Power system for light-hydrogen-electric vehicle and control method

Abstract

The invention discloses a power system for a light-hydrogen-electric vehicle and a control method, and relates to the field of hybrid power systems for light-hydrogen-electric vehicles and energy management. According to the invention, a fuel cell, a power cell and a photovoltaic cell are integrated to form a vehicle multi-energy power system, three energy optimal output proportion calculation methods are given, and an optimal control strategy is explored by using a depth deterministic strategy gradient algorithm. According to the algorithm, the influence of external variables such as road conditions, illumination intensity and traffic flow on energy management is fully considered, so that control errors are reduced, and the strategy reliability is improved. The network information layer interacts with a vehicle system in real time and comprehensively analyzes road conditions, illumination, traffic flow, hydrogen consumption, bus stability, power battery durability and other multi-dimensional factors, so that the optimal balance between hydrogen consumption and energy stability is realized.

Description

Technical Field

[0001] The present invention relates to the field of hybrid power systems for solar-hydrogen-electric vehicles and energy management, and in particular to a power system for solar-hydrogen-electric vehicles and a control method. Background Art

[0002] As the global demand for renewable and clean energy continues to grow, hybrid electric vehicles (HEVs) have attracted widespread attention as a means of transportation with high energy efficiency and low tail gas emissions. Photovoltaic cells and fuel cells, as two important renewable energy technologies, are increasingly used in hybrid electric vehicles. Photovoltaic cells provide a clean power source for vehicles by directly converting solar energy into electrical energy. The application of photovoltaic cells can reduce dependence on traditional fuels, reduce carbon emissions, and thus mitigate the impact of climate change. However, the output of photovoltaic cells is affected by factors such as weather conditions, sunshine intensity, and panel orientation, and its energy output is unstable, so an effective energy management strategy is needed to maximize its energy. At the same time, fuel cells, as an efficient and clean energy technology, also play an important role in hybrid electric vehicles. Fuel cells use the chemical reaction of hydrogen and oxygen to generate electricity, which is emission-free, pollution-free, and has extremely high energy density and environmental friendliness. However, there are still challenges in the stability of fuel cells, hydrogen storage and distribution, etc., which require comprehensive consideration of vehicle system design and energy management strategies. As the core component of hybrid electric vehicles, power batteries are responsible for storing and releasing electrical energy, which is crucial to achieving efficient energy management of vehicles. The energy management strategy of power batteries involves charging, discharging, energy-saving mode switching, etc., and needs to comprehensively consider factors such as the vehicle's power requirements, battery health status, and the source of electrical energy to achieve optimal energy utilization efficiency and vehicle performance.

[0003] To this end, the present invention proposes a power system and control method for a solar-hydrogen-electric vehicle. Summary of the invention

[0004] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a power system and control method for a solar-hydrogen-electric vehicle.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A power system for a light-hydrogen-electric vehicle includes a fuel cell, a power battery, a photovoltaic cell, a drive motor, a DC / DC converter and a DC / AC converter, wherein the DC / DC converter is a non-isolated, multi-phase interleaved variable boost DC / DC converter, the photovoltaic panel of the photovoltaic cell is arranged in the roof area, and the fuel cell and the power battery are placed at the rear of the roof.

[0007] Preferably: the rated output power of the fuel cell should account for 62% to 68% of the rated power of the vehicle motor, and the rated output power of the power battery should account for 32% to 38% of the rated power of the vehicle motor. When working at night, the total rated output power of the fuel cell and the power battery is 100%.

[0008] Preferably: the vehicle longitudinal dynamics model of the solar-hydrogen-electric vehicle power system is f is the rolling resistance coefficient, M is the curb mass, g is the acceleration of gravity, a is the slope, which is generally set to 0 in energy management, ρ is the air density, A represents the frontal area of ​​the vehicle, C is the air resistance coefficient, v is the vehicle speed, δ is the mass coefficient, which is set to 1, and a is the vehicle acceleration.

[0009] Preferably: the model of the fuel cell is E represents the open circuit voltage of the fuel cell, V fc and i fc Represents output voltage and output current, R ohm represents internal resistance, i0 represents exchange current, a, c2, c3, i max are all empirical constants.

[0010] Preferably: the power battery equivalent circuit model is: τ is the time constant, τ = C pol .R pol The battery state of charge (SOC) is an important parameter to characterize the battery state. Its instantaneous change value is expressed as current and battery capacity Q cap ratio.

[0011] Preferably: the model of the drive motor is η m =f(w m ,T m ), When the motor speed w m and torque T m After determining, the efficiency η of the drive motor can be obtained m , n m is the motor speed.

[0012] Preferably: the model of the photovoltaic cell is I sc is the short-circuit current, V oc is the open circuit voltage, I m is the maximum power point current, V m is the maximum power point voltage.

[0013] Preferably: the logic of the DC / DC converter is to define the vehicle demand current change rate α, the vehicle reference current change rate αref , and according to the value of α, the working mode of the fuel cell multi-phase interleaved variable boost DC / DC converter is converted. The conversion formula is: P max , P min is the maximum and minimum value of the load power; V arg is the average voltage under normal operating conditions; f(state) is the operating condition influence coefficient, which realizes the reference current change rate α under different operating conditions ref make corrections;

[0014] The conversion conditions of the DC / DC converter are as follows:

[0015] ①: When α<0, the switches S7 and S8 are disconnected, which is equivalent to an interleaved parallel DC / DC converter;

[0016] ②: When 0<α<α ref When , the switch S8 is turned off, which is equivalent to a three-phase interleaved boost DC / DC converter;

[0017] ③: When α>α ref When the switch S8 is turned off, it is equivalent to a four-phase interleaved boost DC / DC converter; and considering the influence of the actual power of the system, the power factor k is introduced pf (0.9~1.0) to determine the final parameters of the fuel cell. After introducing the power coefficient, the calculation formula for the final parameters of the fuel cell is as follows: pf,fuel =k pf ·P fuel , P total =P pf,fuel +P bat +P pv .

[0018] A control method for a solar-hydrogen-electric vehicle power system comprises the following steps:

[0019] S1: The vehicle's sensor system measures the vehicle's physical signals, which include vehicle status information, power battery information, fuel cell information, and photovoltaic system information. The vehicle status information includes: vehicle speed v, acceleration a, and vehicle position. The power battery information includes: power battery remaining power SOC, power battery charge and discharge efficiency ba, power battery voltage and current, and internal resistance. The fuel cell information includes: fuel cell output power P FC , efficiency η fc , and hydrogen consumption rate △H2, photovoltaic system information includes: light intensity PV, weather temperature T, photovoltaic cell output power P pv ;

[0020] S2: The photovoltaic system uses the maximum power point tracking algorithm MPPT to control the photovoltaic cells to ensure that the output power of the photovoltaic cells is maximized;

[0021] S3: Extract vehicle speed v, acceleration a, hydrogen consumption rate △H2, and motor required power P from the CAN signal in the vehicle as state variables and feed them back to the deep deterministic policy gradient algorithm network as observation values; obtain the vehicle's geographic location in real time through the GPS module and upload it to the cloud server, and after processing by the cloud server, feed back the road signals around the vehicle including road length, road slope, road curvature, road traffic information, real-time light intensity in the area where the vehicle is located, temperature, and future changes;

[0022] S4: Upload vehicle-related data to the cloud server. When the delay time of the feedback data is greater than the threshold T, use a built-in RBF neural network to predict the input data required by DDPG;

[0023] S5: Collect data from the running of the array vehicles, including road length, road slope, road curvature, road traffic information, future changes in light intensity and temperature in the area where the vehicle is located, select 70% of the data for training, 15% for verification, and 15% for testing, then set the relevant parameters of the RBF neural network and train the network, and finally embed the trained network into the DDPG network;

[0024] S6: Preprocess the information collected by the vehicle and the information fed back by the cloud, because the Actor network, the Critic network, and the built-in RBF neural network used when the data feedback is delayed have the same input;

[0025] S7: Select road condition K F , acceleration a, hydrogen consumption rate △H2, motor power requirement P M , bus power change △P, light intensity PV t 、Next moment PV t+Δt , temperature T parameter, define the rate of change of light intensity β, and select some parameters to form the state space S, S=[K F ,α,ΔH2,P M ,ΔP,β,T];

[0026] The pre-allocated output power of the fuel cell and the power battery is selected as the action variable and constitutes the action space a:

[0027] a=[P FC , P ba ]

[0028] Four optimization goals are determined, namely, minimum hydrogen consumption, maximum power battery life, power battery SOC maintenance, and stable bus output power. The reward function R is constructed based on the four optimization goals:

[0029] R=w1.△H2+w2.l+w3.△SOC+w4.smooth

[0030] Among them, w1, w2, w3, and w4 are weight coefficients of hydrogen consumption rate, power battery life, power battery SOC, and bus smoothness, respectively. l is the power battery life, and smooth is the bus smoothness;

[0031] S8: Take the form of power redistribution to transfer the power of the photovoltaic cells to the busbar. The formula is as follows:

[0032]

[0033] P′ ba =P pv -P′ fc

[0034] Among them, ε1, ε2, ε3, ε4 are the correlation coefficients of photovoltaic cell power redistribution, P′ fc is the power shared by the photovoltaic cell for the fuel cell, P′ ba is the power shared by the photovoltaic cell for the power battery, P ba , P fc They are the pre-allocated output power of the power battery and the fuel cell respectively;

[0035] P fc-d =P fc -P′ fc

[0036] P pv-d =P ba -P′ ba

[0037] Where P fc-d is the final fuel cell output power of the system, P pv-d Output power to the final power battery of the system.

[0038] The gradient descent method is used to calculate the update.

[0039] S9: Call the DDPG algorithm in the computer, create and initialize the training environment, construct the acquired test data into a training set, add random noise to the action variables to realize the function of exploring the environment, and constrain the probability distribution of the action. Then start training the network so that the network can obtain the optimal action based on the input state variables.

[0040] Preferably: in the step S9, the DDPG algorithm includes A2C and an experience pool, where A is a policy function responsible for generating continuous actions and interacting with the environment, and C is a value function that provides gradient information, evaluates the performance of the action, and provides feedback updates. The policy function is represented by a t =μ(S t ,|θ Q ), the value function is expressed as Q(S t , a t |θ Q );

[0041] The step S9 specifically includes the following steps:

[0042] S91: Collect multiple state-action pairs t =(S t , a t , R t , S t+1 ), get experience pool D t ={e1, e2, ...e t};

[0043] S92: When the algorithm is learning, it can directly sample from the experience pool;

[0044] S93: The label value calculation and error update of the value function are

[0045] S94: For the policy network, follow Gradients are used to update the network;

[0046] S95: For A2C and experience pool, corresponding evaluation networks and target networks are set to achieve delayed update. The value network is used to select actions, update parameters, and pass parameters to the target network using a small amount of approximation. The target network is used to calculate the learning target, which is the label value yi. Where τ is the update coefficient, and the random noise added to the output action is the Laplace process b t represents the rate at which noise decays as learning proceeds.

[0047] Beneficial effects:

[0048] The present invention integrates fuel cells, power batteries and photovoltaic cells to form a multi-energy power system for vehicles. It provides a method for calculating the optimal output ratio of the three energy sources and uses a deep deterministic policy gradient algorithm to explore the optimal control strategy. The algorithm fully considers the impact of external variables such as road conditions, light intensity, and traffic flow on energy management to reduce control errors and improve strategy reliability. The network information layer interacts with the vehicle system in real time, comprehensively analyzing multi-dimensional factors such as road conditions, light intensity, traffic flow, hydrogen consumption, busbar stability, and power battery durability, thereby achieving the best balance between hydrogen consumption and energy stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a flow chart of a power system and control method for a solar hydrogen electric vehicle provided by an embodiment of the present invention;

[0050] Figure 2 This is an energy layout diagram of a power system for a solar hydrogen electric vehicle provided by an embodiment of the present invention;

[0051] Figure 3 The invention provides a main circuit of a four-phase interleaved DC / DC converter for a hydrogen fuel cell;

[0052] Figure 4 It is a topological structure diagram of a light-hydrogen-electric hybrid vehicle power system provided by an embodiment of the present invention;

[0053] Figure 5 An energy management strategy framework based on DDPG is provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The technical solution of the present invention is further described in detail below in conjunction with specific implementation methods.

[0055] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", and "set" should be understood in a broad sense, for example, they can be fixedly connected or set, or detachably connected or set, or integrally connected or set. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0056] like Figure 1 As shown, the present invention is a power system and control method for a light-hydrogen-electric vehicle, and the specific method steps are as follows:

[0057] First, obtain weather information, including light intensity Pv and temperature T. Use the maximum power point tracking algorithm (MPPT) for photovoltaic cells, so that photovoltaic cells can output maximum power in real time when photovoltaic changes. Since the service life of hydrogen fuel cells is greatly affected by ripple current, and they have slow dynamic response, low voltage and high current and other working characteristics, conventional Boost converters are difficult to meet the power output requirements of hydrogen fuel cells. Therefore, in the first embodiment of this patent, it is proposed to adopt Figure 3 The non-isolated, four-phase interleaved variable boost DC / DC converter shown in the figure adopts a multi-phase current inner loop control method to control the current of each branch respectively.

[0058] In one embodiment, the vehicle operation process has different energy orientation requirements for the power system due to different operating conditions such as normal operating conditions, maximum driving power operating conditions, braking, etc. Therefore, the vehicle demand current change rate θ is defined as follows, and the fuel cell four-phase interleaved boost DC / DC converter is switched in working mode according to the value of α.

[0059]

[0060]

[0061] Where α ref is the vehicle reference current change rate, P max , P min is the maximum and minimum value of the load power; V arg is the average voltage under normal operating conditions; f(state) is the operating condition influence coefficient, which realizes the reference current change rate α under different operating conditions ref When α<0, the switches S7 and S8 are disconnected, which is equivalent to an interleaved parallel DC / DC converter; when 0<α<α ref When , the switch S8 is turned off, which is equivalent to a three-phase interleaved boost DC / DC converter;

[0062] When α>α ref When , the switch S8 is turned off, which is equivalent to a four-phase interleaved boost DC / DC converter;

[0063] The use of the above-mentioned fuel cell four-phase interleaved boost DC / DC converter can bring advantages such as reducing the output current ripple of the fuel cell, increasing the service life of the fuel cell and reducing the voltage stress of each switch tube.

[0064] In one embodiment, according to the hybrid system topology described in this patent, the energy configuration of the vehicle is carried out according to the following scheme: the photovoltaic panels are arranged in the roof area, and the fuel cell and the power battery are placed at the rear of the roof, such as Figure 2As shown. The system is suitable for vehicles with a length greater than 6.5 meters and a width greater than 2.1 meters. The system is designed with fuel cells as the core energy source, where the rated output power of the fuel cell should account for 62% to 68% of the rated power of the vehicle motor, and the rated output power of the power battery should account for 32% to 38% of the rated power of the vehicle motor. In order to ensure that the total rated output power of the fuel cell and the power battery should be maintained at 1% under night conditions when the photovoltaic cells cannot work normally. Considering the influence of the actual power of the system, the power factor k is introduced pf (0.9~1.0) to determine the final parameters of the fuel cell. The specific calculation formula is as follows.

[0065] P pf,fuel =k pf ·P fuel

[0066] P total =P pf,fuel +P bat +P pv

[0067] Through this configuration, not only can the vehicle's dynamic performance be ensured, but the potential of photovoltaic power generation can also be maximized, thereby optimizing energy utilization efficiency.

[0068] In one embodiment, the vehicle's sensor system measures the vehicle's physical signals, mainly including vehicle status information, power battery information, fuel cell information, photovoltaic system information, etc. The vehicle status information includes: vehicle speed v, acceleration a, vehicle position, etc.; the power battery information includes: power battery remaining power soc, power battery charge and discharge efficiency η ba , power battery voltage and current, internal resistance, etc.; the fuel cell information includes: fuel cell output power P FC , efficiency η fc , and hydrogen consumption rate △H2; the photovoltaic system information includes: light intensity PV, weather temperature T, photovoltaic cell output power P pv ;

[0069] Establish a vehicle dynamics model, mainly involving the vehicle longitudinal dynamics model; establish a fuel cell hydrogen consumption model, a power battery equivalent circuit model and a photovoltaic cell model; establish drive motor, DC / DC, DC / AC models; and make corresponding connections based on the hybrid system topology.

[0070] In one embodiment, the vehicle speed v, acceleration a, hydrogen consumption rate △H2, motor demand power P, etc. are extracted from the vehicle's CAN signal as state variables and fed back to the deep deterministic policy gradient algorithm network as observation values; the vehicle's geographic location is acquired in real time through the GPS module and uploaded to the cloud server, and after processing by the cloud server, the road signals around the vehicle are fed back to the vehicle, including road length, road slope, road curvature, road traffic information, real-time light intensity in the area where the vehicle is located, temperature and future changes.

[0071] When uploading vehicle-related data to the cloud server, there will be a certain feedback delay when the network fluctuates, resulting in the vehicle being unable to update parameters in real time, which in turn affects the power following of the bus. Therefore, when the delay time of the feedback data is greater than the threshold T, a built-in RBF neural network is used to predict the input data required by DDPG.

[0072] In one embodiment, when training RBF, first collect the data of the vehicle when it is running, including the road length, road slope, road curvature, road traffic information, the future changes of light intensity and temperature in the area where the vehicle is located, select 70% of the data for training, 15% of the data for verification, and 15% of the data for testing. Then set the relevant parameters of the RBF neural network and train the network, and finally embed the trained network into the DDPG network.

[0073] The information collected by the vehicle and the information fed back by the cloud are preprocessed. Since the Actor network, the Critic network, and the built-in RBF neural network used when data feedback is delayed have the same input, a common set of preprocessing networks is used to extract features from the input data. This can increase the speed of the network's preprocessing of data, thereby accelerating the network's convergence.

[0074] Select road condition K F , acceleration a, hydrogen consumption rate △H2, motor power requirement P M , bus power change △P, light intensity PV t 、Next moment PV t+Δt , temperature T and other parameters, define the light intensity change rate β, and select some parameters to form the state space S.

[0075]

[0076] S=[K F ,a,△H2,P M ,△P,β,T]

[0077] Compared with fixed photovoltaic cells, the output power of vehicle-mounted photovoltaic cells will fluctuate significantly. This is because the vehicle-mounted photovoltaic panels will be shielded by objects on both sides of the road, resulting in large fluctuations in light intensity. Therefore, this patent introduces the variable of light intensity change rate to more accurately analyze the performance of the vehicle-mounted photovoltaic system.

[0078] When the busbar power demand is high and the photovoltaic cell output power is low, the power battery alone may not be able to meet the demand. Therefore, it is necessary to significantly increase the output power of the fuel cell to maintain balance. To this end, the energy management strategy proposed in this patent comprehensively considers information such as changes in road conditions and the rate of change of light intensity to predict changes in busbar power demand and photovoltaic cell output power. And by increasing the opening of the hydrogen valve in advance, the situation of insufficient hydrogen supply when the fuel cell needs to output higher power is avoided, thereby increasing the output power of the fuel cell to cope with high power demand conditions.

[0079] The pre-allocated output power of the fuel cell and the power battery is selected as the action variable and constitutes the action space a:

[0080] a=[P FC , P ba ]

[0081] The present invention has four optimization goals, namely, minimum hydrogen consumption, maximum power battery life, power battery SOC maintenance, and bus output power stability. The reward function R is constructed based on these four optimization goals:

[0082] R=w1.△H2+w2.l+w3.△SOC+w4.smooth

[0083] Among them, w1, w2, w3, and w4 are weight coefficients of hydrogen consumption rate, power battery life, power battery SOC, and bus smoothness, respectively. l is the power battery life, and smooth is the bus smoothness.

[0084] Since photovoltaic cells have many uncontrollable factors and large fluctuations in output power, introducing them into DDPG will greatly increase the training time of the network. Therefore, the present invention adopts the form of power redistribution to transfer the power of photovoltaic cells to the bus. The output power of the fuel cell is limited by factors such as hydrogen consumption rate and power battery SOC. Therefore, when considering the output of photovoltaic cells, it is necessary to combine the above factors for allocation. The formula is:

[0085]

[0086] P′ ba =P pv -P′ fc

[0087] Among them, ε1, ε2, ε3, ε4 are the correlation coefficients of photovoltaic cell power redistribution, P′ fc is the power shared by the photovoltaic cell for the fuel cell, P′ ba is the power shared by the photovoltaic cell for the power battery, P ba , P fc They are the pre-allocated output power of the power battery and fuel cell respectively.

[0088] P fc-d =P fc -P′ fc

[0089] P pv-d =P ba -P′ ba

[0090] Where P fc-d is the final fuel cell output power, P pv-d Output power for the final power battery.

[0091] The gradient descent method is used here for calculation and updating. This algorithm can achieve real-time, accurate and optimal updating of the coefficients in the photovoltaic cell power redistribution formula.

[0092] Call the DDPG algorithm in the computer, create a training environment for it and initialize it. Construct the acquired test data into a training set. Since the DDPG network has only one output for a given input, it will lack the ability to explore the environment, so random noise is added to the action variable to achieve the function of exploring the environment. At the same time, constrain the probability distribution of its actions so that the entropy of the distribution is not too small, so as to try different actions and avoid the occurrence of local optimality. Then start training the network so that the network can obtain the optimal action based on the input state variables.

[0093] The vehicle longitudinal dynamics model established as claimed in claim 1 is:

[0094]

[0095] Among them, f is the rolling resistance coefficient, M is the curb mass, g is the acceleration of gravity, a is the slope, which is generally set to 0 in energy management, ρ is the air density, A represents the frontal area of ​​the vehicle, C is the air resistance coefficient, v is the vehicle speed, δ is the mass coefficient, which is set to 1, and a is the vehicle acceleration.

[0096] The specific form of the fuel cell model is:

[0097]

[0098] P batt =U batt Ibatt

[0099]

[0100]

[0101] Where τ is the time constant, τ = C pol .R pol The battery state of charge (SOC) is an important parameter to characterize the battery state. Its instantaneous change value is expressed as current and battery capacity Q cap ratio.

[0102] The specific form of the driving motor is:

[0103] η m =f(w m ,T m )

[0104]

[0105] When the motor speed w m and torque T m After determining, the efficiency η of the drive motor can be obtained m , n m is the motor speed.

[0106] The specific form of the photovoltaic cell model is:

[0107]

[0108]

[0109]

[0110] Where I sc is the short-circuit current, V oc is the open circuit voltage, I m is the maximum power point current, V m is the maximum power point voltage.

[0111] The method as claimed in claim 2 is characterized in that: the DDPG algorithm includes an actor and a critic, namely A2C, and an experience pool; wherein A is a policy function responsible for generating continuous actions and interacting with the environment, and C is a value function that provides gradient information, evaluates the performance of the action, and provides feedback updates. The policy function is represented by a t =μ(S t ,|θ Q ), the value function is expressed as Q(S t , a t|θ Q ); After training, the performer network can make the critic network output the maximum Q value.

[0112] Since the previous and next samples in deep learning problems are not independent of each other, there may be an obvious probability relationship between the previous and next states. Therefore, we use the experience replay technique to collect multiple state-action pairs. t =(S t , a t , R t , S t+1 ), get experience pool D t ={e1, e2, ...e t}, when the algorithm is learning, it can directly sample from the experience pool, which can smooth the data distribution, solve the problem of too strong correlation between data, and improve the generalization ability of the algorithm. The present invention adopts priority replay in experience replay. Since samples with larger TD errors have a greater impact on the network during back propagation, they are easier to be sampled, which can speed up the convergence speed of the algorithm during training.

[0113] The label value calculation and error update for the value function are:

[0114] y i =r i +γQ′(s i+1 ,μ′(s i+1 |θ μ′ )|θ Q′ )

[0115]

[0116] For the policy network, the network is updated according to the following gradient:

[0117]

[0118] The corresponding evaluation network and target network are set for both Actor and Critic to achieve delayed update. The evaluation network is used to select actions, update parameters, and pass parameters to the target network using a small amount of approximation. The target network is used to calculate the learning target, that is, the label value y i .

[0119] θ Q′ ←τθ Q +(1-τ)θ Q

[0120] θ μ′ ←τθ μ +(1-τ)θ μ′

[0121] Among them, τ is the update coefficient.

[0122] The random noise added to the output action is a Laplace process, which is in the following form:

[0123]

[0124] Among them, b t represents the rate at which noise decays as learning proceeds.

[0125] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A power system for a solar-hydrogen-electric vehicle, characterized in that: It includes fuel cells, power batteries, photovoltaic cells, drive motors, DC / DC converters and DC / AC converters, among which the DC / DC converter is a non-isolated, multi-phase interleaved variable boost DC / DC converter, the photovoltaic panels of the photovoltaic cells are arranged in the roof area, and the fuel cells and power batteries are placed at the rear of the roof.

2. A light-hydrogen-electric vehicle power system according to claim 1, characterized in that: The rated output power of the fuel cell should account for 62% to 68% of the rated power of the vehicle motor, while the rated output power of the power battery should account for 32% to 38% of the rated power of the vehicle motor. When working at night, the total rated output power of the fuel cell and the power battery is 100%.

3. A light-hydrogen-electric vehicle power system according to claim 1, characterized in that: The vehicle longitudinal dynamics model of the light-hydrogen-electric vehicle power system is: f is the rolling resistance coefficient, M is the curb mass, g is the acceleration of gravity, a is the slope, which is generally set to 0 in energy management, ρ is the air density, A represents the frontal area of ​​the vehicle, C is the air resistance coefficient, v is the vehicle speed, δ is the mass coefficient, which is set to 1, and a is the vehicle acceleration.

4. The solar-hydrogen-electric vehicle power system according to claim 1, characterized in that: The fuel cell model is E represents the open circuit voltage of the fuel cell, V fc and i fc Represents output voltage and output current, R ohm represents internal resistance, i0 represents exchange current, a, c2, c3, i max are all empirical constants.

5. The solar-hydrogen-electric vehicle power system according to claim 1, characterized in that: The power battery equivalent circuit model is: τ is the time constant, τ = C pol .R pol The battery state of charge (SOC) is an important parameter to characterize the battery state. Its instantaneous change value is expressed as the current and battery capacity Q cap ratio.

6. The light-hydrogen-electric vehicle power system according to claim 1, characterized in that: The model of the driving motor is η m =f(w m ,T m ), When the motor speed w m and torque T m After determining, the efficiency η of the drive motor can be obtained m , n m is the motor speed.

7. The solar-hydrogen-electric vehicle power system according to claim 1, characterized in that: The model of the photovoltaic cell is I sc is the short-circuit current, V oc is the open circuit voltage, I m is the maximum power point current, V m is the maximum power point voltage.

8. The solar-hydrogen-electric vehicle power system according to claim 1, characterized in that: The DC / DC converter has a logic of defining the vehicle demand current change rate α, the vehicle reference current change rate α ref , and according to the value of α, the working mode of the fuel cell multi-phase interleaved variable boost DC / DC converter is converted. The conversion formula is: P max , P min is the maximum and minimum value of the load power; V arg is the average voltage under normal operating conditions; f(state) is the operating condition influence coefficient, which realizes the reference current change rate α under different operating conditions ref make corrections; The conversion conditions of the DC / DC converter are as follows: ①: When α<0, the switches S7 and S8 are disconnected, which is equivalent to an interleaved parallel DC / DC converter; ②: When 0<α<α ref When , the switch S8 is turned off, which is equivalent to a three-phase interleaved boost DC / DC converter; ③: When α>α ref When the switch S8 is turned off, it is equivalent to a four-phase interleaved boost DC / DC converter; and considering the influence of the actual power of the system, the power factor k is introduced pf (0.9~1.0) to determine the final parameters of the fuel cell. After introducing the power coefficient, the calculation formula for the final parameters of the fuel cell is as follows: pf,fuel =k pf ·P fuel , P total =P pf,fuel +P bat +P pv .

9. A control method for a solar-hydrogen-electric vehicle power system, using a solar-hydrogen-electric vehicle power system according to any one of claims 1 to 8, characterized in that: The following steps are involved: S1: The vehicle's sensor system measures the vehicle's physical signals, which include vehicle status information, power battery information, fuel cell information, and photovoltaic system information. The vehicle status information includes: vehicle speed v, acceleration a, and vehicle position. The power battery information includes: power battery remaining power SOC, power battery charge and discharge efficiency ba, power battery voltage and current, and internal resistance. The fuel cell information includes: fuel cell output power P FC , efficiency η fc , and hydrogen consumption rate △H2, photovoltaic system information includes: light intensity PV, weather temperature T, photovoltaic cell output power P pv ; S2: The photovoltaic system uses the maximum power point tracking algorithm MPPT to control the photovoltaic cells to ensure that the output power of the photovoltaic cells is maximized; S3: Extract vehicle speed v, acceleration a, hydrogen consumption rate △H2, and motor required power P from the CAN signal in the vehicle as state variables and feed them back to the deep deterministic policy gradient algorithm network as observation values; obtain the vehicle's geographic location in real time through the GPS module and upload it to the cloud server, and after processing by the cloud server, feed back the road signals around the vehicle including road length, road slope, road curvature, road traffic information, real-time light intensity in the area where the vehicle is located, temperature, and future changes; S4: Upload vehicle-related data to the cloud server. When the delay time of the feedback data is greater than the threshold T, use a built-in RBF neural network to predict the input data required by DDPG; S5: Collect data from the running of the array vehicles, including road length, road slope, road curvature, road traffic information, future changes in light intensity and temperature in the area where the vehicle is located, select 70% of the data for training, 15% for verification, and 15% for testing, then set the relevant parameters of the RBF neural network and train the network, and finally embed the trained network into the DDPG network; S6: Preprocess the information collected by the vehicle and the information fed back by the cloud, because the Actor network, the Critic network, and the built-in RBF neural network used when the data feedback is delayed have the same input; S7: Select road condition K F , acceleration a, hydrogen consumption rate △H2, motor power requirement P M , bus power change △P, light intensity PV t 、Next moment PV t+Δt , temperature T parameter, define the rate of change of light intensity β, and select some parameters to form the state space S, S=[K F ,α,ΔH2,P M , ΔP, β, T]; The pre-allocated output power of the fuel cell and the power battery is selected as the action variable and constitutes the action space a: a=[P FC 、P ba ] Four optimization goals are determined, namely, minimum hydrogen consumption, maximum power battery life, power battery SOC maintenance, and stable bus output power. The reward function R is constructed based on the four optimization goals: R=w1.△H2+w2.l+w3.△SOC+w4.smooth Among them, w1, w2, w3, and w4 are weight coefficients of hydrogen consumption rate, power battery life, power battery SOC, and bus smoothness, respectively; l is the power battery life, and smooth is the bus smoothness; S8: Take the form of power redistribution to transfer the power of the photovoltaic cells to the busbar. The formula is as follows: P b ′ a =P pv -P f ′ c Among them, ε1, ε2, ε3, ε4 are the correlation coefficients of photovoltaic cell power redistribution, P f ' c is the power shared by the photovoltaic cell for the fuel cell, P b ' a is the power shared by the photovoltaic cell for the power battery, P ba , P fc They are the pre-allocated output power of the power battery and the fuel cell respectively; P fc-d =P fc -P f ′ c P pv-d =P ba -P b ′ a Where P fc-d is the final fuel cell output power of the system, P pv-d The final power battery output power of the system, The gradient descent method is used to calculate the update. S9: Call the DDPG algorithm in the computer, create and initialize the training environment, construct the acquired test data into a training set, add random noise to the action variables to realize the function of exploring the environment, and constrain the probability distribution of the action. Then start training the network so that the network can obtain the optimal action based on the input state variables.

10. The control method of a solar-hydrogen-electric vehicle power system according to claim 9, characterized in that: In the S9 step, the DDPG algorithm includes A2C and the experience pool, where A is the policy function responsible for generating continuous actions and interacting with the environment, and C is the value function, which provides gradient information, evaluates the performance of the action, and provides feedback updates. The policy function is represented by a t =μ(S t ,|θ Q ), the value function is expressed as Q(S t ,a t |θ Q ); The step S9 specifically includes the following steps: S91: Collect multiple state-action pairs t =(S t ,a t ,R t ,S t+1 ), get experience pool D t ={e1,e2,…e t }; S92: When the algorithm is learning, it can directly sample from the experience pool; S93: The label value calculation and error update of the value function is: y i =r i +γQ′(s i+1 ,μ′(s i+1 |θ u′ )|θ Q′ ) L(θ)=∑ i (y i -Q(s i ,a i |θ Q )) 2 ; S94: For the policy network, follow Gradients are used to update the network; S95: For A2C and experience pool, corresponding evaluation networks and target networks are set to achieve delayed update. The value network is used to select actions, update parameters, and pass parameters to the target network using a small amount of approximation. The target network is used to calculate the learning target, which is the label value y i ,θ Q′ ←τθ Q +(1-τ)θ Q θ μ′ ←τθ μ +(1-τ)θ μ′ , where τ is the update coefficient, and the random noise added to the output action is the Laplace process b t represents the rate at which noise decays as learning proceeds.

Images