A fuel cell vehicle energy management method based on neural network target shooting

Abstract

The present invention discloses an energy management method for a fuel cell vehicle based on neural network target shooting, comprising the following steps: S1. Establish a power transmission system model of the fuel cell vehicle, including a vehicle longitudinal dynamics model, a motor model, and an energy source power balance model; S2. Use dynamic programming to solve the optimal energy management problem of the fuel cell vehicle, and establish an optimal data set according to the solution results; S3. Train a neural network based on the optimal data set; S4. Based on the monotonic relationship between the final value and the target value of the state of charge (SoC) of the power battery under the neural network energy management strategy, use the bisection method to perform target shooting on the final SoC under the test conditions until it reaches the expected value. The present invention can accurately control the final state of charge of the power battery under the neural network energy management strategy and achieve approximate optimal energy consumption economy with high computational efficiency.

Description

Technical Field

[0001] The present invention relates to the field of energy management of fuel cell vehicles, and particularly to an energy management method for fuel cell vehicles based on neural network shooting. Background Art

[0002] Neural networks can be used for online energy management of fuel cell vehicles to achieve power distribution between fuel cell systems and power batteries. However, the performance of existing neural network energy management controllers cannot be adjusted during application, resulting in the final SoC of the power battery not being accurately controlled to the desired value. This further leads to the inability of the neural network controller to have the ability to adapt to working conditions, and its application scenarios are also limited. Summary of the Invention

[0003] The purpose of the present invention is to make up for the deficiencies of the existing technology and propose an energy management method for fuel cell vehicles based on neural network shooting. This method introduces a shooting strategy based on neural networks, which can achieve adjustable neural network output and controllable final battery state, thereby providing a basic condition for the working condition adaptability of the neural network energy management strategy.

[0004] To achieve the above purpose, the present invention adopts the following technical solutions: An energy management method for fuel cell vehicles based on neural network shooting, comprising the following steps:

[0005] S1. Establish a power transmission system model for fuel cell vehicles, including a vehicle longitudinal dynamics model, a motor model, and an energy source power balance model;

[0006] S2. Solve the optimal energy management problem of fuel cell vehicles using dynamic programming, and establish an optimal data set according to the solution results;

[0007] Further, the optimal energy management problem of fuel cell vehicles described in step S2 is specifically as follows:

[0008] The optimization objective is to minimize the total hydrogen consumption, the state variable is the power battery SoC, the control variable is the net power of the fuel cell system, the initial state is a fixed value, there are multiple sets of values for the final state, and the constraint conditions include state variable constraints, control variable constraints, and power battery output power constraints;

[0009] The optimal data set described in step S2 is specifically as follows:

[0010] The independent variables are vehicle speed, acceleration, required power, driving mileage ratio, SoC, and target SoC, and the dependent variable is the net power of the fuel cell system.

[0011] S3. Train a neural network based on the optimal data set;

[0012] Further, the neural network described in step S3 is specifically as follows:

[0013] The neural network is a classification network with three hidden layers, the activation function is ReLU, and the optimization algorithm is Adam.

[0014] S4. Based on the monotonic relationship between the final value and the target value of the state of charge (SoC) of the power battery under the neural network energy management strategy, the dichotomy method is used to target the final SoC under the test conditions until it reaches the expected value.

[0015] Further, the dichotomy method for targeting the final SoC under the test conditions described in step S4 specifically includes the following steps:

[0016] S41. Initialize the number of iterations, the minimum value and the maximum value of the target SoC;

[0017] S42. Take the target SoC as the average of its minimum value and maximum value and use it as an input to the neural network.

[0018] S43. Perform energy management using the trained neural network under the test conditions to obtain the final SoC, calculate the difference dSoC between the final SoC and its expected value, and increment the number of iterations by one.

[0019] S44. If dSoC meets the error requirement or the number of iterations reaches the set value, the targeting ends; otherwise, update the minimum value and the maximum value of the target SoC, and repeat steps S42 and S43 until dSoC meets the error requirement or the number of iterations reaches the set value.

[0020] Compared with the prior art, the beneficial effects of the present invention are:

[0021] (1) The energy management method based on neural network proposed by the present invention can adjust the network output, with a simple structure and easy to implement;

[0022] (2) The energy management method based on neural network targeting proposed by the present invention can accurately control the final SoC of the battery and achieve approximately optimal energy consumption economy with high computational efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 is a flowchart of the energy management method for a fuel cell vehicle based on neural network targeting proposed by the present invention;

[0024] Figure 2 is the relationship between the final value and the target value of the SoC of the power battery under the HWFET condition in the present invention.

[0025] Figure 3 is a flowchart of the targeting method in the present invention. Detailed Embodiment

[0026] The following describes in detail the specific embodiments of the present invention in conjunction with the accompanying drawings, but the protection scope of the present invention is not limited to the following description.

[0027] As Figure 1 shown, a fuel cell vehicle energy management method based on neural network target shooting includes the following steps:

[0028] S1. Establish a power transmission system model for a fuel cell vehicle

[0029] The power transmission system model of a fuel cell vehicle includes a vehicle longitudinal dynamics model, a motor model, and an energy source power balance model.

[0030] S11. Establish a vehicle longitudinal dynamics model

[0031] Vehicle longitudinal dynamics is as shown in formula (1):

[0032]

[0033] Among them, a, v, M, f r and A respectively represent the vehicle acceleration, vehicle speed, mass, rolling resistance coefficient, and frontal area; F drv and F brk respectively represent the motor mechanical force and brake pad braking force at the wheels; G represents the acceleration due to gravity; θ represents the road gradient; ρ and C D respectively represent the air density and air resistance coefficient.

[0034] S12. Establish a motor model

[0035] The motor speed ω mot and torque T mot , as shown in formulas (2) and (3):

[0036]

[0037]

[0038] Among them, r whl is the tire rolling radius, i FD and η FD are respectively the transmission ratio and efficiency of the main reducer.

[0039] The electric power P mot,e of the motor, that is, the required power P dmd , is as shown in formula (4):

[0040]

[0041] Among them, ηmot Represents the motor efficiency related to ω mot and T mot where sgn is the sign function.

[0042] S13. Establish an energy source power balance model

[0043] The energy source includes a fuel cell and a power battery. The hydrogen consumption rate of the fuel cell system can be expressed as a function of the net power P fcs of the fuel cell system, as shown in Equation (5):

[0044]

[0045] The power battery adopts an equivalent circuit model, as shown in Equation (6):

[0046]

[0047] where V bat , V OC , I bat , R0 and P bat represent the power battery voltage, open circuit voltage, current, internal resistance, and output power, respectively. V OC and R0 are both functions of the state of charge SoC of the power battery.

[0048] The power battery current I bat , as shown in Equation (7):

[0049]

[0050] The battery system dynamics, as shown in Equation (8):

[0051]

[0052] where Q bat represents the battery capacity.

[0053] The energy source power balance model, as shown in Equation (9):

[0054]

[0055] where η DC / AC and η DC / DC represent the efficiencies of the DC / AC inverter and the DC / DC converter, respectively.

[0056] S2. Solve the optimal energy management problem of a fuel cell vehicle using dynamic programming and establish an optimal data set based on the solution results

[0057] S21. Establish the optimal energy management problem of a fuel cell vehicle

[0058] The optimization objective of the optimal energy management problem for a fuel cell vehicle is to minimize the total hydrogen consumption. The state variable is the SoC of the power battery, the control variable is the net power of the fuel cell system, the initial state is a fixed value, and there are multiple sets of values for the final state. The constraint conditions include state variable constraints, control variable constraints, and power battery output power constraints, as shown in Equation (10):

[0059]

[0060] where t f is the driving cycle duration, SoC0 is the initial SoC, SoC f is the final SoC, and the subscripts min and max represent the minimum and maximum values of the corresponding variables, respectively.

[0061] S22. Establish the optimal data set based on the results of dynamic programming

[0062] The independent variables of the optimal data set are vehicle speed, acceleration, required power, driving range ratio, SoC, and target SoC, and the dependent variable is the net power of the fuel cell system, as shown in Equation (11):

[0063]

[0064] where X is the independent variable, y is the dependent variable, is the driving range ratio,, SoC tgt is the target SoC, the superscript l represents the driving cycle index, and the subscript t represents time. Here SoC tgt = SoC f , because dynamic programming can accurately control the final SoC to reach the expected value.

[0065] S3. Train the neural network based on the optimal data set

[0066] The neural network is a classification network with three hidden layers, the activation function is ReLU, and the optimization algorithm is Adam.

[0067] S4. Based on the monotonic relationship between the final value and the target value of the state of charge SoC of the power battery under the neural network energy management strategy, use the bisection method to target the final SoC under the test driving cycle until it reaches the expected value

[0068] As Figure 2 shown, the relationship between the final value and the target value of the SoC of the power battery under the HWFET driving cycle shows that the neural network controller can accurately control the final SoC through the targeting method.

[0069] As Figure 3As shown, it is a flowchart of the target shooting method. Under the test conditions, the bisection method is used to shoot the target for the final SoC, which specifically includes the following steps:

[0070] S41. Initialize the iteration number Iter, the minimum value SoC of the target SoC tgt,min and the maximum value SoC tgt,max ;

[0071] S42. Take the target SoC as the average of its minimum value and maximum value and use it as an input to the neural network;

[0072] S43. Under the test conditions, use the trained neural network f NN to complete energy management, obtain the final SoC, calculate the difference dSoC between the final SoC and its expected value, and increment the iteration number by one;

[0073] S44. If dSoC meets the error requirement τ or the iteration number reaches the set value, the target shooting ends; otherwise, update the minimum value and maximum value of the target SoC, and repeat steps S42 and S43 until dSoC meets the error requirement τ or the iteration number reaches the set value.

[0074] The above is the preferred implementation manner of the present invention, and it should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Modifications and changes made by those of ordinary skill in the art based on the core idea of the present invention should fall within the protection scope of the appended claims of the present invention.

Claims

1. A fuel cell vehicle energy management method based on neural network target shooting, characterized in that, Including the following steps: S1. Establish a power transmission system model for a fuel cell vehicle, including a vehicle longitudinal dynamics model, a motor model, and an energy source power balance model; S2. Solve the optimal energy management problem of the fuel cell vehicle using dynamic programming, and establish an optimal data set according to the solution results; The optimal energy management problem of the fuel cell vehicle is specifically as follows: The optimization objective is to minimize the total hydrogen consumption. The state variable is the SoC of the power battery, the control variable is the net power of the fuel cell system, the initial state is a fixed value, and there are multiple sets of values for the final state. The constraint conditions include state variable constraints, control variable constraints, and power battery output power constraints; The optimal data set described in step S2 is specifically as follows: The independent variables are vehicle speed, acceleration, required power, driving mileage ratio, SoC, and target SoC, and the dependent variable is the net power of the fuel cell system; S3. Train a neural network based on the optimal data set; S4. Based on the monotonic relationship between the final value and the target value of the state of charge (SoC) of the power battery under the neural network energy management strategy, use the bisection method to target the final SoC under the test conditions until it reaches the expected value; specifically including the following steps: S41. Initialize the number of iterations, the minimum value and the maximum value of the target SoC; S42. Take the target SoC as the average of its minimum value and maximum value and use it as an input to the neural network; S43. Perform energy management using the trained neural network under the test conditions to obtain the final SoC and calculate the difference dSoC between the final SoC and its expected value, and increment the number of iterations by one; S44. If dSoC meets the error requirement or the number of iterations reaches the set value, the targeting ends; otherwise, update the minimum value and the maximum value of the target SoC, and repeat steps S42 and S43 until dSoC meets the error requirement or the number of iterations reaches the set value.

2. The fuel cell vehicle energy management method based on neural network target shooting according to claim 1, characterized in that: The neural network described in step S3 is specifically as follows: The neural network is a classification network with three hidden layers, the activation function is ReLU, and the optimization algorithm is Adam.

Images