CN116278992A - Fuel cell automobile energy management method integrating information physical system - Google Patents

Abstract

The invention provides a fuel cell automobile energy management method integrating an information physical system, which not only considers the energy flow and consumption in the automobile system, but also considers the influence of future road topography and traffic conditions on the automobile energy management, and explores the optimal control in a feasible domain by combining a depth deterministic strategy gradient algorithm, thereby effectively avoiding discrete errors and improving the reliability of strategies. The invention realizes the information interaction between the vehicle system and the network layer through the information physical system, and brings future topographic information, traffic information, battery aging, fuel cell durability constraint, hydrogen consumption and the like obtained through the information physical system into the control frame, thereby having important practical significance for achieving the optimal balance of the system durability and the hydrogen consumption of the real vehicle.

Description

Fuel cell automobile energy management method integrating information physical system

Technical Field

The invention belongs to the technical field of energy management of fuel cell hybrid power systems, and particularly relates to an energy management method of a fuel cell automobile integrating an information physical system.

Background

At present, proton membrane exchange fuel cells are increasingly used by new energy automobiles, especially hybrid electric vehicles, due to the advantages of cleanliness, high energy efficiency and the like. However, the dynamic response of the fuel cell is slow, and the speed and power change during the running process of the vehicle are rapid and drastic, which makes the energy management of the fuel cell hybrid power system formed by matching with the power cell more difficult. In the energy management strategy for the fuel cell hybrid electric vehicle, the problems that the actual performance transition depends on the set conditions and the modeling process exist; although the power distribution of the vehicle power system itself or other high-power components on the vehicle has been comprehensively considered in many prior arts such as chinese patent publication No. CN113085665a, environmental factors outside the vehicle such as road topography, traffic conditions, etc. are not considered, and these factors tend to have a stronger influence on the energy consumption during the running of the vehicle according to practical experience, so that the existing energy management strategies for the fuel cell hybrid power system are not perfect, and still have a great room for improvement.

Disclosure of Invention

In view of the above, the present invention provides a fuel cell vehicle energy management method integrating information physical systems, which specifically includes the following steps:

step one, acquiring vehicle state information, power battery state information and fuel battery state information of a fuel battery hybrid electric vehicle; wherein the vehicle state information includes: vehicle speed v, acceleration acc, driving motor rotation speed omega _motor Torque T of driving motor _motor Efficiency eta of driving motor _motor The method comprises the steps of carrying out a first treatment on the surface of the The power battery state information includes: power battery voltage and current, internal resistance and SOC;the fuel cell state information includes: fuel cell output power P _FC Efficiency eta _FC Rate of change of power Δp _FC ；

Step two, establishing a longitudinal dynamics model of the automobile according to the dynamics of the automobile; sequentially establishing a fuel cell hydrogen consumption model, a power cell equivalent circuit model, a power cell life attenuation model and a driving motor model aiming at a topological structure of a fuel cell hybrid power system;

step three, acquiring real-time driving state information comprising the speed v and the acceleration acc from CAN signals of the vehicle, and acquiring geographic position data of the vehicle through a GPS module; uploading the driving state information and the geographic position data to a cloud server by utilizing a vehicle-mounted network, wherein the cloud server acquires the gradient, curvature and traffic information of a future road which the vehicle is about to pass through based on the information and feeds back the gradient, curvature and traffic information to the vehicle;

step four, selecting a vehicle speed v, an acceleration acc, a power battery SOC, a power battery SOH and a future road gradient i according to a depth deterministic strategy gradient (Deep Deterministic Policy Gradient, DDPG) algorithm _f Future road curvature c _f Future road traffic information t _f As state variables, and constitute a state space S:

S＝[v,acc,SOC,SOH,i _f ,c _f ,t _f ]

selecting a fuel cell power change rate DeltaP _FC As an action variable, and constitutes an action space a:

a＝[ΔP _FC |ΔP _FC ∈[-3,+3]kW]

four optimization targets including overall vehicle hydrogen consumption, power battery life, power battery SOC maintenance and fuel battery power limit are set, and corresponding reward functions r are constructed:

wherein p is ₁ For the unit price of hydrogen per kilogram,

for hydrogen mass, p ₂ For power battery replacement price, α and β are weighting coefficients for power battery SOC maintenance and fuel battery power variation limit, respectively, SOC _tar Δp, target value for battery SOC maintenance _FCmax Maximum value for fuel cell power conversion limit;

initializing the DDPG algorithm, constructing a training set by utilizing historical data or vehicle state information corresponding to standard working conditions, and training the algorithm, so that the trained algorithm can obtain optimal action variables according to real-time state variables.

Further, the specific form of the automobile longitudinal dynamics model established in the second step is as follows:

η _t ＝η _DC/AC ·η _EM ·η _tra

P _tol ＝P _FC ·η _DC/DC +P _bat

wherein P is _tol For the total power required for the vehicle to travel, eta _t Is the efficiency of the vehicle, m is the weight of the vehicle, g is the gravitational acceleration, f is the rolling resistance coefficient, α is the road gradient, A is the frontal area, C _D Is the air resistance coefficient, v is the vehicle speed, delta is the conversion coefficient of the vehicle rotating mass, eta _DC/AC 、η _EM 、η _tra 、η _DC/DC Efficiency of DC/AC converter, drive motor, drive train and DC/DC converter, respectively, P _FC 、P _bat The output power of the fuel cell and the power cell respectively;

the specific form of the fuel cell hydrogen consumption model is as follows:

wherein,,

for the instantaneous hydrogen consumption of the fuel cell system, +.>

For the heating value of hydrogen, < >>

The theoretical power generated for the consumed hydrogen, t is a time variable;

the power battery equivalent circuit model is specifically formed by:

wherein V is _ocv For the open-circuit voltage of the power battery, I _bat Is the current of the power battery, R ₀ Is the internal resistance of the power battery, Q _bat Is the power battery capacity;

the specific form of the power battery life attenuation model is as follows:

wherein Q is _loss For power battery capacity loss, c is power battery discharge multiplying power, B (c) is compensation factor, E _a (c) For activation energy, R is an ideal gas constant, T is the absolute temperature of the power battery, A (c) is the ampere-hour throughput of the power battery, and N (c) is the equivalent charge and discharge quantity of the power battery;

the driving motor model is specifically formed by:

η _motor ＝f(ω _motor ,T _motor )

when the rotating speed omega of the motor _motor And torque T _motor After the determination, the efficiency eta of the driving motor can be obtained _motor 。

Further, the DDPG algorithm specifically comprises a actor network mu, a commentator network Q and an experience pool; the said interview network outputs the comprehensive score Q (s, a) for action-rewards based on state variable s and action variable a; the actor network can maximize Q (s, a) output by the critics network through training;

the experience pool is used for forming and storing a state variable s, a motion variable a, a rewarding value r and a next state variable s' corresponding to a certain state into experience samples, and when the number of the experience samples in the experience pool exceeds the storable number of the experience pool, old data can be covered; the algorithm training is specifically carried out by using small batches of samples randomly extracted from an experience pool;

the actor network updates by performing gradient descent of the form of the objective function corresponding to the optimization objective:

J(θ ^μ )＝E[Q(s,μ(s))]

wherein J (θ) ^μ ) As an objective function, θ ^μ As a function of the network parameters of the actor,

representing the gradient, E (·) is the mathematical expectation, η is the learning rate of the actor network; symbol ≡ represents the item to the left of which the item to the right is determined;

the actor network and the commentator network respectively have corresponding parameters theta ^μ′ Target actor network mu' of (1) and with parameter theta ^Q′ Target critics network Q'; the target actor network outputs a corresponding action variable a ' based on a next state variable s ', and s ' and a ' are input to the target critic network together to output Q ' (s ', a '); the evaluator network is used for minimizing the TD error between the current Q value and the time sequence differential target thereof, and the specific form is as follows:

y _target (t)＝r(s,a)+γQ′(s',a'|θ ^Q′ )

δ(t)＝y _target (t)-Q(s,a|θ ^Q )

wherein y is _target (t) is a time sequence differential target, and delta (t) is a TD error;

the update of the commentator network is likewise effected using the gradient descent method of the following form:

wherein, beta is the learning rate of the criticizer network;

the target actor network and the target criticizer network adopt a delay updating mode, and only after the actor network and the criticizer network are updated for a preset number of times, the target actor network and the target criticizer network are updated, and corresponding network parameters are updated in a soft mode by the following steps:

θ ^μ′ ←τθ ^μ +(1-τ)θ ^μ′

θ ^Q′ ←τθ ^Q +(1-τ)θ ^Q′

where τ is the soft update factor.

According to the fuel cell automobile energy management method integrating the information physical system, provided by the invention, not only are the energy flow and consumption in the automobile system considered, but also the influence of future road topography and traffic conditions on the automobile energy management is considered, and the optimal control in the feasible domain is explored by combining the depth deterministic strategy gradient algorithm, so that the discrete error is effectively avoided, and the reliability of the strategy is improved. According to the invention, the information interaction between the vehicle system and the network layer is realized through the information physical system, and future topographic information obtained through the information physical system, battery aging, fuel cell durability constraint, hydrogen consumption and the like are brought into the control frame, so that the method has important practical significance for achieving the optimal balance of the system durability and the hydrogen consumption of the real vehicle.

Drawings

FIG. 1 is a flow chart of a method provided by the present invention;

FIG. 2 is an alternative topology of a fuel cell hybrid power system to which the present invention is applicable;

fig. 3 is a schematic block diagram of the DDPG algorithm.

Detailed Description

The following description of the embodiments of the present invention will be made apparent and fully in view of the accompanying drawings, in which some, but not all embodiments of the invention are shown. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

The fuel cell automobile energy management method integrating the information physical system provided by the invention, as shown in figure 1, specifically comprises the following steps:

step one, acquiring vehicle state information, power battery state information and fuel battery state information of a fuel battery hybrid electric vehicle; wherein the vehicle state information includes: vehicle speed v, acceleration acc, driving motor rotation speed omega _motor Driving motorTorque T _motor Efficiency eta of driving motor _motor The method comprises the steps of carrying out a first treatment on the surface of the The power battery state information includes: power battery voltage and current, internal resistance and SOC; the fuel cell state information includes: fuel cell output power P _FC Efficiency eta _FC Rate of change of power Δp _FC ；

Step two, establishing a longitudinal dynamics model of the automobile according to the dynamics of the automobile; sequentially establishing a fuel cell hydrogen consumption model, a power cell equivalent circuit model, a power cell life attenuation model and a driving motor model aiming at a topological structure of a fuel cell hybrid power system;

step three, acquiring real-time driving state information comprising the speed v and the acceleration acc from CAN signals of the vehicle, and acquiring geographic position data of the vehicle through a GPS module; uploading the driving state information and the geographic position data to a cloud server by utilizing a vehicle-mounted network, wherein the cloud server acquires the gradient, curvature and traffic information of a future road which the vehicle is about to pass through based on the information and feeds back the gradient, curvature and traffic information to the vehicle;

step four, selecting a vehicle speed v, an acceleration acc, a power battery SOC, a power battery SOH and a future road gradient i according to a depth deterministic strategy gradient (Deep Deterministic Policy Gradient, DDPG) algorithm _f Future road curvature c _f Future road traffic information t _f As state variables, and constitute a state space S:

S＝[v,acc,SOC,SOH,i _f ,c _f ,t _f ]

selecting a fuel cell power change rate DeltaP _FC As an action variable, and constitutes an action space a:

a＝[ΔP _FC |ΔP _FC ∈[-3,+3]kW]

four optimization targets including overall vehicle hydrogen consumption, power battery life, power battery SOC maintenance and fuel battery power limit are set, and corresponding reward functions r are constructed:

wherein p is ₁ For the unit price of hydrogen per kilogram,

for hydrogen mass, p ₂ For power battery replacement price, α and β are weighting coefficients for power battery SOC maintenance and fuel battery power variation limit, respectively, SOC _tar Δp, target value for battery SOC maintenance _FCmax Maximum value for fuel cell power conversion limit;

initializing the DDPG algorithm, constructing a training set by utilizing historical data or vehicle state information corresponding to standard working conditions, and training the algorithm, so that the trained algorithm can obtain optimal action variables according to real-time state variables.

Fig. 2 shows an alternative topology of a fuel cell hybrid system to which the method provided by the invention can be applied.

In a preferred embodiment of the present invention, the longitudinal dynamics model of the automobile established in the second step is in the specific form of:

η _t ＝η _DC/AC ·η _EM ·η _tra

P _tol ＝P _FC ·η _DC/DC +P _bat

wherein P is _tol For the total power required for the vehicle to travel, eta _t Is the efficiency of the vehicle, m is the weight of the vehicle, g is the gravitational acceleration, f is the rolling resistance coefficient, α is the road gradient, A is the frontal area, C _D Is the air resistance coefficient, v is the vehicle speed, delta is the conversion coefficient of the vehicle rotating mass, eta _DC/AC 、η _EM 、η _tra 、η _DC/DC Efficiency of DC/AC converter, drive motor, drive train and DC/DC converter, respectively, P _FC 、P _bat The output power of the fuel cell and the power cell respectively;

the specific form of the fuel cell hydrogen consumption model is as follows:

wherein,,

for the instantaneous hydrogen consumption of the fuel cell system, +.>

For the heating value of hydrogen, < >>

The theoretical power generated for the consumed hydrogen, t is a time variable;

the power battery equivalent circuit model is specifically formed by:

wherein V is _ocv For the open-circuit voltage of the power battery, I _bat Is the current of the power battery, R ₀ Is the internal resistance of the power battery, Q _bat Is the power battery capacity;

the specific form of the power battery life attenuation model is as follows:

wherein Q is _loss For power battery capacity loss, c is power battery discharge multiplying power, B (c) is compensation factor, E _a (c) For activation energy, R is an ideal gas constant, T is the absolute temperature of the power battery, A (c) is the ampere-hour throughput of the power battery, and N (c) is the equivalent charge and discharge quantity of the power battery;

the driving motor model is specifically formed by:

η _motor ＝f(ω _motor ,T _motor )

when the rotating speed omega of the motor _motor And torque T _motor After the determination, the efficiency eta of the driving motor can be obtained _motor 。

In a preferred embodiment of the present invention, as shown in fig. 3, the DDPG algorithm specifically includes a actor network μ, a reviewer network Q, and an experience pool; the said interview network outputs the comprehensive score Q (s, a) for action-rewards based on state variable s and action variable a; the actor network can maximize Q (s, a) output by the critics network through training;

the experience pool is used for forming and storing a state variable s, a motion variable a, a rewarding value r and a next state variable s' corresponding to a certain state into experience samples, and when the number of the experience samples in the experience pool exceeds the storable number of the experience pool, old data can be covered; the algorithm training is specifically carried out by using small batches of samples randomly extracted from an experience pool;

the actor network updates by performing gradient descent of the form of the objective function corresponding to the optimization objective:

J(θ ^μ )＝E[Q(s,μ(s))]

wherein J (θ) ^μ ) As an objective function, θ ^μ As a function of the network parameters of the actor,

representing the gradient, E (·) is the mathematical expectation, η is the learning rate of the actor network; symbol ≡ represents the item to the left of which the item to the right is determined;

the actor network and the commentator network respectively have corresponding parameters theta ^μ′ Target actor network mu' of (1) and with parameter theta ^Q′ Target critics network Q'; the target actor network outputs a corresponding action variable a ' based on a next state variable s ', and s ' and a ' are input to the target critic network together to output Q ' (s ', a '); the evaluator network is used for minimizing the TD error between the current Q value and the time sequence differential target thereof, and the specific form is as follows:

y _target (t)＝r(s,a)+γQ′(s',a'|θ ^Q′ )

δ(t)＝y _target (t)-Q(s,a|θ ^Q )

wherein y is _target (t) is a time sequence differential target, and delta (t) is a TD error;

the update of the commentator network is likewise effected using the gradient descent method of the following form:

wherein, beta is the learning rate of the criticizer network;

the target actor network and the target criticizer network adopt a delay updating mode, and only after the actor network and the criticizer network are updated for a preset number of times, the target actor network and the target criticizer network are updated, and corresponding network parameters are updated in a soft mode by the following steps:

θ ^μ′ ←τθ ^μ +(1-τ)θ ^μ′

θ ^Q′ ←τθ ^Q +(1-τ)θ ^Q′

where τ is the soft update factor.

It should be understood that, the sequence number of each step in the embodiment of the present invention does not mean that the execution sequence of each process should be determined by the function and the internal logic of each process, and should not limit the implementation process of the embodiment of the present invention.

Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.

Claims (3)

1. A fuel cell automobile energy management method integrating an information physical system is characterized in that: the method specifically comprises the following steps:

step one, acquiring vehicle state information, power battery state information and fuel battery state information of a fuel battery hybrid electric vehicle; wherein the vehicle state information includes: vehicle speed v, acceleration acc, driving motor rotation speed omega _motor Torque T of driving motor _motor Efficiency eta of driving motor _motor The method comprises the steps of carrying out a first treatment on the surface of the The power battery state information includes: power battery voltage and current, internal resistance and SOC; the fuel cell state information includes: fuel cell output power P _FC Efficiency eta _FC Rate of change of power Δp _FC ；

Step two, establishing a longitudinal dynamics model of the automobile according to the dynamics of the automobile; sequentially establishing a fuel cell hydrogen consumption model, a power cell equivalent circuit model, a power cell life attenuation model and a driving motor model aiming at a topological structure of a fuel cell hybrid power system;

step three, acquiring real-time driving state information comprising the speed v and the acceleration acc from CAN signals of the vehicle, and acquiring geographic position data of the vehicle through a GPS module; uploading the driving state information and the geographic position data to a cloud server by utilizing a vehicle-mounted network, wherein the cloud server acquires the gradient, curvature and traffic information of a future road which the vehicle is about to pass through based on the information and feeds back the gradient, curvature and traffic information to the vehicle;

step four, selecting a vehicle speed v, an acceleration acc, a power battery SOC, a power battery SOH and a future road gradient i according to a DDPG algorithm _f Future road curvature c _f Future road traffic information t _f As state variables, and constitute a state space S:

S＝[v,acc,SOC,SOH,i _f ,c _f ,t _f ]

selecting a fuel cell power change rate DeltaP _FC As an action variable, and constitutes an action space a:

a＝[ΔP _FC |ΔP _FC ∈[-3,+3]kW]

four optimization targets including overall vehicle hydrogen consumption, power battery life, power battery SOC maintenance and fuel battery power limit are set, and corresponding reward functions r are constructed:

r＝p ₁ ·[m _H2 (t)]+p ₂ ·Q _bat ·ΔSOH+α·[SOC(t)-SOC _tar ] ² +β·|ΔP _FC /ΔP _FCmax |

wherein p is ₁ For unit price per kilogram of hydrogen, m _H2 For hydrogen mass, p ₂ For power battery replacement price, α and β are weighting coefficients for power battery SOC maintenance and fuel battery power variation limit, respectively, SOC _tar Δp, target value for battery SOC maintenance _FCmax Maximum value for fuel cell power conversion limit;

initializing the DDPG algorithm, constructing a training set by utilizing historical data or vehicle state information corresponding to standard working conditions, and training the algorithm, so that the trained algorithm can obtain optimal action variables according to real-time state variables.

2. The method of claim 1, wherein: the specific form of the automobile longitudinal dynamics model established in the second step is as follows:

η _t ＝η _DC/AC ·η _EM ·η _tra

P _tol ＝P _FC ·η _DC/DC +P _bat

wherein P is _tol For the total power required for the vehicle to travel, eta _t Is the efficiency of the vehicle, m is the weight of the vehicle, g is the gravitational acceleration, f is the rolling resistance coefficient, α is the road gradient, A is the frontal area, C _D Is the air resistance coefficient, v is the vehicle speed, delta is the conversion coefficient of the vehicle rotating mass, eta _DC/AC 、η _EM 、η _tra 、η _DC/DC Efficiency of DC/AC converter, drive motor, drive train and DC/DC converter, respectively, P _FC 、P _bat The output power of the fuel cell and the power cell respectively;

the specific form of the fuel cell hydrogen consumption model is as follows:

wherein,,

for the instantaneous hydrogen consumption of the fuel cell system, +.>

For the heating value of hydrogen, < >>

The theoretical power generated for the consumed hydrogen, t is a time variable;

the power battery equivalent circuit model is specifically formed by:

wherein V is _ocv For the open-circuit voltage of the power battery, I _bat Is the current of the power battery, R ₀ Is the internal resistance of the power battery, Q _bat Is the power battery capacity;

the specific form of the power battery life attenuation model is as follows:

wherein Q is _loss For power battery capacity loss, c is power battery discharge multiplying power, B (c) is compensation factor, E _a (c) For activation energy, R is ideal gas constant, T is absolute temperature of power battery, A (c) is ampere-hour throughput of power battery, and N (c) is power batteryEquivalent charge and discharge quantity;

the driving motor model is specifically formed by:

η _motor ＝f(ω _motor ,T _motor )

when the rotating speed omega of the motor _motor And torque T _motor After the determination, the efficiency eta of the driving motor can be obtained _motor 。

3. The method of claim 1, wherein: the DDPG algorithm specifically comprises a actor network mu, a criticism network Q and an experience pool; the said interview network outputs the comprehensive score Q (s, a) for action-rewards based on state variable s and action variable a; the actor network can maximize Q (s, a) output by the critics network through training;

the experience pool is used for forming and storing a state variable s, a motion variable a, a rewarding value r and a next state variable s' corresponding to a certain state into experience samples, and when the number of the experience samples in the experience pool exceeds the storable number of the experience pool, old data can be covered; the algorithm training is specifically carried out by using small batches of samples randomly extracted from an experience pool;

the actor network updates by performing gradient descent of the form of the objective function corresponding to the optimization objective:

J(θ ^μ )＝E[Q(s,μ(s))]

wherein J (θ) ^μ ) As an objective function, θ ^μ The method is characterized in that the method is an actor network parameter, wherein, the actor network parameter is V represents a gradient, E (·) is a mathematical expectation, and eta is a learning rate of the actor network; symbol ≡ represents the item to the left of which the item to the right is determined;

the actor network and the commentator network respectively have corresponding parameters theta ^μ′ Target actor network mu' of (1) and with parameter theta ^Q′ Target critics network Q'; the target actor network outputs a corresponding action variable a ' based on a next state variable s ', and s ' and a ' are input to the target critic network together to output Q ' (s ', a '); the evaluator network is used for minimizing the TD error between the current Q value and the time sequence differential target thereof, and the specific form is as follows:

y _target (t)＝r(s,a)+γQ′(s',a'|θ ^Q′ )

δ(t)＝y _target (t)-Q(s,a|θ ^Q )

wherein y is _target (t) is a time sequence differential target, and delta (t) is a TD error;

the update of the commentator network is likewise effected using the gradient descent method of the following form:

wherein, beta is the learning rate of the criticizer network;

the target actor network and the target criticizer network adopt a delay updating mode, and only after the actor network and the criticizer network are updated for a preset number of times, the target actor network and the target criticizer network are updated, and corresponding network parameters are updated in a soft mode by the following steps:

θ ^μ′ ←τθ ^μ +(1-τ)θ ^μ′

θ ^Q′ ←τθ ^Q +(1-τ)θ ^Q′

where τ is the soft update factor.

Images