A method for energy management control of FCHEV based on multi-objective dynamic programming neural network

Abstract

The application relates to an FCHEV energy management control method based on a multi-target dynamic programming neural network, which comprises the following steps: constructing a vehicle dynamics model; establishing a dynamic programming algorithm with the optimal operation cost as a target; inputting standard working condition data into the vehicle dynamics model, combining the dynamic programming algorithm to solve, obtaining corresponding optimal fuel cell and energy storage device output power under different driving conditions, and constructing an optimal data set; training a neural network by using the optimal data set to obtain a trained network model; inputting actual working condition data into the trained network model, combining a Kalman filter, and outputting fuel cell and energy storage device power distribution results; and controlling the working states of the fuel cell and the energy storage device on the automobile according to the fuel cell and energy storage device power distribution results. Compared with the prior art, the application designs a new energy management strategy, which can effectively balance the economy, durability and real-time performance of the fuel cell vehicle.

Description

TECHNICAL FIELD

[0001] The present application relates to fuel cell vehicle energy management control, and particularly to a FCHEV energy management control method based on multi-objective dynamic programming neural network. BACKGROUND

[0002] Fuel cell vehicles (FCHEV) have become one of the main development objects of new energy vehicles due to their advantages of no pollution, high energy conversion rate, no need for charging, etc. However, fuel cells also have their shortcomings, such as poor cold start performance, slow system response, high cost, short service life, etc. In order to overcome these shortcomings, fuel cells need to work with other energy storage devices, such as batteries and supercapacitors. In the conventional working condition, the fuel cell supplies energy, and in the severe working condition of high load change, start-stop, etc., the corresponding energy storage device supplies energy. Considering the multi-energy matching of fuel cell vehicles, the load management and energy output among multiple energy sources are the focus of current design and research.

[0003] The existing energy management strategies have global optimization algorithms based on dynamic programming, Pontryagin minimum principle, etc. However, the above algorithms need to obtain global prior information in advance, and the calculation amount is large, which does not have real-time performance. The emerging neural network technology can be used for real-time energy management of fuel cell vehicles, and the neural network trained by the dynamic programming algorithm can better track the results of the dynamic programming algorithm.

[0004] However, the objective function of the current global optimization algorithm is mainly the fuel economy, and the impact of start-stop and frequent load change on the durability of the fuel cell during vehicle driving is not considered. In practical applications, the fuel cell vehicle cannot well balance the economy, durability and real-time performance. SUMMARY

[0005] The purpose of the present application is to overcome the defects of the prior art and provide a FCHEV energy management control method based on multi-objective dynamic programming neural network, which can balance the economy, durability and real-time performance of the fuel cell vehicle.

[0006] The purpose of the present application can be achieved by the following technical scheme: a FCHEV energy management control method based on multi-objective dynamic programming neural network, comprising the following steps:

[0007] S1, constructing a vehicle dynamics model;

[0008] S2, establishing a dynamic programming algorithm with optimal operating cost as the target;

[0009] S3, input the standard working condition data into the vehicle dynamics model, and solve by combining the dynamic programming algorithm to obtain the optimal fuel cell and energy storage device output power corresponding to different driving conditions, so as to build an optimal data set;

[0010] S4, train the neural network using the optimal data set to obtain a trained network model;

[0011] S5, input the actual working condition data into the trained network model, and combine the Kalman filter to output the fuel cell and energy storage device power distribution results;

[0012] S6, according to the fuel cell and energy storage device power distribution results, the working state of the fuel cell and energy storage device on the automobile is controlled.

[0013] Further, the vehicle dynamics model in step S1 includes a vehicle transmission model, a fuel cell model, an energy storage model, and a motor model.

[0014] Further, the vehicle transmission model adopts a longitudinal dynamics model, specifically:

[0015]

[0016] Wherein, P is the demand power of the vehicle, η T is the efficiency of the transmission system, m is the mass of the vehicle, g is the acceleration of gravity, f is the rolling resistance coefficient, c D is the air resistance coefficient, A is the wind area, u is the speed of the vehicle without wind, δ is the rotational mass conversion coefficient of the vehicle, and α is the road slope.

[0017] Further, the fuel cell model adopts a net output power and efficiency model, specifically:

[0018] η=f(P fc )

[0019] Wherein, P fc is the net output power of the fuel cell, and η is the efficiency of the fuel cell.

[0020] Further, the energy storage model adopts an internal resistance model, specifically:

[0021]

[0022]

[0023] Wherein, I bat is the current of the energy storage device, U oc is the open circuit voltage of the energy storage device, R bat is the internal resistance of the energy storage device, and P batFor the output power of the energy storage device, when P bat > 0, it indicates that the energy storage device is discharging, R bat is the discharging resistance; when P bat < 0, it indicates that the energy storage device is charging, R bat is the charging resistance, SOC(t) is the SOC (State of Charge) of the energy storage device, η C is the coulomb efficiency of the energy storage device; C bat is the capacity of the energy storage device.

[0024] Further, the motor model is specifically:

[0025]

[0026] Wherein, P m is the demand power of the motor, T m is the output torque of the motor, ω is the motor speed, and η M is the efficiency of the motor.

[0027] Further, the dynamic programming algorithm in the step S2 includes a target function and a constraint condition, the target function is the lowest running cost of hydrogen consumption and fuel cell attenuation in four typical working conditions of start-stop, variable load, idle and full load;

[0028] The constraint condition includes the net output power constraint of the fuel cell, the output power constraint of the energy storage device, and the SOC constraint of the energy storage device.

[0029] Further, the target function is specifically:

[0030]

[0031]

[0032]

[0033] Wherein, C is the FCHEV running cost, is the price of each kilogram of hydrogen, C fc is the fuel cell cost, m is the consumption of hydrogen, and DEG is the attenuation degree of the fuel cell;

[0034] is the molar mass of hydrogen; I fc is the fuel cell stack current; n is the number of fuel cell monomers; F is the Faraday constant; H2 is the low heat value of hydrogen, P fc is the output power of the fuel cell; η fc (P fc ) is the efficiency of the fuel cell under the current fuel cell power;

[0035] p 1 1,p ′ 2,p ′ 3, ′ 4 respectively are the decay rates of four typical operating conditions start-stop, variable load, idle, full load, n is the number of start-stop, Δt is the simulation step, Δt low , Δt high respectively are the durations of idle and full load, fc,-1 is the power of the fuel cell at the previous time, P fc, is the power of the fuel cell at the current time, P rate is the rated power of the fuel cell, P low is the power of the fuel cell at idle.

[0036] Further, the constraint condition is specifically:

[0037]

[0038] wherein, P fc, , P fc, are the minimum and maximum net output powers of the fuel cell, P b,min , P b,max are the minimum and maximum output powers of the energy storage device, SOC min , SOC max are the upper and lower boundary values of the SOC of the energy storage device.

[0039] Further, the network model trained in the step S4 comprises a three-layer network structure: an input layer, a hidden layer and an output layer, wherein the input data of the input layer are speed, acceleration, demand power, average speed and integral of demand power; the hidden layer is designed as 10 neurons, and the output of the output layer is the power distribution of the fuel cell and the energy storage device, that is, the output power of the fuel cell corresponding to a total power demand.

[0040] Compared with the prior art, the present application has the following advantages:

[0041] Firstly, the present application establishes a whole vehicle dynamics model, then constructs a dynamic programming algorithm with the optimal comprehensive operation cost as the target, selects different standard operating conditions to calculate the optimal output powers of the fuel cell and the storage battery under the corresponding driving conditions to establish the corresponding optimal data set, then trains the neural network using the optimal data set and adopts a Kalman filter to smooth the output results of the neural network model to eliminate noise, so that the optimization in economy and durability can be realized, and the real-time performance of energy management can be met.

[0042] Second, when establishing the dynamic programming algorithm in this invention, the operating cost function of hydrogen consumption and fuel cell degradation under four typical operating conditions is used as the basis. It can comprehensively consider fuel economy and battery durability. Then, the state variable is selected as SOC and the control variable is the output power of the fuel cell. Constraints such as SOC and fuel cell output power range are established. This can realize a multi-objective dynamic programming algorithm, which is beneficial to obtain the optimal dataset by combining different standard operating conditions, thereby ensuring the accuracy of neural network training.

[0043] Third, in this invention, the standard operating conditions include start-stop, variable load, idling, and full load, which can fully consider the impact of start-stop and frequent load changes during vehicle driving on the durability of fuel cells. Attached Figure Description

[0044] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0045] Figure 2 This is a schematic diagram illustrating the application process of an example.

[0046] Figure 3 This is a schematic diagram of the fuel cell vehicle powertrain system in the embodiment;

[0047] Figure 4 This is a schematic diagram illustrating the process of using dynamic programming to obtain the optimal dataset. Detailed Implementation

[0048] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0049] Example

[0050] like Figure 1 As shown, an FCHEV energy management control method based on a multi-objective dynamic programming neural network includes the following steps:

[0051] S1. Construct the vehicle dynamics model;

[0052] S2. Establish a dynamic programming algorithm with the goal of optimizing operating cost;

[0053] S3. Input the standard operating condition data into the vehicle dynamics model and solve it using dynamic programming algorithm to obtain the optimal output power of fuel cell and energy storage device under different driving conditions, thereby constructing the optimal dataset;

[0054] S4. Train the neural network using the optimal dataset to obtain the trained network model;

[0055] S5. Input the actual operating condition data into the trained network model and combine it with the Kalman filter to output the power distribution results of the fuel cell and energy storage device.

[0056] S6、According to the power distribution result of the fuel cell and the energy storage device, the working state of the fuel cell and the energy storage device on the automobile is controlled correspondingly.

[0057] The above technical solution is applied to the actual situation, as shown in Figure 2 .

[0058] Step 1, a vehicle dynamics model is established, which contains a driving cycle information input module, a vehicle transmission system, a motor system, a fuel cell system, an energy storage system (in this embodiment, a storage battery is used as the energy storage device), a DC / DC, etc.

[0059] Step 2, a target function of a dynamic programming algorithm is established based on a hydrogen consumption, a fuel cell running cost function in four typical working conditions, and a decay. At the same time, a state variable is selected as SOC, a control variable is selected as the output power of the fuel cell, and a constraint condition such as a SOC and a fuel cell output power range is established.

[0060] Step 3, different standard working conditions are selected as the driving cycle input into the vehicle dynamics model, the required power is solved by using the dynamic programming algorithm, the optimal output power of the fuel cell and the storage battery under the corresponding driving condition is calculated, and the corresponding optimal data set is established.

[0061] Step 4, a BP neural network is trained by using the optimal data set. The neural network includes an input layer, a hidden layer and an output layer. The hidden layer is designed to have 10 neurons. The structure parameters, learning rate and activation function of the neural network can be adjusted as needed to optimize the network performance.

[0062] Step 5, finally, the output result of the trained neural network enters a Kalman filter, and the output result of the neural network is smoothed. The Kalman filter takes the output result of the neural network as the input, and outputs a more smoothed fuel cell output power curve. The related parameters of the Kalman filter can be adjusted to obtain fuel cell output power with different smoothing degrees.

[0063] Specifically, in step 1, as shown in Figure 3 , a vehicle longitudinal dynamics model is established as follows:

[0064]

[0065] In the formula, P is the demand power of the vehicle, η T is the efficiency of the transmission system, m is the mass of the vehicle, g is the acceleration of gravity, f is the rolling resistance coefficient, c D is the air resistance coefficient, A is the windward area, u is the driving speed of the vehicle without wind, δ is the rotational mass conversion coefficient of the vehicle, and α is the road slope.

[0066] The established motor model is an equivalent model as follows:

[0067]

[0068] where P m is the motor demand power, T m is the motor output torque, ω is the motor speed, and η M is the motor efficiency. The fuel cell adopts a net output power and efficiency model

[0069] η = f (P fc )

[0070] The battery adopts an internal resistance model as follows:

[0071]

[0072] where I bat is the battery current, U oc is the open circuit voltage of the battery, R bat is the internal resistance of the battery, and P bat is the output power of the battery. When P bat > 0, it indicates that the battery is discharging, and R bat is the discharging resistance; when P bat < 0, it indicates that the battery is charging, and R bat is the charging resistance.

[0073] Based on the ampere-hour integral method, the SOC of the battery is calculated as follows:

[0074]

[0075] where η C is the coulombic efficiency of the battery, and C bat is the capacity of the battery.

[0076] The DC / DC converter only considers its efficiency, and does not consider its dynamic response capability.

[0077] In step 2, the cost-based objective function is defined as follows:

[0078]

[0079] C is the operating cost of the FCHEV, C fc is the price of each kilogram of hydrogen, C fc is the fuel cell cost, m is the consumption of hydrogen, and DEG is the attenuation degree of the fuel cell. The calculation method is as follows

[0080]

[0081] where m is the consumption of hydrogen. I is the molar mass of hydrogen gas. fc is the fuel cell stack current; n is the number of fuel cell cells; F is the Faraday constant; H2 Because of the low calorific value of hydrogen, P fc η is the output power of the fuel cell. fc (P fc () represents the efficiency of the fuel cell at the current fuel cell power.

[0082]

[0083] Where p ′ 1, p ′ 2, p ′ 3, ′ 4 represents the attenuation rate under four typical operating conditions: start-stop, variable load, idling, and full load; n represents the number of start-stop cycles; and Δt represents the simulation step size. low , Δt high These represent the durations of idling and full load, respectively. fc,-1 P represents the power of the fuel cell at the previous moment. fc, Let P be the power of the fuel cell at the current moment. rate P is the rated power of the fuel cell. low This represents the power output of the fuel cell at idle speed.

[0084] The constraints of the dynamic programming algorithm are as follows:

[0085]

[0086] Where P fc,min ,P fc,max P represents the minimum and maximum net output power of the fuel cell. b,min ,P b,max This refers to the minimum and maximum output power of the battery. SOC min SOC max These are the upper and lower boundary values ​​of the SOC.

[0087] In step 3, as Figure 4 As shown, different operating conditions are selected as inputs, and the results are used for model simulation to obtain dynamic programming results. The operating conditions are then input into the dynamic programming algorithm to obtain the optimal solution. The state transition equation of the dynamic programming algorithm is as follows:

[0088]

[0089] In step 4, the inputs to the neural network are velocity, acceleration, power demand, average velocity, and the integral of power demand; the output is the power allocation between the fuel cell and the battery, that is, how much power the fuel cell should output under a total power demand.

[0090] In conclusion, in the technical scheme, the objective function of the dynamic programming algorithm comprehensively considers the economy of fuel and the durability of the fuel cell, and the training set with optimal economy and durability is obtained through the multi-objective dynamic programming algorithm, so that the neural network trained thereby is more optimal in considering the comprehensive operation cost of the fuel cell. Meanwhile, the neural network is adopted as the vehicle energy management strategy, and the real-time performance is considered, so that the vehicle demand power is more rapidly responded to real-time changes. In addition, the Kalman filter is used to optimize the output result of the neural network, to optimize the fluctuation of the neural network, and to eliminate the noise of the neural network output, so that the output of the neural network is closer to the optimal comprehensive operation cost.

Claims

1. A FCHEV energy management control method based on a multi-objective dynamic programming neural network, characterized in that, Includes the following steps: S1. Construct the vehicle dynamics model; S2. Establish a dynamic programming algorithm with the goal of optimizing operating cost; S3. Input the standard operating condition data into the vehicle dynamics model and solve it using dynamic programming algorithm to obtain the optimal output power of fuel cell and energy storage device under different driving conditions, thereby constructing the optimal dataset; S4. Train the neural network using the optimal dataset to obtain the trained network model; S5. Input the actual operating condition data into the trained network model and combine it with the Kalman filter to output the power distribution results of the fuel cell and energy storage device. S6. Control the operating status of the fuel cell and energy storage device on the vehicle according to the power distribution results of the fuel cell and energy storage device. The dynamic programming algorithm in step S2 includes an objective function and constraints. The objective function is to minimize the operating cost of hydrogen consumption and fuel cell degradation under four typical operating conditions: start-stop, variable load, idling, and full load. The constraints include the net output power constraint of the fuel cell, the output power constraint of the energy storage device, and the SOC constraint of the energy storage device. The objective function is specifically: ， ， ， in, For FCHEV operating costs, The price per kilogram of hydrogen. For the cost of fuel cells, m This represents the amount of hydrogen consumed. DEG The degree of degradation of the fuel cell; is the molar mass of hydrogen gas; This refers to the current in the fuel cell stack. This represents the number of individual fuel cell units. It is Faraday's constant; Because of the low calorific value of hydrogen, This refers to the output power of the fuel cell; This represents the efficiency of the fuel cell at the current fuel cell power level. , , The attenuation rates are for four typical operating conditions: start-stop, variable load, idling, and full load. n For the number of start-stop cycles, For simulating step size, , These represent the durations of idling and full load, respectively. This represents the power of the fuel cell at the previous moment. The power of the fuel cell at the current moment. This refers to the rated power of the fuel cell. This represents the power output of the fuel cell at idle speed.

2. The FCHEV energy management control method based on a multi-objective dynamic programming neural network according to claim 1, characterized in that, The vehicle dynamics model in step S1 includes a vehicle transmission model, a fuel cell model, an energy storage model, and a motor model.

3. The FCHEV energy management control method based on a multi-objective dynamic programming neural network according to claim 2, characterized in that, The vehicle transmission model adopts a longitudinal dynamics model, specifically: ， in, For the power required by the whole vehicle, For the efficiency of the transmission system. For the overall vehicle quality, It is the acceleration due to gravity. The rolling resistance coefficient, The air drag coefficient, For windward area, The speed of the car when there is no wind. This is the conversion factor for the rotational mass of a vehicle. This refers to the road slope.

4. The FCHEV energy management control method based on a multi-objective dynamic programming neural network according to claim 2, characterized in that, The fuel cell model adopts a net output power and efficiency model, specifically: ， in, This refers to the net output power of the fuel cell. For the efficiency of fuel cells.

5. The FCHEV energy management control method based on a multi-objective dynamic programming neural network according to claim 2, characterized in that, The energy storage model adopts an internal resistance model, specifically: ， ， in, For energy storage devices, Open-circuit voltage for energy storage devices; The internal resistance of the energy storage device; For the output power of energy storage devices, when When, it indicates that the energy storage device is discharging. For discharge resistance; when When, it indicates that the energy storage device is charging. For charging resistors, For the SOC of energy storage devices, Coulomb efficiency of energy storage devices; This refers to the capacity of the energy storage device.

6. The FCHEV energy management control method based on a multi-objective dynamic programming neural network according to claim 2, characterized in that, The motor model is specifically as follows: ， in, For the required power of the motor, For the motor output torque, This refers to the motor speed. For motor efficiency.

7. The FCHEV energy management control method based on a multi-objective dynamic programming neural network according to claim 1, characterized in that, The specific constraints are as follows: ， in, For the minimum and maximum net output power of the fuel cell, The minimum and maximum output power of the energy storage device, For energy storage devices The upper and lower boundary values.

8. The FCHEV energy management control method based on a multi-objective dynamic programming neural network according to claim 1, characterized in that, The network model trained in step S4 includes a three-layer network structure: an input layer, a hidden layer, and an output layer. The input data of the input layer are velocity, acceleration, power demand, average velocity, and the integral of the power demand. The hidden layer is designed with 10 neurons. The output of the output layer is the power allocation between the fuel cell and the energy storage device, that is, the output power of the fuel cell under a total power demand.

Images