Energy management method for fuel cell vehicles based on predictive nonlinear model control

Abstract

The present invention relates to a fuel cell vehicle energy management method based on predictive nonlinear model control, comprising: establishing a predictive model based on an optimized Markov chain based on historical vehicle motion information; obtaining the vehicle's current motion information, online predicting the vehicle's operating state within a limited future time domain based on the predictive model, and adaptively updating the predictive model; establishing a fuel cell vehicle power system model, using the vehicle's operating state within the predicted time domain as input and outputting a vehicle power demand sequence; using the vehicle power demand sequence within the predicted time domain as input to a dynamic programming algorithm that introduces a variable load penalty coefficient and outputting an optimal power allocation sequence; applying the optimal power allocation sequence to the vehicle, and then obtaining the vehicle's current motion information again at the next moment, re-predicting and optimizing until the vehicle stops. Compared with existing technologies, the present invention has the advantages of achieving a balance between fuel cell vehicle economy and durability.

Description

Technical Field

[0001] The present invention relates to the field of energy management strategies, and in particular to an energy management method for fuel cell vehicles based on predictive nonlinear model control. Background Art

[0002] Energy management strategy is one of the core technologies in fuel cell vehicle development. An excellent energy management strategy can fully leverage the advantages of a fuel cell vehicle's various power sources, thereby achieving optimal vehicle performance. Currently, the main energy management strategies can be categorized as rule-based and optimization-based.

[0003] Rule-based energy management strategies are simple to control and are unaffected by input conditions. They are among the earliest control methods used in fuel cell vehicles. However, due to their simple rules, they cannot achieve optimal performance.

[0004] Optimization-based energy management strategies can be broadly categorized into two types: globally optimal and locally optimal. Globally optimal energy management strategies, due to their high computational complexity, cannot be applied to real-time energy management strategies. Locally optimal algorithms primarily include energy management methods based on minimizing equivalent fuel consumption, neural networks, or model predictive control. Model predictive control, which transforms global optimality into local optimality through online prediction, exhibits strong robustness and is well-suited for hybrid power system energy management. However, achieving both more accurate predictions and precise control during the rolling optimization process to achieve a balance between fuel cell vehicle economy and durability remains a challenging task. Summary of the Invention

[0005] The purpose of the present invention is to provide a fuel cell vehicle energy management method based on a predictive nonlinear model control. The method uses an optimized Markov chain prediction method, reasonably selects the prediction time interval and adopts a multi-step prediction method rather than a single-step prediction method to ensure the accuracy of the prediction. During the variable working conditions, a simple uniform speed prediction method is used and combined with the update of the working area to make the prediction model more accurate and have a certain degree of adaptability. Finally, by introducing a dynamic programming algorithm that considers variable load penalties within the prediction time domain, the fuel cell vehicle's economy and durability are made more excellent.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] A fuel cell vehicle energy management method based on predictive nonlinear model control comprises the following steps:

[0008] S1: Based on the historical movement information of the vehicle, a prediction model based on the optimized Markov chain is established;

[0009] S2: During vehicle travel, the current vehicle motion information is obtained. Based on the current motion information, the vehicle's operating state within a limited future time domain is predicted online using a prediction model, and the prediction model is adaptively updated.

[0010] S3: Establish a fuel cell vehicle power system model, use the vehicle operating status in the prediction time domain obtained by the prediction model as input, and output the vehicle demand power sequence in the prediction time domain;

[0011] S4: The vehicle demand power sequence within the forecast horizon is used as the input of a dynamic programming algorithm that introduces a variable load penalty coefficient, and the optimal power allocation sequence within the finite forecast horizon is output;

[0012] S5: Apply the optimal power allocation sequence within the limited prediction time domain to the vehicle, obtain the vehicle's current motion information again at the next moment, return to step S2 to re-predict and perform optimization of S3-S4, and repeat the optimization until the vehicle stops.

[0013] Said S1 comprises the following steps:

[0014] S11: Analyze the historical vehicle movement information offline, select different prediction time domains to analyze the Markov characteristics of the running state, and select the optimal prediction time domain;

[0015] S12: Divide the motion state space according to the vehicle speed and acceleration in the vehicle motion information, establish a set of operating points, and obtain the initial state transition matrix through offline learning. The state transition probability calculation formula is as follows:

[0016]

[0017] in: is the state transition probability, is the number of states that transition from state i to state j after m time intervals, It represents the sum of the number of transitions from state i to each state after m time intervals;

[0018] S13: Update the state transfer matrix according to the root mean square error of the prediction results, and select multi-step state transfer matrices to construct a multi-scale prediction model based on the optimized Markov chain.

[0019] The fuel cell vehicle power system model includes a fuel cell model, a motor model, a battery model and a transmission system model.

[0020] The fuel cell model adopts the basic efficiency-power curve model:

[0021] η=f(P fc )

[0022] Where η is the fuel cell efficiency, P fc Output power for the fuel cell.

[0023] The motor model is a MAP diagram model established according to the selected parameters:

[0024] η m =f(n m , T m )

[0025] Among them, η m is the motor efficiency, n m is the motor speed, T m is the motor torque.

[0026] The battery adopts the internal resistance model, and the calculation is as follows:

[0027]

[0028] Among them, I bat is the battery current, V oc is the open circuit voltage of the battery, R bat is the battery resistance, P bat is the battery output power, and the battery state of charge SOC is calculated as:

[0029]

[0030] Among them, C cap is the capacity of the battery.

[0031] The transmission system model is:

[0032]

[0033] Among them, P is the required 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 frontal area, u is the speed of the car in no wind, δ is the conversion coefficient of the car's rotational mass, and α is the road slope.

[0034] The objective function of the dynamic programming algorithm with the introduction of the variable load penalty coefficient is:

[0035]

[0036] Where m is the hydrogen consumption, λ is the load penalty coefficient, which is used to balance the durability of the fuel cell and battery, and D load-change is the difference between the fuel cell output at the previous moment and the current moment, and N is the size of the prediction time domain.

[0037] The calculation method of the hydrogen consumption is:

[0038]

[0039] in, is the molar mass of hydrogen, I fc is the fuel cell stack current, n is the number of fuel cell units, F is the Faraday constant, E H2 is the lower heating value of hydrogen, η is the fuel cell efficiency, P fc Output power for the fuel cell.

[0040] The constraint condition of the dynamic programming algorithm that introduces the load penalty coefficient is to predict the possible change value in the time domain space:

[0041]

[0042] Among them, P fc For the fuel cell output power, please add parameter explanation P b , SOC is the state of charge of the battery, P fc-t，min 、P b-t，min , SOC t-min is the preset lower limit of the corresponding parameter, P fc-t，max 、P b-t，max , SOC t-max is the preset upper limit of the corresponding parameter.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] (1) The present invention improves the accuracy, stability and adaptive performance of the prediction model by improving the prediction model and reasonably selecting the prediction time span, adopting the form of multi-step prediction, and using a simple uniform speed prediction method during the variable working conditions.

[0045] (2) The present invention improves the robustness and real-time performance of the fuel cell output power of a fuel cell vehicle through continuously rolling updated prediction and optimal control.

[0046] (3) The present invention balances the pursuit of durability and economy of fuel cell vehicles within a limited time domain by introducing a dynamic programming algorithm with a variable load penalty coefficient, and the performance approaches the global optimum. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 Schematic diagram of the method flow of the present invention;

[0048] Figure 2 A prediction flow chart of a prediction model for optimizing a Markov chain in an embodiment;

[0049] Figure 3Schematic diagram of the updating process of the state transfer matrix in one embodiment;

[0050] Figure 4 Schematic diagram of a vehicle topology structure in an embodiment. DETAILED DESCRIPTION

[0051] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0052] This embodiment provides a fuel cell vehicle energy management method based on predictive nonlinear model control, such as Figure 1 As shown, the following steps are included:

[0053] S1: Based on the historical movement information of the vehicle, a prediction model based on the optimized Markov chain is established, such as Figure 2 This is shown in the offline learning section.

[0054] S11: Analyze the historical vehicle movement information offline, select different prediction time domains to analyze the Markov characteristics of the running state, and select the optimal prediction time domain;

[0055] S12: Divide the motion state space according to the vehicle speed and acceleration in the vehicle motion information, establish a set of operating points, and obtain the initial state transition matrix through offline learning. The state transition probability calculation formula is as follows:

[0056]

[0057] in: is the state transition probability, is the number of states that transition from state i to state j after m time intervals, It represents the sum of the number of transitions from state i to each state after m time intervals;

[0058] S13: Update the state transfer matrix according to the root mean square error of the prediction results, and select multi-step state transfer matrices to construct a multi-scale prediction model based on the optimized Markov chain.

[0059] In this embodiment, the transition probability of each state under different time scales of the cyclic working condition is obtained statistically to establish a state transfer matrix. That is, if the prediction time domain is n seconds, an n-step transfer matrix is ​​used instead of the iteration of a single-step transfer matrix.

[0060] The update process of the state transfer matrix is ​​as follows Figure 3 As shown. For the current state j at time k, the state i at time km is known, then the state transfer matrix is ​​calculated according to the formula to update.

[0061] To reduce prediction distortion when the transfer matrix is ​​immature, a "uniform speed prediction" approach is used to improve prediction accuracy. Specifically, when the matrix is ​​immature and the variance between the predicted speed and the actual speed is too large, the vehicle is assumed to be traveling at a uniform speed for the next period of the prediction.

[0062] S2: During the vehicle's driving process, the current motion information of the vehicle is obtained, and based on the current motion information, the vehicle's operating status in the future limited time domain is predicted online based on the prediction model, and the prediction model is adaptively updated, such as Figure 2 as shown in the Online Update section.

[0063] S3: Establish a fuel cell vehicle power system model, use the vehicle operating status in the prediction time domain obtained by the prediction model as input, and output the vehicle demand power sequence in the prediction time domain.

[0064] The fuel cell vehicle power system model includes fuel cell model, motor model, battery model and transmission system model, such as Figure 4 As shown, the fuel cell stack receives hydrogen from the hydrogen tank and oxygen from the air. The reaction generates electricity, which is fed through the DC / DC converter and then into the high-voltage bus. The drive motor converts the electricity into mechanical energy. The excess energy also charges the battery, which serves as energy storage and can cope with high load demands.

[0065] Among them, the fuel cell model adopts the basic efficiency-power curve model:

[0066] η=f(P fc )

[0067] Where η is the fuel cell efficiency, P fc Output power for the fuel cell.

[0068] The motor model is a MAP diagram model established based on the selected parameters:

[0069] η m =f(n m , T m )

[0070] Among them, η m is the motor efficiency, n m is the motor speed, T m is the motor torque.

[0071] The battery adopts the internal resistance model, and the calculation is as follows:

[0072]

[0073] Among them, I bat is the battery current, Voc is the open circuit voltage of the battery, R bat is the battery resistance, P bat is the battery output power, and the battery state of charge SOC is calculated as:

[0074]

[0075] Among them, C cap is the capacity of the battery.

[0076] The transmission system model is:

[0077]

[0078] Among them, P is the required 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 frontal area, u is the speed of the car in no wind, δ is the conversion coefficient of the car's rotational mass, and α is the road slope.

[0079] S4: The vehicle demand power sequence within the prediction time domain is used as the input of the dynamic programming algorithm that introduces the variable load penalty coefficient, and the optimal power allocation sequence within the limited prediction time domain is output.

[0080] The objective function of the dynamic programming algorithm with the introduction of the variable load penalty coefficient is:

[0081]

[0082] Where m is the hydrogen consumption, λ is the load penalty coefficient, which is used to balance the durability of the fuel cell and battery, and D load-change is the difference between the fuel cell output at the previous moment and the current moment, and N is the size of the prediction time domain.

[0083] The calculation method of hydrogen consumption is:

[0084]

[0085] in, is the molar mass of hydrogen, I fc is the fuel cell stack current, n is the number of fuel cell units, F is the Faraday constant, E H2 is the lower heating value of hydrogen, η is the fuel cell efficiency, P fc Output power for the fuel cell.

[0086] The constraints of the dynamic programming algorithm that introduces the variable load penalty coefficient are to predict the possible change values ​​in the time domain space:

[0087]

[0088] Among them, P fc For the fuel cell output power, please add parameter explanation P b , SOC is the state of charge of the battery, P fc-t，min 、P b-t，min , SOC t-min is the preset lower limit of the corresponding parameter, P fc-t，max 、P b-t，max , SOC t-max is the preset upper limit of the corresponding parameter.

[0089] S5: Apply the optimal power allocation sequence within the limited prediction time domain to the vehicle, obtain the vehicle's current motion information again at the next moment, return to step S2 to re-predict and perform optimization of S3-S4, and repeat the optimization until the vehicle stops.

[0090] Table 1 compares the method of the present invention with the thermostat strategy and the power-following strategy. Compared to the thermostat strategy, the present invention achieves a better balance between economy and durability, although fuel cell degradation is slightly increased. However, battery degradation, hydrogen consumption per 100 kilometers, and operating costs per 100 kilometers are reduced, while the power-following strategy significantly reduces fuel cell degradation and operating costs per 100 kilometers. Although battery degradation and hydrogen consumption per 100 kilometers are slightly increased, overall performance is still improved.

[0091] Table 1

[0092]

[0093] The above describes in detail the preferred embodiments of the present invention. It should be understood that those skilled in the art can make numerous modifications and variations based on the concepts of the present invention without inventive effort. Therefore, any technical solutions that can be derived by those skilled in the art through logical analysis, reasoning, or limited experimentation based on the concepts of the present invention and the prior art should be within the scope of protection defined by the claims.

Claims

1. A fuel cell vehicle energy management method based on predictive nonlinear model control, characterized in that: The following steps are involved: S1: Based on the historical movement information of the vehicle, a prediction model based on the optimized Markov chain is established; Said S1 comprises the following steps: S11: Analyze the historical vehicle movement information offline, select different prediction time domains to analyze the Markov characteristics of the running state, and select the optimal prediction time domain; S12: Divide the motion state space according to the vehicle speed and acceleration in the vehicle motion information, establish a set of operating points, and obtain the initial state transition matrix through offline learning. The state transition probability calculation formula is as follows: in: is the state transition probability, From the state i Departure, passing m Time interval transfer to state The number of Indicates the slave state i Departure, passing m The sum of the number of states transferred to each state in a time interval is obtained; the transfer probability of each state under different time scales of the cycle condition is obtained statistically, and the state transfer matrix is ​​established, that is, the prediction time domain has Seconds, use Step transfer matrix; S13: Update the state transfer matrix according to the root mean square error of the prediction results, and select multi-step state transfer matrices to construct a multi-scale prediction model based on the optimized Markov chain; Among them, when the matrix is ​​immature, when the variance between the predicted speed and the actual speed exceeds the preset threshold, it is considered that the vehicle is traveling at a constant speed in the next predicted time period; S2: During vehicle travel, the current vehicle motion information is obtained. Based on the current motion information, the vehicle's operating state within a limited future time domain is predicted online using a prediction model, and the prediction model is adaptively updated. S3: Establish a fuel cell vehicle power system model, use the vehicle operating status in the prediction time domain obtained by the prediction model as input, and output the vehicle demand power sequence in the prediction time domain; S4: The vehicle demand power sequence within the forecast horizon is used as the input of a dynamic programming algorithm that introduces a variable load penalty coefficient, and the optimal power allocation sequence within the finite forecast horizon is output; The objective function of the dynamic programming algorithm with the introduction of the variable load penalty coefficient is: in, is the hydrogen consumption, is the variable load penalty coefficient, which is used to balance the durability of the fuel cell and battery. The difference between the fuel cell output last time and the current time, is the size of the predicted time domain; S5: Apply the optimal power allocation sequence within the limited prediction time domain to the vehicle, obtain the vehicle's current motion information again at the next moment, return to step S2 to re-predict and perform optimization of S3-S4, and repeat the optimization until the vehicle stops.

2. The fuel cell vehicle energy management method based on predictive nonlinear model control according to claim 1, characterized in that: The fuel cell vehicle power system model includes a fuel cell model, a motor model, a battery model and a transmission system model.

3. The fuel cell vehicle energy management method based on predictive nonlinear model control according to claim 2, characterized in that: The fuel cell model adopts the basic efficiency-power curve model: in, is the fuel cell efficiency, Output power for the fuel cell.

4. The method for fuel cell vehicle energy management based on predictive nonlinear model control according to claim 2, characterized in that: The motor model is a MAP diagram model established according to the selected parameters: in, is the motor efficiency, is the motor speed, is the motor torque.

5. The fuel cell vehicle energy management method based on predictive nonlinear model control according to claim 2, characterized in that: The battery adopts the internal resistance model, and the calculation is as follows: in, is the battery current, is the battery open circuit voltage, is the battery resistance, is the battery output power, and the battery state of charge SOC is calculated as: in, is the capacity of the battery.

6. The method for fuel cell vehicle energy management based on predictive nonlinear model control according to claim 2, characterized in that: The transmission system model is: in, The power required for the vehicle, is the efficiency of the transmission system, is the vehicle mass, is the acceleration due to gravity, is the rolling resistance coefficient, is the air resistance coefficient, is the windward area, is the speed of the car when there is no wind, is the vehicle rotation mass conversion factor, is the road slope.

7. The method for fuel cell vehicle energy management based on predictive nonlinear model control according to claim 1, characterized in that: The calculation method of the hydrogen consumption is: in, is the molar mass of hydrogen, is the fuel cell stack current, is the number of fuel cell units, is the Faraday constant, is the lower calorific value of hydrogen, is the fuel cell efficiency, Output power for the fuel cell.

8. The method for fuel cell vehicle energy management based on predictive nonlinear model control according to claim 1, characterized in that: The constraint condition of the dynamic programming algorithm that introduces the load penalty coefficient is to predict the possible change value in the time domain space: in, is the fuel cell output power, is the state of charge of the battery, 、 、 is the preset lower limit of the corresponding parameter, 、 、 is the preset upper limit of the corresponding parameter.

Images