal hyperparameter combination;

[0284] Based on fitness values, the top three gray wolves in the current population are selected as leaders, namely... Wolf, wolves and Wolves, namely, gray wolves with the lowest fitness value, gray wolves with the second lowest fitness value, and gray wolves with the third lowest fitness value;

[0285] All other gray wolf individuals are updated based on the positions of these three leaders, using the following formula:

[0286] ;

[0287] In the formula, This is the position vector for the next generation of gray wolves. , and Contemporary Leaders Wolf, wolves and The wolf's position vector; The convergence factor is It is a random vector. It is a distance vector;

[0288] 3413) Iterative Looping and Output of Optimal Solution;

[0289] Repeat the fitness assessment and location update process described above until the preset termination condition is met. The wolf's position (i.e., the hyperparameter combination with the minimum fitness value) is used as the optimal hyperparameter for the improved SAC algorithm found by the improved gray wolf algorithm, and is used to guide subsequent reinforcement learning training.

[0290] Specifically, step 35) includes: repeating steps 32) to 34) and updating the parameters of the critic network and actor network using gradient descent based on the calculated loss function until the preset number of training iterations is reached or the convergence condition is met.

[0291] This invention has many specific applications. The above description is only a preferred embodiment of this invention. It should be noted that for those skilled in the art, several improvements can be made without departing from the principle of this invention, and these improvements should also be considered within the scope of protection of this invention.

Claims

1. A flying car hybrid propulsion system, characterized in that, The application relates to a hybrid propulsion system of a flying car. The hybrid propulsion system comprises a power output module, an energy source module and a power distribution module. The power output module comprises a flight propulsion unit and a ground driving unit. The flight propulsion unit comprises a rotor motor, a flight propulsion mode gearbox and a coaxial reverse unit. The rotor motor is electrically connected to the energy source module at an input end and mechanically connected to an input end of the flight mode gearbox at an output end. The coaxial reverse unit is mechanically connected to an output end of the flight propulsion mode gearbox. The ground driving unit comprises a hub motor, a ground driving mode gearbox and a wheel. The hub motor is electrically connected to the energy source module at an input end and mechanically connected to an input end of the ground driving mode gearbox at an output end. The wheel is mechanically connected to an output end of the ground driving mode gearbox. The energy source module comprises a hydrogen fuel cell stack, a hydrogen storage tank, a lithium ion battery pack and a super capacitor. The hydrogen fuel cell stack is connected to the hydrogen storage tank at an input end and electrically connected to the energy source module at an output end, serving as a main energy source of the flying car. The lithium ion battery pack is electrically connected to the energy source module at an output end and connected to the super capacitor in parallel to form a composite power source, serving as a vice energy source of the flying car. The super capacitor is electrically connected to the energy source module at an output end. The power distribution module comprises an energy management unit, a DC / AC converter, a DC / DC converter and a power bus. The energy management unit dynamically calculates and decomposes the power output and charging / discharging strategy of each energy source according to the total power demand of land and air travel, the real-time state of each energy source and the current land and air travel mode.

2. A method of energy management for a flying car hybrid propulsion system, based on the system of claim 1, characterized in that, The DC / DC converter comprises a fuel cell side DC / DC converter, a lithium ion battery side DC / DC converter and a super capacitor side DC / DC converter. One end of the fuel cell side DC / DC converter is electrically connected to the output end of the hydrogen fuel cell stack, and the other end is electrically connected to the power bus. One end of the lithium ion battery side DC / DC converter is electrically connected to the output end of the lithium ion battery pack, and the other end is electrically connected to the power bus. One end of the super capacitor side DC / DC converter is electrically connected to the output end of the super capacitor, and the other end is electrically connected to the power bus. The DC / AC converter comprises a rotor motor side DC / AC converter and a hub motor side DC / AC converter. One end of the rotor motor side DC / AC converter is electrically connected to the power bus, and the other end is electrically connected to the rotor motor. One end of the hub motor side DC / AC converter is electrically connected to the power bus, and the other end is electrically connected to the hub motor. The method steps are as follows: Step 1): a power coupling model of the hybrid propulsion system of the flying car is established; Step 2): state variables, action space, reward function and performance constraints of the improved SAC algorithm are designed; Step 3): the super parameters of the improved SAC algorithm are optimized, the power coupling model of the hybrid propulsion system of the flying car is solved, offline iterative training is carried out under representative working conditions, actions are generated according to input states and corresponding rewards are obtained, and multi-objective optimization effects are obtained.

3. The energy management method of claim 2, wherein, The step 1) specifically comprises: establishing a power balance model, a ground longitudinal dynamics model, an aerodynamics model, a hydrogen fuel cell model, a lithium ion battery model and a super capacitor model of the flying car hybrid propulsion system; as follows: 11) establishing a power balance model; The relationship between the total demand power and the power of each component is: ； wherein, Ptot is the total demand power; Preq and Preq are the demand powers of the rotor electric machine and the hub electric machine, respectively; Preq and Preq are the demand powers of the rotor electric machine and the hub electric machine, respectively; Preq and Preq are the demand powers of the rotor electric machine and the hub electric machine, respectively; Preq and Preq are the demand powers of the rotor electric machine and the hub electric machine, respectively; Preq and Preq are the demand powers of the rotor electric machine and the hub electric machine, respectively; Preq and Preq are the demand powers of the rotor electric machine and the hub electric machine, respectively; Preq and Preq are the demand powers of the rotor electric machine and the hub electric machine, respectively; 12) establishing a ground longitudinal dynamics model; Mechanical power demand of a flying car when driving on the ground in longitudinal direction The formula is: ； ； wherein, is the operating efficiency of the wheel hub motor; is the ground travel speed of the flying car; is the mechanical efficiency of the transmission system, is the full load mass of the flying car; is the acceleration of gravity; is the rolling resistance coefficient of the wheel; is the road slope angle; is the air resistance coefficient; is the ground travel windward area of the flying car; is the air density; is the rotating mass conversion coefficient; is the ground travel acceleration of the flying car; 13) establishing an aerodynamics model; Mechanical power demand of a flying car while in flight The formula is: ； ； In the formula, is the propeller rotor motor operating efficiency; are the induced power and the profile power, respectively, which can be obtained according to the blade element theory and the momentum theory; is the power of the flying car in the vertical motion stage; is the power of the flying car in the horizontal flight motion stage;​ 14) establishing a hydrogen fuel cell model; The relationship between the output power of the hydrogen fuel cell stack and the voltage and current is: ； wherein P is the output power of the hydrogen fuel cell stack, and U and I are the output voltage and current, respectively, of the hydrogen fuel cell stack, Pauxis the power of the auxiliary system; 15) establishing a lithium ion battery model; The relationship between the output power, voltage, current and state of charge of the lithium ion battery pack in the charging and discharging cycle is: ； wherein Poutis the output power of the lithium-ion battery for charge and discharge cycles, and Voutand Ioutare the output voltage and current of the lithium-ion battery, respectively, SOCis the state of charge of the lithium-ion battery, and C0and Cnare the initial and nominal cell capacity of the lithium-ion battery, respectively, ηcellis the cell Coulombic efficiency, OCVis the open circuit voltage, Rcellis the cell internal resistance, and Rcell,disand Rcell,charg are the cell internal resistance in discharge and charge state, respectively; 16) establishing a super capacitor model; The relationship between the output power, voltage, current and state of charge of the super capacitor is: ； In the formula, This refers to the output power of the supercapacitor. and These are the output voltage and current of the supercapacitor, respectively. This refers to the state of charge of a supercapacitor. and These are the maximum safe operating voltage and the minimum safe operating voltage, respectively. For battery coulomb efficiency; Ideal voltage; This is the rated capacitor; This is the equivalent series resistance.

4. The energy management method of claim 3, wherein, The calculation formula of each sub-power of the mechanical demand power when flying is: ； wherein is the propeller rotor radius, is the propeller rotor disc rotational speed, is the propeller induced rotational speed, is the propeller angular velocity, is the propeller rotor cross-sectional area coefficient, is the flying car rotor airfoil average drag coefficient, is the flying car rotor thrust ratio, and are the flying car vertical flight speed and acceleration, respectively, and are the flying car horizontal flight speed and acceleration, respectively, and are the flying car vertical air resistance coefficient and frontal area, respectively, and are the flying car horizontal air resistance coefficient and frontal area, respectively, is the flying car rotational mass conversion coefficient.

5. The energy management method of claim 2, wherein, The step 2) specifically comprises: designing state variables, action space, reward function and performance constraints of the improved SAC algorithm. 21) defining state space and action space; The state variable of the improved SAC algorithm is defined as: ； wherein is the state of charge of the lithium-ion battery, is the state of health of the lithium-ion battery, is the state of health of the hydrogen fuel cell stack, is the state of charge of the supercapacitor; The action space of the improved SAC algorithm is defined as: ； wherein to improve the action variable of the SAC algorithm, for the discharge power of a lithium-ion battery, for the discharge power of a lithium-ion battery, wherein ; 22) defining a reward function; The reward function of the improved SAC algorithm is defined as: ； In the formula, , and These are the prices of hydrogen, the replacement price of hydrogen fuel cell stacks, and the replacement price of lithium-ion battery packs; , , and The weighting coefficients for relevant quantities reflect the relationship between monetary cost and battery. The relative importance of values; The reference state of charge for the lithium-ion battery pack, i.e., the initial state of charge. ; As a punishment for violating the rules; 23) defining performance constraints; The performance constraints of the improved SAC algorithm are: ； wherein, and are minimum and maximum values of the state of charge of the lithium-ion battery pack, respectively; and are minimum and maximum values of the state of charge of the lithium-ion battery pack, respectively; and are minimum and maximum values of the output power of the lithium-ion battery pack, respectively; and are minimum and maximum values of the output power of the hydrogen fuel cell stack, respectively; and are minimum and maximum values of the output power of the supercapacitor, respectively, and are minimum and maximum values of the bus voltage, respectively; and are minimum and maximum values of the capacity loss percentage of the lithium-ion battery pack, respectively; and are minimum and maximum values of the health state decrease percentage of the hydrogen fuel cell stack, respectively.

6. The energy management method of claim 2, wherein, The step 3) specifically comprises: 31) initializing parameters; 32) collecting environmental data; 33) critic network update; 34) using the improved grey wolf algorithm to optimize the super parameters of the improved SAC algorithm, using the lognormal distribution to improve the action generation strategy, dynamically calculating and decomposing the power output of each energy source and the charging and discharging strategy, updating the actor network and the temperature factor; 35) cycle and convergence.

7. The energy management method of claim 6, wherein, The step 31) specifically comprises: 311) initialize critic network and its parameters , , initialize target critic network and its parameters , and let , ; initialize actor network and its parameters , the output of the actor network are two latent normal distribution parameters and of a log-normal distribution policy; initialize temperature factor and experience replay pool ; 312) initializing the parameters of the improved grey wolf optimization algorithm; The improved grey wolf optimization algorithm is used to adaptively optimize the key super parameters of the improved SAC algorithm, ensuring the convergence speed and global optimization performance of the training.

8. The energy management method of claim 7, wherein, The step 32) specifically comprises: 321) current state Next, actor network Improving action generation policy output latent normal distribution parameters according to lognormal distribution and ; 322) sampling an output action from a lognormal distribution ; 323) the current action is input to the simulation environment, which calculates an immediate reward and a next state according to a reward function and a hybrid propulsion system power coupling model for the flying car, respectively; 324) storing an experience tuple comprising the current state, action, reward, and next state into an experience replay pool .

9. The energy management method of claim 8, wherein, The step 33) specifically comprises: 331) randomly sampling a plurality of experience tuples from an experience replay pool ;​ 332) Compute critic network according to soft Bellman iteration formula values as follows: ； wherein is the critic network Q-value for the next state and next action; , and are the current state, action and reward, respectively; and are the next state's action and reward, respectively; is the discount factor; is the temperature factor; is the actor network policy function; 333) by minimizing a loss function to update the critic network parameters as follows: ； wherein is an experience replay pool, is an experience tuple randomly sampled from the experience replay pool is an experience tuple randomly sampled from the experience replay pool, is the expected parameter update of randomly sampling an experience tuple from the experience replay pool; is a target critic network; 334) updating the target critic network using a soft update mechanism parameters of the target critic network : ； In the formula, is a soft update step size to smooth and stabilize the training process.

10. The energy management method for a flying car hybrid propulsion system of claim 9, wherein, The step 34) specifically comprises: 341) using the improved grey wolf algorithm to optimize the super parameters of the improved SAC algorithm; 342) using the lognormal distribution to improve the action generation strategy; The formula for improving the action generation of the lognormal distribution is: ； where is the current action; is the input noise vector; is the element-wise multiplication; is the current state is the mean of the output action latent normal distribution; is the current state is the standard deviation of the output action latent normal distribution; ; 343) by minimizing a loss function to update the parameters of the actor network as follows:​ ； In the formula, For experience replay pool Random sampling status Expectations; According to the actor network In state Take action below The average value; For entropy terms; For state Take action below The probability of; 344) Minimizing the loss function The optimization of the loss function is equivalent to performing an information projection by gradient descent on the divergence, which is given by ； where is divergence; denotes finding that minimizes the KL divergence between itself and the target distribution; is the probability distribution over the current policy of the actor network, denoting the state and the probability of taking each possible action; is the log partition function of the normalizing distribution; is the critic network's value estimate for the state and all possible actions; is the exponential symbol; is the temperature factor; 345) by minimizing a loss function Adjusting the temperature factor to balance exploration and exploitation, as follows: ； In the formula, According to the actor network In state Take action below The average value, The target entropy value.

Images