Fuel cell hybrid electric vehicle health-aware energy management method considering air conditioning control

Abstract

The application provides a fuel cell hybrid vehicle health-aware energy management method considering air conditioner control, which not only considers the important influence of the air conditioner system on the energy consumption and performance of the vehicle, but also considers that the fuel cell and the power battery will continuously age and decline during use, and the energy management needs to be adjusted accordingly. Therefore, the battery attenuation factor is also included in the multi-objective optimization process, thereby effectively ensuring the superiority of the energy management strategy. Based on the established multiple models and optimization objectives related to hydrogen consumption, cabin comfort, battery life and performance, the TD3 algorithm is used to obtain the optimal action variable, effectively solving the problem of Q value overestimation in the reinforcement learning process, and the training efficiency and optimization ability are obviously better than those of the prior art.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of fuel cell hybrid electric vehicle energy management, and particularly relates to a fuel cell hybrid electric vehicle health-aware energy management method considering air conditioner control. BACKGROUND

[0002] As a clean power source with zero emissions, fuel cells are currently more commonly used in hybrid electric vehicles in combination with power batteries, which can achieve high energy density and long vehicle driving range. However, how to manage the energy of the fuel cell hybrid power system to achieve comprehensive optimization of vehicle performance and energy consumption is crucial for such power systems. The existing energy management strategies for fuel cell hybrid electric vehicles can be roughly divided into rule-based strategies, optimization-based strategies, and learning-based strategies. However, the management effect of these strategies is severely limited by the conditions set and the modeling process, and most of them ignore the high-power auxiliary components such as air conditioning systems on the vehicle, making the energy management effect generally unable to meet expectations. Although the existing technology such as the Chinese patent with publication number CN115503559A considers the air conditioning system in energy management, it ignores the real-time health status of the battery and the continuous aging degradation problem in use, and cannot effectively alleviate aging using energy management strategies. At the same time, the Soft actor critic (SAC) algorithm used also has the defects of difficulty in parameter adjustment and overestimation of target Q value during training. Therefore, how to improve the energy management strategy of fuel cell hybrid electric vehicles, further reduce the comprehensive energy consumption of high-power auxiliary components such as air conditioning systems, and simultaneously improve vehicle driving performance and ride comfort is an urgent technical problem to be solved in the field. SUMMARY

[0003] Therefore, in view of the technical problems in the field, the present application provides a fuel cell hybrid electric vehicle health-aware energy management method considering air conditioner control. The method is based on the TwinDelayed Deep Deterministic policy gradient algorithm (TD3), which is a deterministic strategy that has been applied in some existing technologies for energy management of plug-in hybrid electric vehicles, but rarely involves fuel cell hybrid electric vehicles. Compared with the SAC algorithm mentioned above, the TD3 algorithm has higher accuracy of Q function.

[0004] The method specifically includes the following steps:

[0005] Step one, obtain the vehicle state information, air conditioning system state information, power battery state information and fuel cell state information of the fuel cell hybrid vehicle; wherein the vehicle state information includes: vehicle speed v, acceleration acc, driving motor speed ω m , driving motor torque T m and cabin internal temperature T in ; the air conditioning system state information is specifically air conditioning system refrigeration / heating capacity Q ac ; the power battery state information includes: power battery voltage, power battery current, power battery SOC and SOH; the fuel cell state information includes: fuel cell output power P fc , fuel cell efficiency η fc and power change rate ΔP fc ;

[0006] Step two, according to the vehicle dynamics, the longitudinal dynamics model of the vehicle is established; for the fuel cell hybrid power system topology, the whole vehicle hydrogen mass consumption model, the air conditioning system model reflecting the air conditioning refrigeration / heating performance and demand, the cabin heat load model reflecting the temperature environment in the cabin and its each related physical quantity, the fuel cell aging model, the power battery life decline model and the driving motor model reflecting the performance of the driving motor are established in turn;

[0007] Step three, for TD3 algorithm, vehicle speed v, acceleration acc, power battery SOC, power battery SOH, cabin internal temperature T in and fuel cell output power P fc are selected as state variables of the algorithm, and constitute a state space S:

[0008] S=[v,acc,SOC,SOH,T in ,P fc ]

[0009] The fuel cell power change rate ΔP fc and the air conditioning system refrigeration / heating capacity Q ac are selected as action variables, and constitute an action space a:

[0010] a=[ΔP fc ,Q ac ]

[0011] Five optimization objectives including hydrogen consumption cost, fuel cell recession cost, power battery recession cost, power battery SOC maintenance and cabin internal temperature change are set, and the corresponding reward function r is constructed:

[0012]

[0013] Wherein, τ, ψ, γ represent the price of hydrogen per kg, the purchase price of power battery, and the purchase price of fuel cell, respectively; is the hydrogen mass; ΔSOH is the change of power battery state of health; V fc is the fuel cell voltage degradation rate; ω1 and ω2 are weight factors for adjusting the power battery SOC maintenance and the temperature change inside the vehicle cabin, respectively; SOC tar is the target SOC to be maintained; T tar is the target temperature set inside the vehicle cabin; t is the time variable;

[0014] Step four, after initializing the TD3 algorithm, the TD3 algorithm is trained using the corresponding training set constructed by pre-acquiring the state information of each vehicle, so that the trained algorithm can obtain the action variable that maximizes the reward function value according to the real-time state variable.

[0015] Further, the specific form of the automobile longitudinal dynamics model established in step two is:

[0016]

[0017] Wherein, F t is the traction force, m is the mass of the vehicle, g is the acceleration of gravity, f is the rolling resistance coefficient, α is the road slope angle, ρ is the air density, A is the windward area, C D is the air resistance coefficient, v is the vehicle speed, and δ is the vehicle rotational mass conversion coefficient;

[0018] The specific form of the hydrogen mass consumption model of the vehicle is:

[0019]

[0020] Wherein, is the instantaneous hydrogen consumption of the fuel cell system, is the chemical energy density of hydrogen; P fc is the output power of the fuel cell;

[0021] The specific form of the air conditioning system model is:

[0022]

[0023] Wherein, P ac is the demand power of the air conditioning system, Q ac is the refrigeration / heat quantity of the air conditioning system, and COP is the refrigeration / heat performance coefficient of the air conditioning system;

[0024] The specific form of the vehicle cabin heat load model is:

[0025] Q = Q s + Q c + Qp +Q a +Q o

[0026]

[0027]

[0028] Q p =Q d +116ζn

[0029] Q a = m a Cp air (T out -T in )

[0030]

[0031] wherein Q is the total heat load of the vehicle cabin, Q s is the solar radiation heat load, Q c is the conduction heat load, Q p is the passenger metabolic heat load, Q a is the air exchange heat load, Q o is the heat load generated by the electrical auxiliary equipment of the vehicle, η is the conduction coefficient of solar radiation through the vehicle window, ρ w is the average heat absorption rate of the vehicle window, a in and a out are the convective heat transfer coefficients inside and outside the vehicle cabin, P sum is the radiant intensity per unit area, C is the sunshade correction coefficient of the vehicle window glass, S is the area of the vehicle window, θ is the angle between the normal of the vehicle window surface and the position of the sun in the sky, K is the heat transfer coefficient of each part of the vehicle body, A body is the area of each part of the vehicle body, ρ c is the average heat absorption rate of the heat exchange surface, T out and T in are the external environment temperature and the internal temperature of the vehicle cabin, Q d is the heat generated by the driver, ζ is the heat dissipation cluster correction coefficient of the random age and gender of the passengers, n is the number of passengers, m a is the amount of air entering the vehicle cabin from outside, Cp air is the specific heat capacity of air, ρ air is the air density in the vehicle cabin, V air is the air volume in the vehicle cabin; t is the time variable;

[0032] The specific form of the fuel cell aging model is:

[0033] V fc = N fcv1 + t1v2 + ΔP fc v3 + t2v4

[0034] wherein V fc is the degradation voltage of the fuel cell per unit time, i.e., the voltage degradation rate, N fc is the number of start-stop cycles per unit time of the fuel cell system, v1 is the voltage degradation rate of the fuel cell under one start-stop cycle, t1 is the running time of the fuel cell under low power load (less than or equal to 20% of the rated power) per unit time, t2 is the running time of the fuel cell under high power load (greater than or equal to 80% of the rated power) per unit time, v2 is the voltage degradation rate of the fuel cell under low power load, v3 is the voltage degradation rate of the fuel cell under high power load, and v4 is the voltage degradation rate per kilowatt of the fuel cell under load change;

[0035] The specific form of the power battery life degradation model is:

[0036]

[0037]

[0038]

[0039] wherein P bat is the output power of the power battery, V ocv is the open-circuit voltage of the power battery, I bat is the current of the power battery, R0 is the internal resistance of the power battery, SOC0 is the initial value of SOC, SOH0 is the initial value of SOH, N EOL is the total cycle number of the power battery, c is the charge-discharge rate of the power battery, Q bat is the capacity of the power battery, and t is the time variable.

[0040] The specific form of the driving motor model is:

[0041]

[0042] wherein P motor is the power of the driving motor, T m is the torque of the driving motor, ω m is the speed of the driving motor, and η m is the efficiency of the driving motor.

[0043] Further, the TD3 algorithm includes an action network μ and two critic networks Q1 and Q2, wherein the action network selects a determined action a according to the input state s, and the target action is added with noise for relieving overfitting caused by estimation bias, i.e.,

[0044]

[0045] where ξ is the added noise, μ' is the target action network, subscript t+1 denotes the t+1 state, θ μ′ is the target action network parameter; the symbol ← indicates that the item on the left is determined by the item on the right;

[0046] In the learning process of the TD3 algorithm, the target Q value is calculated by selecting the smaller value from the outputs of the two target critic networks, that is:

[0047]

[0048]

[0049] where y t (t) is the target Q value under a specific state s t and action a t , r is the instantaneous reward, φ is the discount factor, Q' i is the target critic network, and are the parameters of the critic network and the target critic network respectively, is the learning rate, is the loss function , and the gradient of the target critic network with respect to the critic network is represented by ∇θ J(θ μ ); the above steps can solve the problem of overestimation of the critic network Q value;

[0050] To improve the quality and stability of algorithm updates, the action network update frequency is set to be lower than the critic network, and the action network is updated using the gradient of the target function, and the update mechanism is as follows:

[0051]

[0052] where β is the learning rate, is the gradient of the target function J(θ μ ) with respect to θ μ .

[0053] The fuel cell hybrid vehicle health-aware energy management method considering air conditioner control provided by the application not only considers the important influence of the air conditioner system on the vehicle energy consumption and performance, but also takes the battery attenuation factor into the multi-objective optimization process, so as to effectively ensure the accuracy of the energy management strategy. Based on the established multiple models and optimization objectives related to hydrogen consumption, cabin comfort, battery life and performance, the TD3 algorithm is trained to obtain the optimal action variable, effectively solving the problem of Q value overestimation in the reinforcement learning process, and the training efficiency and optimization ability are obviously better than those of the prior art. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 The flowchart of the method provided by the application is shown in the figure;

[0055] Figure 2 The fuel cell hybrid power system topology diagram is shown in the figure;

[0056] Figure 3 The schematic diagram of the operation principle of the TD3 algorithm used in the application is shown in the figure. DETAILED DESCRIPTION

[0057] The technical solutions of the application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the application.

[0058] The fuel cell hybrid vehicle health-aware energy management method considering air conditioner control provided by the application specifically includes the following steps:

[0059] Step 1: Obtain the vehicle state information, air conditioner system state information, power battery state information and fuel cell state information of the fuel cell hybrid vehicle; wherein the vehicle state information includes vehicle speed v, acceleration acc, driving motor speed ω m , driving motor torque T m and cabin internal temperature T in ; the air conditioner system state information is specifically the air conditioner system refrigeration / heating capacity Q ac ; the power battery state information includes power battery voltage, power battery current, power battery SOC and SOH; the fuel cell state information includes fuel cell output power P fc , fuel cell efficiency η fc and power change rate ΔP fc ;

[0060] Step two, according to the vehicle dynamics, the longitudinal dynamics model of the vehicle is established; for the fuel cell hybrid power system topology, the hydrogen mass consumption model of the whole vehicle is established, the air conditioning system model reflecting the air conditioning refrigeration / heat performance and demand, the cabin heat load model reflecting the temperature environment in the cabin and its related physical quantities, the fuel cell aging model, the power battery life decline model and the driving motor model reflecting the performance of the driving motor are established in turn;

[0061] Step three, for the TD3 algorithm, the vehicle speed v, the acceleration acc, the power battery SOC, the power battery SOH, the cabin internal temperature T in and the fuel cell output power P fc are selected as the state variables of the algorithm, and the state space S is composed:

[0062] S=[v,acc,SOC,SOH,T in ,P fc ]

[0063] The fuel cell power change rate ΔP fc and the air conditioning system refrigeration / heat quantity Q ac are selected as the action variables, and the action space a is composed:

[0064] a=[ΔP fc ,Q ac ]

[0065] Five optimization objectives including hydrogen consumption cost, fuel cell degradation cost, power battery degradation cost, power battery SOC maintenance and cabin temperature change are set, and the corresponding reward function r is constructed:

[0066]

[0067] Wherein, τ, ψ, γ respectively represent the price of each kilogram of hydrogen, the purchase price of power battery, the purchase price of fuel cell; is the hydrogen mass; ΔSOH is the change of power battery health state; V fc is the fuel cell voltage degradation rate; ω1 and ω2 are weight factors for adjusting power battery SOC maintenance and cabin internal temperature change respectively; SOC tar is the target SOC to be maintained; T tar is the target temperature set in the cabin; t is the time variable;

[0068] Step four, after initializing the TD3 algorithm, the TD3 algorithm is trained by using the corresponding training set constructed by pre-acquiring each vehicle state information, so that the trained algorithm can obtain the action variable of the maximum reward function value according to the real-time state variable.

[0069] The execution process of the above method is as follows:Figure 1 As shown, it can be widely applied to fuel cell hybrid power systems or other various topological forms as shown. Figure 2 As shown, it can be widely applied to fuel cell hybrid power systems or other various topological forms as shown.

[0070] In a preferred embodiment of the present application, the specific form of the vehicle longitudinal dynamics model established in step two is:

[0071]

[0072] Wherein, F t is the traction force, m is the mass of the vehicle, g is the acceleration of gravity, f is the rolling resistance coefficient, a is the road slope angle, p is the air density, A is the windward area, C D is the air resistance coefficient, v is the vehicle speed, and d is the vehicle rotational mass conversion coefficient.

[0073] The specific form of the vehicle hydrogen mass consumption model is:

[0074]

[0075] Wherein, is the instantaneous hydrogen consumption of the fuel cell system, is the chemical energy density of hydrogen; P fc is the output power of the fuel cell;

[0076] The specific form of the air conditioning system model is:

[0077]

[0078] Wherein, P ac is the demand power of the air conditioning system, Q ac is the refrigeration / heating capacity of the air conditioning system, and COP is the refrigeration / heating performance coefficient of the air conditioning system.

[0079] The specific form of the vehicle cabin heat load model is:

[0080] Q = Q s + Q c + Q p + Q a + Q o

[0081]

[0082]

[0083] Q p = Q d + 116ζn

[0084] Q a = m aCp air (T out -T in )

[0085]

[0086] wherein Q is the total thermal load of the vehicle cabin, Q s is the solar radiation thermal load, Q c is the conduction thermal load, Q p is the passenger metabolic thermal load, Q a is the air exchange thermal load, Q o is the thermal load generated by the electrical auxiliary equipment of the vehicle, η is the conduction coefficient of solar radiation through the vehicle window, ρ w is the average heat absorption rate of the vehicle window, a in and a out are the convective heat transfer coefficients inside and outside the vehicle cabin, P sum is the radiation intensity per unit area, C is the sunshade correction coefficient of the vehicle window glass, S is the area of the vehicle window, θ is the angle between the normal of the vehicle window surface and the position of the sun in the sky, K is the heat transfer coefficient of each part of the vehicle body, A body is the area of each part of the vehicle body, ρ c is the average heat absorption rate of the heat exchange surface, T out and T in are the external environment temperature and the internal temperature of the vehicle cabin, Q d is the heat generated by the driver, ζ is the heat dissipation cluster correction coefficient of the random age and gender of the passengers, n is the number of passengers, m a is the amount of air entering the vehicle from outside, Cp air is the specific heat capacity of air, ρ air is the air density in the vehicle cabin, V air is the air volume in the vehicle cabin; t is the time variable;

[0087] The specific form of the fuel cell aging model is:

[0088] V fc = N fc v1+t1v2++ΔP fc v3+t2v4

[0089] wherein V fc is the decay voltage of the fuel cell per unit time, i.e., the voltage decay rate, N fcis the number of start-stop cycles per unit time of the fuel cell system, v1 is the voltage attenuation rate of the fuel cell in one start-stop cycle, t1 is the running time of the fuel cell under low power load (less than or equal to 20% of the rated power) per unit time, t2 is the running time of the fuel cell under high power load (greater than or equal to 80% of the rated power) per unit time, v2 is the voltage attenuation rate of the fuel cell under low power load, v3 is the voltage attenuation rate of the fuel cell under high power load, and v4 is the voltage attenuation rate per kilowatt of the fuel cell during load change;

[0090] The specific form of the power battery life degradation model is:

[0091]

[0092]

[0093]

[0094] wherein P bat is the output power of the power battery, V ocv is the open-circuit voltage of the power battery, I bat is the current of the power battery, R0 is the internal resistance of the power battery, SOC0 is the initial value of SOC, SOH0 is the initial value of SOH, N EOL is the total cycle number of the power battery, c is the charge-discharge rate of the power battery, Q bat is the capacity of the power battery; and t is a time variable.

[0095] The specific form of the driving motor model is:

[0096]

[0097] wherein P motor is the power of the driving motor, T m is the torque of the driving motor, ω m is the speed of the driving motor, and η m is the efficiency of the driving motor.

[0098] In a preferred embodiment of the present application, the TD3 algorithm is as shown in Figure 3 , specifically comprising an action network μ and two critic networks Q1 and Q2, wherein the action network selects a determined action a according to the input state s, and noise is added in the target action to alleviate overfitting caused by estimation bias, that is:

[0099]

[0100] wherein, ξ is the added noise, μ' is the target action network, subscript t+1 represents the t+1 state, θ μ′ is the target action network parameter; the symbol <- represents that the item on the left is determined by the item on the right;

[0101] In the learning process of the TD3 algorithm, the target Q value is calculated by selecting the smaller value from the outputs of the two target critic networks, that is:

[0102]

[0103]

[0104] wherein, y t (t) is the target Q value under a specific state s t and action a t , r is the instantaneous reward, φ is the discount factor, Q' i is the target critic network, and are the parameters of the critic network and the target critic network respectively, is the learning rate, is the loss function , and the gradient of the target critic network with respect to the critic network is represented by ; the above steps can solve the problem of overestimation of the critic network Q value;

[0105] In order to improve the quality and stability of the algorithm update, the action network update frequency is set to be lower than the critic network, and the action network is updated by using the gradient of the target function to update the parameters, and the update mechanism is as follows:

[0106]

[0107] wherein, β is the learning rate, is the gradient of the target function J(θ μ ) with respect to θ μ .

[0108] It should be understood that the size of the serial number of each step in the embodiments of the present application does not mean the order of execution, and the execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0109] Although the embodiments of the present application have been shown and described, it can be understood by those of ordinary skill in the art that various changes, modifications, replacements and variations can be made to these embodiments without departing from the principles and spirits of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A fuel cell hybrid electric vehicle health-aware energy management method considering air conditioning control, characterized in that: Specifically comprising the following steps: Step one, obtaining vehicle state information, air conditioning system state information, power battery state information and fuel cell state information of the fuel cell hybrid vehicle; wherein the vehicle state information includes: vehicle speed v, acceleration acc, driving motor speed ω m , driving motor torque T m and cabin internal temperature T in ; the air conditioning system state information is specifically air conditioning system refrigeration / heating capacity Q ac ; the power battery state information includes: power battery voltage, power battery current, power battery SOC and SOH; the fuel cell state information includes: fuel cell output power P fc , fuel cell efficiency η fc and power change rate ΔP fc ; Step two, according to the vehicle dynamics to establish a vehicle longitudinal dynamics model; For fuel cell hybrid power system topology in turn to establish the whole vehicle hydrogen mass consumption model, air conditioning system model reflecting the performance and demand of air conditioning refrigeration / heat, vehicle cabin thermal load model reflecting the temperature environment and each related physical quantity in the vehicle cabin, fuel cell aging model, power battery life decline model and driving motor model reflecting the performance of driving motor; Step three, select the vehicle speed v, acceleration acc, power battery SOC, power battery SOH, vehicle cabin temperature T in and fuel cell output power P fc as the state variables of the algorithm, and form the state space S: S = [v, acc, SOC, SOH, T in , P fc ] Selecting the fuel cell power change rate ΔP fc With the air conditioning system refrigeration / heating capacity Q ac As the action variable, and constitutes the action space a: a = [ΔP fc , Q ac ] Set five optimization objectives including hydrogen consumption cost, fuel cell degradation cost, power battery degradation cost, power battery SOC maintenance and vehicle cabin temperature change, and construct the corresponding reward function r: Wherein, τ, ψ, γ respectively represent the price of hydrogen per kg, the purchase price of power battery, the purchase price of fuel cell; is the hydrogen quality; ΔSOH is the power battery health state change amount; V fc is the fuel cell voltage degradation rate; ω1 and ω2 are weight factors for adjusting the power battery SOC maintenance and the temperature change inside the vehicle cabin; SOC tar is the target SOC to be maintained; T tar is the target temperature set inside the vehicle cabin; t is the time variable; Step four, after initializing the TD3 algorithm, using the pre-acquired vehicle state information and the corresponding training set constructed to train the TD3 algorithm, so that the trained algorithm can obtain the action variable that maximizes the reward function value according to the real-time state variable.

2. The method of claim 1, wherein: The specific form of the vehicle longitudinal dynamics model established in step two is: Among them, F t Let m be the traction force, g be the vehicle mass, f be the acceleration due to gravity, α be the rolling resistance coefficient, ρ be the road surface slope angle, A be the air density, and C be the frontal area. D δ is the air resistance coefficient, v is the vehicle speed, and δ is the vehicle rotational mass conversion factor. The specific form of the whole vehicle hydrogen mass consumption model is: wherein, is the instantaneous hydrogen consumption of the fuel cell system, is the chemical energy density of hydrogen; η fc is the fuel cell efficiency, P fc is the fuel cell output power; The specific form of the air conditioning system model is: where P ac is the required power of the air conditioning system, Q ac is the refrigeration / heating capacity of the air conditioning system, and COP is the refrigeration / heating performance coefficient of the air conditioning system. The specific form of the vehicle cabin thermal load model is: Q = Q s +Q c +Q p +Q a +Q o Q p = Q d + 116ζn Q a = m a Cp air (T out -T in ) where Q is the total heat load of the vehicle cabin, Q s is the solar radiation heat load, Q c is the conduction heat load, Q p is the passenger metabolic heat load, Q a is the air exchange heat load, Q o is the heat load generated by the electrical auxiliary equipment of the vehicle, η is the conduction coefficient of solar radiation through the vehicle window, ρ w is the average heat absorption rate of the vehicle window, a in and a out are the convective heat transfer coefficients inside and outside the vehicle cabin, P sum is the radiant intensity per unit area, C is the sunshade correction coefficient of the vehicle window glass, S is the area of the vehicle window, θ is the angle between the normal of the vehicle window surface and the position of the sun in the sky, K is the heat transfer coefficient of each part of the vehicle body, A body is the area of each part of the vehicle body, ρ c is the average heat absorption rate of the heat exchange surface, T out and T in are the external environment temperature and the internal temperature of the vehicle cabin, respectively, Q d is the heat generated by the driver, ζ is the heat dissipation cluster correction coefficient of the random age and gender of the passengers, n is the number of passengers, m a is the amount of air entering the vehicle from outside, Cp air is the specific heat capacity of air, ρ air is the air density in the vehicle cabin, V air is the air volume in the vehicle cabin; t is the time variable; The specific form of the fuel cell aging model is: V fc = N fc v1+t1v2++ΔP fc v3+t2v4 wherein V fc is the voltage degradation rate of the fuel cell per unit time, N fc is the number of start-stop cycles per unit time, v1 is the voltage degradation rate of the fuel cell in one start-stop cycle, t1 is the running time of the fuel cell under low power load per unit time, t2 is the running time of the fuel cell under high power load per unit time, v2 is the voltage degradation rate of the fuel cell under low power load, v3 is the voltage degradation rate of the fuel cell under high power load, and v4 is the voltage degradation rate per kilowatt of the fuel cell during load change. The specific form of the power battery life decline model is: wherein P bat is the power battery output power, V ocv is the power battery open circuit voltage, I bat is the power battery current, R0 is the power battery internal resistance, SOC0 is the initial value of SOC, SOH0 is the initial value of SOH, N EOL is the total cycle number of the power battery, c is the power battery charge and discharge rate, Q bat is the power battery capacity; t is the time variable; The specific form of the driving motor model is: where P motor is the drive motor power, T m is the drive motor torque ω m is the drive motor speed, and η m is the drive motor efficiency.

3. The method of claim 1, wherein: The TD3 algorithm includes an action network μ and two critic networks Q1 and Q2, where the action network selects a determined action according to an input state, and the target action Noise is added in the target action for mitigating overfitting caused by the estimation bias, that is: wherein ξ is the added noise, μ' is the target action network, the subscript t+1 denotes the t+1th state, θ μ′ is the target action network parameter; the symbol denotes that the item on the left is determined by the item on the right; In the learning process of TD3 algorithm, the target Q value is calculated by selecting the smaller value from the output of the two target critic networks, that is: where y t (t) is the target Q value in a specific state s t and action a t , r is the instantaneous reward, φ is the discount factor, Q' i is the target critic network, and are the parameters of the critic network and the target critic network, respectively, is the learning rate, is the loss function , and its gradient with respect to ; the problem of overestimation of the critic network Q value can be solved by the above steps; In order to improve the quality and stability of algorithm update, the action network update frequency is set lower than the critic network, specifically the action network is updated by using the gradient of the target function to update the parameters, and the update mechanism is as follows: where β is a learning rate, is the gradient of the objective function J(θ μ ) with respect to θ μ .

Images