Fuel cell hybrid locomotive energy scheduling method and system

Abstract

The invention discloses a fuel cell hybrid train energy dispatching method and system. The method comprises the steps that S1, a hydrogen fuel hybrid train power system model is constructed; s2, acquiring required power in a train running process, and decomposing the required power into a low-frequency component and a high-frequency component by using a low-pass filter with adjustable cut-off frequency; constructing a double-layer reinforcement learning agent, and performing real-time dynamic optimization adjustment by taking the cut-off frequency of the low-pass filter as an action variable of the reinforcement learning agent; s3, designing a multi-objective reward function and a system total cost function, and synchronously optimizing fuel cell efficiency, fuel cell power fluctuation, hydrogen consumption and lithium battery SOC to approach initial values; and updating double-layer reinforcement learning agent network parameters through a depth deterministic strategy gradient algorithm, and finally outputting a real-time power distribution instruction of the fuel cell and the lithium battery. The method has the advantages of reducing the fluctuation of the output power of the hydrogen fuel cell, reducing the hydrogen consumption and the like.

Description

TECHNICAL FIELD

[0001] The present application mainly relates to the technical field of vehicle energy management, and in particular to a fuel cell hybrid locomotive energy scheduling method and system. BACKGROUND

[0002] Due to the increase in population and the increase in the scale of human production activities, the ecological environment is damaged and the energy crisis problem is caused. The new power system combined with fuel cells and hydrogen is regarded as a potential solution to replace the traditional internal combustion engine technology and applied to the field of rail transportation to reduce pollution emissions and promote the development of green transportation. Due to the soft output characteristics, slow power response and inability to recover the energy feedback during train braking of hydrogen fuel cells, only using hydrogen fuel cells as power supply cannot meet the demand of the load. Usually, the battery and super capacitor are selected as auxiliary power supply to form a hybrid power supply with hydrogen fuel cells. When the train accelerates or runs at a constant speed, the hydrogen fuel cell serves as the main power supply to provide the required power for the load, and if there is a large power demand, the auxiliary power supply will supplement the peak power exceeding the power supply capacity of the hydrogen fuel cell; when the train decelerates, the load demand power is negative, and the auxiliary power supply will recover and store the feedback energy to improve the energy utilization rate and reduce the overall operation cost of the train. Designing a reasonable and reliable energy management strategy to coordinate the energy flow between multiple power supplies and between the power supply and the load is the key research content of the fuel cell hybrid train.

[0003] Current theoretical and experimental researches mainly focus on reducing fuel consumption or improving the service life of fuel cells in a single aspect, and the energy scheduling methods based on rules and optimization have defects such as dependence on engineering experience, inability to guarantee optimal control, large amount of calculation, poor real-time performance, and possible local optimization. Based on this, the present application proposes an energy scheduling method based on frequency decoupling and reinforcement learning by constructing a power system model of a hydrogen fuel hybrid train, considering fuel cell life and energy saving, and distributing the optimal fuel cell output power.

[0004] The power system model of the hydrogen fuel hybrid train is usually modeled by using the formula approximation method, and the model accuracy is not high.

[0005] The energy scheduling methods based on rules and optimization have defects such as dependence on engineering experience, large amount of calculation and poor real-time performance, and the energy scheduling method based on reinforcement learning substitutes the approximate formula in the environmental modeling aspect, which has poor accuracy, resulting in poor strategy availability, and usually considers a single target. SUMMARY

[0006] In view of the technical problems existing in the prior art, the present application provides a fuel cell hybrid locomotive energy scheduling method and system for reducing the fluctuation of hydrogen fuel cell output power and hydrogen consumption.

[0007] To solve the above technical problems, the technical scheme provided by the present application is: A fuel cell hybrid vehicle energy scheduling method, comprising the following steps: S1, a hydrogen fuel hybrid train power system model is constructed, the power system model includes a fuel cell model, a lithium battery model and a traction motor model; S2, the demand power in the train running process is obtained, and a low-pass filter with adjustable cutoff frequency is used to decompose the demand power into low-frequency components and high-frequency components; A double-layer reinforcement learning agent is constructed, and the cutoff frequency of the low-pass filter is used as the action variable of the reinforcement learning agent for real-time dynamic optimization and adjustment; The double-layer reinforcement learning agent includes an outer agent and an inner agent; the outer agent is used to extract the global features of the demand power and output an initial normalized cutoff frequency value; the inner agent receives the initial normalized cutoff frequency value, optimizes the power distribution strategy in real time, and takes the low-frequency component as the fuel cell output power and the high-frequency component as the lithium battery output power; S3, a multi-objective reward function and a system total cost function are designed to simultaneously optimize the fuel cell efficiency, fuel cell power fluctuation, hydrogen consumption, and lithium battery SOC close to the initial value; The double-layer reinforcement learning agent network parameters are updated through a deep deterministic policy gradient algorithm, and finally the real-time power distribution instructions of the fuel cell and the lithium battery are output.

[0008] Preferably, in step S3, when the braking power is detected: if the lithium battery SOC is greater than or equal to the safety threshold, the regenerative braking energy is distributed to the braking resistor for consumption, otherwise the energy is recovered by the lithium battery.

[0009] Preferably, in step S3, the reward function r And the system total cost cost total Respectively:

[0010]

[0011] In the formula, The fuel cell efficiency is The remaining capacity of the current power battery is The standard deviation of the fuel cell output power is The fuel consumption of the adjacent two time steps is The corresponding weight coefficient is The cost of starting the fuel cell once is The number of fuel cell start-ups is And for the operating costs of fuel cells and power cells, and for the output power of fuel cells and power cells.

[0012] Preferably, in the fuel cell model, the output voltage of a single fuel cell is given by the following equation: ; wherein, V is the output voltage of a single PEMFC; N is the Nernst voltage; R is the voltage loss due to activation polarization effect; R is the voltage loss due to ohmic polarization effect; R is the voltage loss due to concentration polarization effect.

[0013] Preferably, in the lithium battery model, the output voltage of a lithium battery during charging and discharging is given by the following equation: Discharge:

[0014] Charge:

[0015] wherein, V is the output voltage of the battery, V is the constant output voltage of the battery, R is the polarization constant, C is the battery capacity, Q is the actual battery charge, A is the amplitude of the exponential region, A is the inverse of the amplitude of the exponential region, R is the internal resistance, I is the battery current, I is the filtered battery current; the state of charge of the current battery is given by the following equation:

[0016] wherein, Q is the remaining charge of the battery at the initial time; Q is the state of charge of the battery at the next time; Q is the total charge at the current time; Q is the total charge at the next time; Q represents the charge that the battery has taken in under charging conditions; Q represents the charge that the battery has given out under discharging conditions.

[0017] ​Preferably, in the traction motor model, the demand torque and the rotating speed of the traction motor are obtained, and then the demand power of the traction motor is obtained.

[0018] The application further discloses a computer program product comprising a computer program which, when executed by a processor, performs the steps of the method as described above.

[0019] The application further discloses a computer readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the steps of the method as described above.

[0020] The application further discloses a fuel cell hybrid vehicle energy scheduling system comprising a memory and a processor connected to each other, wherein the memory has a computer program stored thereon, and the computer program, when executed by the processor, performs the steps of the method as described above.

[0021] Compared with the prior art, the application has the following advantages: The application designs a "global-instantaneous" double-agent architecture: the outer layer solves the initial value convergence problem, and the inner layer realizes dynamic power distribution, breaking the feature extraction limitation of the traditional single-agent reinforcement learning; a dynamic frequency decoupling mechanism: the filter cutoff frequency fc is used as a learnable action variable to replace the fixed threshold, so that the power distribution is self-adaptive to the working condition change; a multi-objective reward function design: for the first time, fuel cell life (fluctuation suppression), energy consumption (hydrogen consumption), battery state (SOC) and other multi-dimensional objectives are integrated, solving the single-objective optimization problem in the industry.

[0022] Based on the mechanism model of the fuel cell and the power battery, the application constructs a hybrid power system simulation model, improves the accuracy of the model, and further improves the usability of the energy management strategy, interacts with the environment and the agent strategy, and supports the cyclic training of reinforcement learning.

[0023] Based on the mechanism model, the application trains and interacts with the reinforcement learning agent, reduces the fuel cell energy consumption through the design of the reward function, and introduces a filter to decouple the demand power, allocates the low-frequency part to the fuel cell output, improves the service life of the fuel cell, and reduces the fluctuation of the hydrogen fuel cell output power and the hydrogen consumption. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 FIG. 1 is a fuel cell hybrid energy scheduling system architecture diagram of the application.

[0025] Figure 2 FIG. 2 is a hybrid power system architecture diagram of the application.

[0026] Figure 3Fig. 1 is a fuel cell output characteristic curve diagram of the present application; (a) is a volt-ampere characteristic curve diagram; (b) is an output power-current curve diagram.

[0027] Figure 4 Fig. 2 is a schematic diagram of the influence of temperature on the output characteristics of the fuel cell in the present application; (a) is a schematic diagram of the influence of temperature on output voltage; (b) is a schematic diagram of the influence of temperature on output efficiency.

[0028] Figure 5 Fig. 3 is a lithium battery discharge curve diagram in the present application; (a) is a discharge curve diagram under nominal current; (b) is a discharge curve diagram when the discharge current = 100A, 500A, 1000A.

[0029] Figure 6 Fig. 4 is a schematic diagram of a traction motor calculation model in the present application.

[0030] Figure 7 Fig. 5 is a schematic diagram of a demand frequency decoupling method in the present application.

[0031] Figure 8 Fig. 6 is a schematic diagram of a frequency decoupling power distribution method in the present application.

[0032] Figure 9 Fig. 7 is a DDPG reinforcement learning algorithm training architecture diagram in the present application. DETAILED DESCRIPTION

[0033] The present application is further described below in conjunction with the accompanying drawings and specific embodiments.

[0034] As shown in Figure 1 , the fuel cell hybrid vehicle energy scheduling method provided by the embodiments of the present application comprises the following steps: S1, a hydrogen fuel hybrid train power system model is constructed, and the power system model comprises a fuel cell model, a lithium battery model and a traction motor model; As shown in Figure 2 , the power source characteristics of the hydrogen fuel hybrid train are analyzed, and a corresponding vehicle power system model is established, which specifically comprises: Hydrogen fuel cell model: the response characteristics of the fuel cell are analyzed, and the output characteristics of the hydrogen fuel cell are affected by factors such as temperature, gas pressure and load current. The output voltage of the single fuel cell is shown in formula (1): (1) In the formula, is the output voltage of the single proton exchange membrane fuel cell; is the Nernst voltage, which is the theoretical maximum voltage; is the voltage loss caused by the activation polarization effect; is the voltage loss caused by the ohmic polarization effect; It is the voltage loss caused by concentration polarization effect.

[0035] The Nernst voltage The expression is shown in formula (2): (2) Where, The internal temperature of the hydrogen fuel cell during operation, in ; is the Faraday constant, which is 96485; is the universal ideal gas constant, which is 8.314; is the partial pressure of hydrogen; is the partial pressure of oxygen.

[0036] Voltage loss caused by activation polarization effect As shown in formula (3): (3) Where, is the conversion factor, which takes a value of 0.5; is the actual current density, in units of ; is the exchange current density, in units of , the empirical value is .

[0037] Voltage loss due to ohmic polarization effect As shown in formula (4): (4) Where, is the external output current of the hydrogen fuel cell; is the equivalent resistance of the proton exchange membrane; It is the membrane resistance parameter, which is used to characterize the resistance encountered by protons passing through the exchange membrane.

[0038] Voltage loss caused by concentration polarization effect As shown in formula (5): (5) Where, is a constant, determined by the operating state of the hydrogen fuel cell; is the maximum current density; It is the ratio of the actual current to the membrane area during the operation of the hydrogen fuel cell, which is used to characterize the load degree of the hydrogen fuel cell. The unit is .

[0039] The output characteristic curve of the proton exchange membrane fuel cell is as follows: Figure 3As shown in Figure 2, when the output power gradually increases from zero, the output voltage drops rapidly. At this time, it is in the activation polarization stage, as shown in Figure 2. Figure 3 As shown in (a); when the output power increases to a certain value, the output voltage decreases linearly with the current, and is now in the ohmic polarization stage, as shown in Figure 3 As shown in (b).

[0040] The effect of temperature on the output voltage and efficiency of fuel cells is as follows: Figure 4 As shown in Figure 2, within a certain temperature range, as the temperature increases, the output voltage and efficiency both increase. Figure 4 As shown in (a) and (b).

[0041] Lithium battery model: The output voltage expression of lithium battery during charging and discharging is as follows: The output voltage expression during discharge is: (6) The output voltage expression during charging is: (7) Where, is the battery output voltage (V), is the constant output voltage of the battery (V), is the polarization constant (V / Ah), is the battery capacity (Ah), is the actual battery capacity (Ah), is the exponential region amplitude (V), is the inverse of the amplitude in the exponential zone ( / Ah), is the internal resistance (Ω), is the battery current (A), Filter current for the battery (A).

[0042] The SOC estimation method used in this invention is the charge integration method. Under the premise of knowing the remaining battery capacity at the previous moment, the charge and discharge current of the battery over a period of time is integrated to obtain the current state of charge of the battery. The calculation method is as follows: (8) Where, is the initial moment The battery's state of charge; The next moment The battery's state of charge; is the total charge at the current moment; is the total charge at the next moment; Respectively represent the amount of charge charged into / discharged from the battery under charging / discharging conditions.

[0043] The discharge curve of the hydrogen fuel hybrid train power battery at the rated current (nominal current) of 155.217A is as follows: Figure 5 As shown in (a); the discharge curves when the discharge current = 100A, 500A, and 1000A are as follows Figure 5 As shown in (b).

[0044] The discharge curve can be divided into three regions. The first region (such as Figure 5 In (a), the yellow area is also called the exponential area), and the terminal voltage drops rapidly. If the discharge current increases, the voltage drops faster. The second area is also called the rated working area (such as Figure 5 (a) The gray area in the middle, also called the nominal area), the voltage drops slowly, and the smaller the discharge rate, the longer the rated working area lasts; the third area (such as Figure 5 In (a), the blank area A shows that when the battery is almost fully discharged, the voltage drops rapidly.

[0045] Traction motor model: resistance during train operation Including air resistance and friction , the calculation method is as follows: (9) Where, is the air resistance coefficient; is the air density, take ; is the frontal area of ​​the train; is the actual running speed of the train; is the coefficient of sliding friction; m is the mass of the train; is the acceleration due to gravity, take .

[0046] The difference between the actual speed of the train and the reference speed is controlled proportionally and differentially to obtain the traction force required for the actual speed to track the reference speed. As shown in formula (10), the total traction force required by the train is For resistance And the sum of the tractive effort required to track the reference speed: (10) Using the total traction force required by the train obtained above and formula (11), the required torque of the traction motor can be calculated: : (11) Where, is the radius of the train wheel; The transmission ratio is defined as the ratio of the main wheel speed to the follower wheel speed. In trains, it is also defined as the ratio of the wheel torque to the traction motor torque. is the efficiency of the gearbox.

[0047] Traction motor speed The actual speed of the train The relationship is shown in formula (12): (12) Where, is the speed of the traction motor, in units of ; The reduction ratio, that is, the transmission ratio of the reduction device, refers to the ratio of the instantaneous input speed to the output speed in the reduction device.

[0048] Substituting the required torque and speed of the traction motor obtained above into formula (13), the required power of the traction motor can be obtained : (13) Since the voltage of the DC bus in the system is known, the required current of the traction motor can be calculated.

[0049] In summary, the traction motor calculation model is as follows Figure 6 shown.

[0050] S2, such as Figure 1 As shown in the figure, the energy scheduling of hydrogen fuel hybrid train adopts a hierarchical structure, which is mainly composed of calculation modules such as operating speed planning, required power calculation, and energy scheduling unit. Figure 1 In the system, the operating speed planning module calculates the global speed curve based on the upper-level timetable, section speed limit, and train parameters; the required power calculation module calculates the required power of the route based on the global speed curve and the train power balance equation, and further passes it to the energy scheduling module for power distribution. The goal is to reduce the fluctuation of fuel cell output power, improve service life, and reduce hydrogen consumption of the route. The final power distribution curve is sent to the real-time controller for final control implementation.

[0051] The energy management strategy based on frequency decoupling starts from the characteristics of hydrogen fuel cells and lithium batteries, and divides the required power into two parts: low-frequency component and high-frequency component. The separated low-frequency component is used as the output power of the hydrogen fuel cell, and the high-frequency component is used as the output power of the lithium battery. Figure 7 Specifically, the required power is divided into a low-frequency component and a high-frequency component through a second-order low-frequency filter with an adjustable normalized cutoff frequency.

[0052] Construct a two-layer reinforcement learning agent and normalize the cutoff frequency of the low-pass filter The action variable of the reinforcement learning agent is dynamically optimized and adjusted in real time. The calculation formula is as follows, and the value is between 0 and 1: (14) As shown in Figure 9 , the double-layer reinforcement learning agent is composed of an inner agent and an outer agent, which are respectively responsible for learning the instantaneous characteristics and global characteristics of the vehicle demand power. The outer agent is responsible for predicting the threshold of frequency decoupling, and the result of frequency decoupling is sent to the inner agent as an initial value, which can effectively solve the problems of initial value convergence and global feature extraction; the inner agent takes the initial normalized cutoff frequency value output by the outer agent as the starting point, and optimizes the power distribution strategy in real time, taking the low-frequency component as the fuel cell output power and the high-frequency component as the lithium battery output power.

[0053] S3, design a multi-objective reward function and a system total cost function, and simultaneously optimize the fuel cell efficiency, fuel cell power fluctuation, hydrogen consumption and lithium battery SOC close to the initial value; update the agent network parameters through the DDPG algorithm (Deep Deterministic Policy Gradient, Deep Deterministic Policy Gradient algorithm); Specifically, the DDPG algorithm updates the parameters of the Actor and Critic networks using the target network. The optimization of the Critic network is to minimize the network error, and the process of training the Critic network is to find the optimal solution of the network parameters . The process of training the Actor network is to find the optimal solution of the Actor network parameters , and the random gradient descent method is used to optimize the Actor network parameters.

[0054] In the actual training of the method, in each large cycle, first, initialize the state variable and reset the train running condition. After the simulation operation starts, in each small cycle, first, get the action a t from the Actor network a t , take the average value as the mean value, and the parameter var as the variance to form a normal distribution, and randomly select an action as the normalized cutoff frequency of the second-order low-pass filter .

[0055] The filter is used for filtering the demand power, and a low-frequency part is distributed to the output power of the fuel cell by the energy management system. The simulation running pause time and the output power distributed to the fuel cell are written into the simulation model, and the hydrogen consumption, the lithium battery SOC, the actual output power of the fuel cell, the fuel cell output efficiency, the total cost of the system and the like are obtained from the model. The fuel cell output efficiency, the total cost of the system cost total , the lithium battery SOC, the fuel cell output standard deviation and the hydrogen consumption are taken as the reward function r components, and the calculation method is as follows: (15) (16) In the formula, is the fuel cell efficiency, is the remaining power of the current power battery, is the fuel cell output power standard deviation, is the fuel consumption of adjacent two time steps; is the corresponding weight coefficient of each item; is the cost of starting the fuel cell once, is the starting frequency of the fuel cell, and are the operation costs of the fuel cell and the power battery respectively, and are the output powers of the fuel cell and the power battery respectively; if the fuel cell output efficiency is higher, the total cost of the system is lower, the lithium battery SOC is closer to the initial value, the fuel cell output power standard deviation is smaller, and the hydrogen consumption is lower, the reward value obtained by the intelligent agent is larger, and it is more beneficial to achieve the optimization goal of the system.

[0056] The present application uses the normalized cutoff frequency, the fuel cell efficiency and the lithium battery SOC as state variables s t ; when the system simulation is paused for one second, the state variables obtained from the simulation model will constitute the state space at the next time s t+1 ; the four-tuple ( s t , a t , r t , s t+1 ) is stored in the DDPG experience replay buffer pool, the capacity of the buffer pool is 10000, when the storage space of the buffer pool is full, the normal distribution variance parameter is reduced to reduce the exploration rate of the intelligent agent; and a plurality of groups of data are selected from the buffer pool to train the intelligent agent, and the network parameters are updated. Wherein st is the state instruction of the current agent, including normalized cutoff frequency, fuel cell efficiency, lithium battery SOC; s t+1 is the state instruction of the next moment agent; a t is the action instruction of the current agent, specifically adjusting the normalized cutoff frequency; r t is the multi-objective reward function of the current agent.

[0057] The learning rate of the Actor network and the Critic network is set to 1x10 -3 The Actor_target network and the Critic_target network parameters are updated at a smoothing rate of 0.01, and the Q target value is updated using a discount factor of 0.9 according to the Bellman equation. Figure 9 The training architecture of the reinforcement learning algorithm is shown in

[0058] Where is to prevent the lithium battery from overcharging when the train brakes, the model of the application also introduces a braking resistor, and the specific power distribution method is shown in Figure 8 Where the demand power is subtracted from the lithium battery output power and the fuel cell output power, and the remaining power is consumed by the braking resistor, thereby preventing the lithium battery from overcharging.

[0059] The application designs a "global-instantaneous" double-layer agent architecture, the outer agent solves the initial value convergence problem, and the inner agent realizes dynamic power distribution, breaking through the feature extraction limitation of the traditional single agent of reinforcement learning; through the dynamic frequency decoupling mechanism, the filter cutoff frequency is taken as a learnable action variable, replacing the fixed threshold, so that the power distribution is self-adaptive to the working condition change; a multi-objective reward function is designed, which first integrates fuel cell life (fluctuation suppression), energy consumption (hydrogen consumption), battery state (SOC) and other multi-dimensional objectives, solving the single objective optimization drawbacks in the industry.

[0060] The application is based on the mechanism model of fuel cells and power batteries, constructs a hybrid power system simulation model, improves the accuracy of the model, and further improves the usability of the energy management strategy, interacts with the environment and the agent strategy, and supports the cycle training of reinforcement learning.

[0061] The application is based on the mechanism model to construct a hybrid system environment model and train the reinforcement learning agent, reduce the fuel cell energy consumption through the design of the reward function, and introduce a filter to decouple the demand power, distribute the low-frequency part to the fuel cell output, and improve the service life of the fuel cell, reduce the fluctuation of the hydrogen fuel cell output power and hydrogen consumption.

[0062] The application provides an energy scheduling method based on frequency decoupling and reinforcement learning, which considers fuel cell life and energy saving simultaneously, utilizes working condition data on actual vehicle running routes for training, effectively reduces hydrogen consumption under fixed working conditions, simultaneously reduces fluctuation of fuel cell output power, and improves service life.

[0063] The energy scheduling method based on frequency decoupling and reinforcement learning trains the energy scheduling problem through reinforcement learning, obtains an energy scheduling strategy, i.e., an output power distribution strategy, which can improve fuel cell life and reduce fuel consumption through decoupling and continuous trial-and-error updating of demand power, and verifies effectiveness of the energy scheduling method through simulation verification of positive line real vehicle data backfilling, thereby laying an engineering foundation.

[0064] The application further discloses a computer program product, including a computer program, which executes steps of the above method when run by a processor. The application further discloses a computer readable storage medium, which stores a computer program, which executes steps of the above method when run by a processor. The application further discloses a fuel cell hybrid vehicle energy scheduling system, including a memory and a processor connected to each other, the memory stores a computer program, which executes steps of the above method when run by the processor. The product, medium and system of the application correspond to the above method, and also have the advantages of the above method.

[0065] The present application can realize all or part of the processes in the above-mentioned embodiment methods, and can also be completed by computer program instruction related hardware. The computer program can be stored in a computer readable storage medium. When the computer program is executed by a processor, the steps of the above-mentioned method embodiment can be realized. The computer program includes computer program code, which can be in the form of source code, object code, executable files or some intermediate forms, etc. The computer readable storage medium includes any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. The memory is used to store computer programs and / or modules. The processor realizes various functions by running or executing the computer programs and / or modules stored in the memory, and calling the data stored in the memory. The memory can include high-speed random access memory, and can also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one magnetic disk storage device, flash memory device, or other volatile solid-state storage device, etc.

[0066] The above is only the preferred embodiment of the present application, and the protection scope of the present application is not limited to the above-mentioned embodiment. Any technical solution falling within the concept of the present application shall fall within the protection scope of the present application. It should be noted that some improvements and refinements made by ordinary skilled in the art without departing from the principles of the present application shall be considered as the protection scope of the present application.

Claims

1. A fuel cell hybrid vehicle energy scheduling method, characterized in that: The following steps are involved: S1. Constructing a hydrogen fuel hybrid train power system model, wherein the power system model includes a fuel cell model, a lithium battery model, and a traction motor model; S2. Obtain the required power during train operation and decompose the required power into low-frequency and high-frequency components using a low-pass filter with an adjustable cutoff frequency; A two-layer reinforcement learning agent is constructed, and the cutoff frequency of the low-pass filter is used as the action variable of the reinforcement learning agent for real-time dynamic optimization and adjustment. The dual-layer reinforcement learning agent includes an outer agent and an inner agent; the outer agent is used to extract the global characteristics of the required power and output an initial normalized cutoff frequency value; the inner agent receives the initial normalized cutoff frequency value and optimizes the power allocation strategy in real time, using the low-frequency component as the fuel cell output power and the high-frequency component as the lithium battery output power; S3. Design a multi-objective reward function and a total system cost function to simultaneously optimize fuel cell efficiency, fuel cell power fluctuation, hydrogen consumption, and lithium battery SOC close to the initial value; The network parameters of the two-layer reinforcement learning agent are updated through a deep deterministic policy gradient algorithm, and finally the real-time power allocation instructions of the fuel cell and lithium battery are output.

2. The fuel cell hybrid vehicle energy scheduling method according to claim 1, characterized in that: In step S3, when the braking power is detected: if the lithium battery SOC ≥ the safety threshold, the regenerative braking energy is allocated to the braking resistor for consumption; otherwise, the energy is recovered by the lithium battery.

3. The fuel cell hybrid vehicle energy scheduling method according to claim 1 or 2, characterized in that: In step S3, the reward function r Total system cost cost total They are: Where, is the fuel cell efficiency, is the remaining power of the current power battery, is the standard deviation of fuel cell output power, is the fuel consumption in two adjacent time steps; is the corresponding weight coefficient; The cost of starting a fuel cell once, is the number of fuel cell starts, and The operating costs of fuel cells and power batteries, and is the output power of the fuel cell and power battery.

4. The fuel cell hybrid vehicle energy scheduling method according to claim 1 or 2, characterized in that: In the fuel cell model, the output voltage of a single fuel cell is expressed as follows: ； Where, is the output voltage of a single PEMFC; is the Nernst voltage; Voltage loss caused by activation polarization effect; The voltage loss is caused by the ohmic polarization effect; It is the voltage loss caused by concentration polarization effect.

5. The fuel cell hybrid vehicle energy scheduling method according to claim 1 or 2, characterized in that: In the lithium battery model, the output voltage expression of the lithium battery during charging and discharging is as follows: Discharge: Charge: Where, is the battery output voltage, is the constant output voltage of the battery, is the polarization constant, is the battery capacity, is the actual battery capacity, is the exponential region amplitude, is the inverse of the exponential region amplitude, is the internal resistance, is the battery current, Filters current for batteries; Current battery state of charge The expression is: Where, is the remaining charge of the battery at the initial moment; The next moment The battery's state of charge; is the total charge at the current moment; is the total charge at the next moment; Represents the amount of charge charged into the battery under charging conditions; Represents the amount of charge released by the battery under discharge conditions.

6. The fuel cell hybrid vehicle energy scheduling method according to claim 1 or 2, characterized in that: In the traction motor model, the required torque and speed of the traction motor are obtained, and then the required power of the traction motor is obtained.

7. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are performed.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the computer program performs the steps of the method according to any one of claims 1 to 6.

9. A fuel cell hybrid vehicle energy dispatch system, comprising a memory and a processor connected to each other, wherein a computer program is stored in the memory, characterized in that: When the computer program is executed by a processor, the computer program performs the steps of the method according to any one of claims 1 to 6.

Images