阅读说明：本技术 一种混合动力汽车能量管理的两阶段双模型预测控制方法 (Two-stage double-model prediction control method for energy management of hybrid electric vehicle ) 是由 周维 张宁峰 张维刚 于 2020-12-28 设计创作，主要内容包括：本发明公开了一种混合动力汽车能量管理的两阶段双模型预测控制方法,包括：将预测域划分为连续的两个阶段,第一阶段构建降阶的一阶RC模型,第二阶段构建纯内阻模型；分别获取两个阶段内每个时间步的状态可行域边界,并将每个时间步的状态可行域进行离散；基于前向动态规划算法筛选全局成本最优的路径并获得其对应在最后时间步N的最优状态点x～*(N)和从时间步N-1转移到时间步N的最优控制输入P-e～*(N-1),并依次反向递推得到从初始状态转移到时间步1的最优控制输入P-e～*(0)；以P-e～*(0)作为当前时刻发动机的目标输出功率进行功率分配控制；然后随时间步的滚动重复上述步骤。本方法在预测域前段利用降阶的高精度一阶RC模型处理电池功率约束,在预测域后段则采用简单的纯内阻模型,从而在不增加计算复杂度的基础上实现高效且安全的能量分配。(The invention discloses a two-stage and two-model predictive control method for energy management of a hybrid electric vehicle, which comprises the following steps: dividing a prediction domain into two continuous stages, wherein a first-stage RC model with reduced order is constructed in the first stage, and a pure internal resistance model is constructed in the second stage; respectively acquiring the state feasible domain boundary of each time step in two stages, and dispersing the state feasible domain of each time step; screening a global cost optimal path based on a forward dynamic programming algorithm and obtaining an optimal state point x corresponding to the optimal path at the last time step N * (N) and transition from time step N-1 to time stepN optimal control input P e * (N-1) and obtaining the optimal control input P from the initial state to the time step 1 by reverse recursion in turn e * (0) (ii) a With P e * (0) Performing power distribution control as the target output power of the engine at the current moment; the above steps are then repeated with a scrolling of time steps. The method utilizes a reduced-order high-precision one-order RC model to process the battery power constraint at the front stage of the prediction domain, and adopts a simple pure internal resistance model at the rear stage of the prediction domain, so that the high-efficiency and safe energy distribution is realized on the basis of not increasing the computational complexity.)

1. A two-stage double-model predictive control method for energy management of a hybrid electric vehicle is characterized by comprising the following steps:

obtaining an estimated current batterySoC state and polarization voltage state v of a packet₁And obtaining the internal resistance R of the battery pack₀Internal polarization resistance R₁Polarization time constant τ₁And a prediction domain length N;

dividing a prediction domain into two continuous stages, constructing a reduced-order first-order RC model aiming at the first-stage prediction domain, and constructing a pure internal resistance model (Rint model) aiming at the second-stage prediction domain; based on first-order RC model and polarization time constant tau₁Obtaining the first stage prediction domain length N₁Further obtain the second stage prediction domain length N₂；

Respectively acquiring a feasible state boundary of each time step in a first-stage prediction domain and a second-stage prediction domain;

respectively dispersing the state space based on the feasible state boundary of each time step in the first-stage prediction domain and the second-stage prediction domain, and expressing the state point in the feasible region of each time step obtained after dispersion as x_i(k) (ii) a Acquiring all state points x from initial state to last time step N_i(N) optimal feasible path, screening the path with optimal global cost and obtaining the optimal state point x corresponding to the path at the time step N^*(N) and optimal control input P transitioning from time step N-1 to time step N_e ^*(N-1) and reversely recurrently obtaining the optimal control input P for transferring from the initial state to the time step 1_e ^*(0)；

With P_e ^*(0) Performing power distribution control as the target output power of the engine at the current moment; the above steps are repeated with a rolling motion of time steps.

2. The energy management control method for hybrid electric vehicle according to claim 1, wherein the first-stage prediction domain length N is obtained based on the first-stage RC model and the polarization time constant τ₁Further obtain the second stage prediction domain length N₂The method specifically comprises the following steps:

the maximum n is solved that satisfies the following formula,

in the formula, the polarization time constant τ₁＝R₁C₁，C₁Is a polarization capacitor; delta_relIs a preset relative tolerance; e is a natural constant;

the maximum N obtained by solving is the length N of the prediction domain of the first stage₁Second stage prediction field length N₂＝N-N₁。

3. The energy management control method of claim 1, wherein the obtaining of the feasible region boundary of each time step in the first-stage prediction region and the second-stage prediction region respectively comprises:

solving for the feasible region boundary [ soc ] of each time step in the first-stage prediction domain by the following formula_min(k),v_1,max(k)]And [ soc)_max(k),v_1,min(k)]，k＝1,2,…,N₁，

Wherein i (k) is calculated by the following formula,

solving the feasible region boundary soc of each time step in the second-stage prediction domain by the following formula_min(k) And soc_max(k)，k＝N₁+1,N₁+2,…,N，

In the above formula, k represents a time step, SoC (k) represents a SoC state of the battery; q_nomRepresents a rated capacity of the battery; i (k) represents a current; v_oc(soc (k)) represents the open-circuit voltage of the battery; v. of₁(k) Represents the cell polarization voltage; p_dmd(k) Representing the required power of the bus; p_e(k) Representing engine power; r₀(soc (k)) represents the ohmic internal resistance of the cell.

4. The energy management control method of claim 3, wherein discretizing the state space based on the feasible region boundary of each time step in the first-stage prediction domain and the second-stage prediction domain respectively comprises:

discretizing the state space based on the feasible region boundaries for each time step within the first stage prediction domain comprises:

based on the determined feasible region boundary of each time step in the first-stage prediction domain, obtaining an approximate linear equation of the feasible region of each time step in the first-stage prediction domain:

firstly, dispersing a state variable SoC intoAnother state variable polarization voltage v₁Based onAnd the above-mentioned approximate linear equation of feasible domain of each time step is directly dispersed into

Discretizing the state space based on the feasible region boundaries for each time step within the second-stage prediction domain comprises:

discretizing state variable SoC into

5. The energy management control method for hybrid electric vehicle according to claim 4, wherein the obtaining of all state points x from the initial state to the last time step N_i(N) optimal feasible path, screening the path with optimal global cost and obtaining the optimal state point x corresponding to the path at the time step N^*(N) and optimal control input P transitioning from time step N-1 to time step N_e ^*(N-1) and reversely recurrently obtaining the optimal control input P for transferring from the initial state to the time step 1_e ^*(0) The method specifically comprises the following steps:

the feasible domains obtained after the dispersion of each time step in the first-stage prediction domain and the second-stage prediction domain are uniformly expressed as state vectorsFeasible regionThe ith inner state point is denoted as x_i(k)；

Uniformly dispersing the controlled variable in the constraint range of the controlled variable intoIs calculated atUnder action, the single state point x can be transferred from a time step k to a step k +1_iState set of (k +1)

In the formula, g () represents an inverse function of a state variable SoC state transition equation, which in the first stage and the second stage respectively is:

calculating from initial state to state point x_iCost of feasible path of (k +1)

In the formula (I), the compound is shown in the specification,representing last time step state vectorTo the state point x_iThe cost of the transfer of (k +1),state vector representing initial state to last time stepThe optimal cost of (2); whereinAndcalculated by the following formula:

in the formula eta_e(P_e(k) ) is an optimal efficiency curve for the hybrid system; q_lhvIs the low heat value of the fuel oil; zeta is a SoC track deviation penalty coefficient; SoC _ ref (k) is the SoC reference trace, preferably may be 0.5;

screening from initial state to state point x_iOptimal cost of feasible path of (k +1)And obtains the corresponding optimal control input for the transition from the state of time step k to time step k +1

Repeating the above process to obtain all state points x from the initial state to the time step N_i(N) optimal feasible path, screening the path with optimal global cost and obtaining the optimal state point corresponding to the path at the time step NAnd an optimal control input P for transitioning from time step N-1 to time step N_e ^*(N-1) and reversely recurrently obtaining the optimal control input P for transferring from the initial state to the time step 1_e ^*(0)。

6. The hybrid vehicle energy management control method according to any one of claims 1 to 5, characterized in that the following state and control constraints due to physical limitations and safety are satisfied throughout the prediction domain:

soc_lb≤soc(k)≤soc_hb

P_e,min≤P_e(k)≤P_e,max

ΔP_e,min≤P_e(k+1)-P_e(k)≤ΔP_e,max

P_b,min(k)≤P_b(k)≤P_b,max(k)

wherein, soc_lbAnd soc_hbIs SoC minimum and maximum; p_b,minAnd P_b,maxThe minimum and maximum instantaneous charge and discharge power of the battery pack is mainly restricted by the terminal voltage of the charge and discharge; p_e,minAnd P_e,maxIs the minimum and maximum output power of the engine; delta P_e,minAnd Δ P_e,maxIs the maximum allowable ramp down and ramp up rate of the engine.

7. The hybrid vehicle energy management control method according to claim 6,

for the first-order RC model corresponding to the first-stage prediction domain, the voltage v is polarized₁Is a state variable, therefore P_b,minAnd P_b,maxThe represented peak power constraint can be directly translated into a voltage constraint, calculated as:

V_low≤V_oc(soc(k))-i(k)R₀(soc(k))-v₁(k)≤V_high

for the Rint model corresponding to the second stage prediction domain, P_b,minAnd P_b,maxThe calculation is as follows:

wherein, V_lowAnd V_highThe discharge cut-off voltage and the charge cut-off voltage of the battery pack are respectively; v. of₁(n) is the polarization voltage state of the last time step of the first phase.

Technical Field

The invention relates to the technical field of automobile energy management, in particular to a hybrid electric vehicle energy management control method.

Background

The hybrid electric vehicle can not only greatly improve the fuel economy of the vehicle, but also improve the dynamic property of the vehicle. The power system of the hybrid electric vehicle mainly comprises an engine, a motor, a battery pack, a control system and other components, and is divided into plug-in hybrid power and non-plug-in hybrid power according to whether the battery can be charged by an external power supply or not.

The hybrid electric vehicle is provided with two sets of energy systems (fuel oil and electric energy), and the energy management of the hybrid electric vehicle improves the working efficiency of the power system and effectively reduces the fuel oil consumption of the vehicle by adjusting the output power distribution of an engine and a battery. In recent years, model predictive control has been widely used in the field of energy management of hybrid vehicles. For example, patent CN 110696815 a, "a method for energy management prediction for internet-connected hybrid electric vehicles," plans a reference SoC trajectory by using an optimal SoC feature obtained by offline dynamic planning in combination with a neural network, tracks the reference SoC trajectory by using a model prediction method based on dynamic planning, and solves a power distribution optimization problem in a prediction domain. The patent CN 109017809A energy distribution method based on cross-country working condition prediction establishes a vehicle electric transmission power model, a power battery simple internal resistance model, an engine generator model and a system state equation, calculates the future required power of the vehicle based on the information of working condition prediction vehicle speed, gradient, rolling resistance and the like, and adopts a model prediction control strategy of an embedded dynamic programming algorithm to give the optimal energy distribution at the next moment. Although the energy management method based on the model predictive control can improve the fuel economy of the whole vehicle, the battery model adopts a simple pure internal resistance model, and the calculation accuracy of the battery power constraint boundary is low. Particularly, under the condition of extremely sparse control grid division required by real-time calculation, the calculation accuracy of the lower battery power constraint boundary influences the optimality of the result, so that the oil-saving potential of the hybrid electric vehicle cannot be exerted to the maximum extent. If a first-order RC model is directly adopted in the prediction domain, the calculation complexity is greatly increased, and real-time calculation cannot be carried out in the vehicle-mounted embedded controller. Therefore, it is necessary to reasonably simplify the battery model on the premise of ensuring the calculation accuracy of the battery power constraint boundary, so that the algorithm can run in real time and the fuel consumption of the vehicle is reduced to the maximum extent.

Disclosure of Invention

The invention provides a two-stage and two-model predictive control method for energy management of a hybrid electric vehicle, which aims to solve the problem that the fuel-saving potential of the hybrid electric vehicle cannot be furthest exerted because the power constraint of a battery cannot be processed with high precision in the existing energy management control scheme.

The two-stage double-model predictive control method for the energy management of the hybrid electric vehicle comprises the following steps:

acquiring estimated SoC state and polarization voltage state v of current battery pack₁And obtaining the internal resistance R of the battery pack₀Internal polarization resistance R₁Polarization time constant τ₁And a prediction domain length N;

dividing a prediction domain into two continuous stages, constructing a first-stage RC model aiming at the first-stage prediction domain, and constructing a pure internal resistance model aiming at the second-stage prediction domain; based on first-order RC model and polarization time constant tau₁Obtaining the first stage prediction domain length N₁Further obtain the second stage prediction domain length N₂；

Respectively acquiring a feasible domain boundary of each time step in the first-stage prediction domain and the second-stage prediction domain;

respectively dispersing the state space based on the feasible domain boundary of each time step in the first-stage prediction domain and the second-stage prediction domain, and expressing the state point in the feasible domain of each time step obtained after dispersion as x_i(k)；

Acquiring all state points x from an initial state to a time step N_i(N) optimal feasible path, screening the path with optimal global cost and obtaining the optimal state point x corresponding to the path at the time step N^*(N) and transition from time step N-1 to timeOptimal control input P for step N_e ^*(N-1) and reversely recurrently obtaining the optimal control input P for transferring from the initial state to the time step 1_e ^*(0)；

With P_e ^*(0) Performing power distribution control as a target output power of the engine; the above steps are repeated with a rolling motion of time steps.

Further, the first-stage prediction domain length N is obtained based on the first-stage RC model and the polarization time constant tau₁Further obtain the second stage prediction domain length N₂The method specifically comprises the following steps:

the maximum n is solved that satisfies the following formula,

in the formula, the polarization time constant τ₁＝R₁C₁，C₁Is a polarization capacitor; delta_relPresetting a relative tolerance; e is a natural constant;

the maximum N obtained by solving is the length N of the prediction domain of the first stage₁Second stage prediction field length N₂＝N-N₁。

Further, the obtaining of the feasible region boundary of each time step in the first-stage prediction domain and the second-stage prediction domain respectively specifically includes:

solving for the feasible region boundary [ soc ] of each time step in the first-stage prediction domain by the following formula_min(k),v_1,max(k)]And [ soc)_max(k),v_1,min(k)]，k＝1,2,…,N₁，

Wherein i (k) is calculated by the following formula,

solving the feasible region boundary soc of each time step in the second-stage prediction domain by the following formula_min(k) And soc_max(k)，k＝N₁+1,N₁+2,…,N，

In the above formula, k represents a time step, SoC (k) represents a SoC state of the battery; q_nomRepresents a rated capacity of the battery; i (k) represents a current; v_oc(soc (k)) represents the open-circuit voltage of the battery; v. of₁(k) Represents the cell polarization voltage; p_dmd(k) Representing the required power of the bus; p_e(k) Representing engine power; r₀(soc (k)) represents the ohmic internal resistance of the cell.

Further, the discretizing the state space based on the feasible domain boundary of each time step in the first-stage prediction domain and the second-stage prediction domain respectively specifically includes:

discretizing the state space based on the feasible region boundaries for each time step within the first stage prediction domain comprises:

based on the determined feasible region boundary of each time step in the first-stage prediction domain, obtaining an approximate linear equation of the feasible region of each time step in the first-stage prediction domain:

firstly, dispersing a state variable SoC into(k＝1,2,…,N₁) Another state variable polarization voltage v₁Based onAnd the above-mentioned approximate linear equation of feasible domain of each time step is directly dispersed into

Discretizing the state space based on the feasible region boundaries for each time step within the second-stage prediction domain comprises:

discretizing state variable SoC into(k＝N₁+1,N₁+2,…,N)。

Further, the obtaining all state points x from the initial state to the time step N_i(N) optimal feasible path, screening the path with optimal global cost and obtaining the optimal state point x corresponding to the path at the time step N^*(N) and optimal control input P transitioning from time step N-1 to time step N_e ^*(N-1) and reversely recurrently obtaining the optimal control input P for transferring from the initial state to the time step 1_e ^*(0) The method specifically comprises the following steps:

the feasible domains obtained after the dispersion of each time step in the first-stage prediction domain and the second-stage prediction domain are uniformly expressed as the shapesState vector(k-1, 2, …, N), feasible domainThe ith inner state point is denoted as x_i(k)；

Uniformly dispersing the controlled variable in the constraint range of the controlled variable intoIs calculated atUnder the action of the action, the ith state point x can be transferred from the time step k to the step k +1_iState set of (k +1)

In the formula, g () represents an inverse function of a state variable SoC state transition equation, which in the first stage and the second stage respectively is:

calculating from initial state to state point x_iCost of feasible path of (k +1)

In the formula (I), the compound is shown in the specification,representing last time step state vectorTo the state point x_iThe cost of the transfer of (k +1),state vector representing initial state to last time stepThe optimal cost of (2); whereinx_i(k +1) andcalculated by the following formula:

in the formula eta_e(P_e(k) ) is an optimal efficiency curve for the hybrid system; q_lhvIs the low heat value of the fuel oil; zeta is a SoC track deviation penalty coefficient; SoC _ ref (k) is the SoC reference trace, preferably, may be 0.5;

screening from initial state to state point x_iOptimal cost of feasible path of (k +1)And obtains the corresponding optimal control input for the transition from the state of time step k to time step k +1

Repeating the above process to obtain all state points x from the initial state to the time step N_i(N) optimal feasible path, screening the path with optimal global cost and obtaining the optimal state point x corresponding to the path at the time step N^*(N) and optimal control input P transitioning from time step N-1 to time step N_e ^*(N-1) and reversely recurrently obtaining the optimal control input P for transferring from the initial state to the time step 1_e ^*(0)。

Further, the following state and control constraints due to physical constraints and safety are satisfied throughout the prediction domain:

soc_lb≤soc(k)≤soc_hb

P_e,min≤P_e(k)≤P_e,max

ΔP_e,min≤P_e(k+1)-P_e(k)≤ΔP_e,max

P_b,min≤P_b(k)≤P_b,max

wherein, soc_lbAnd soc_hbIs SoC minimum and maximum; p_b,minAnd P_b,maxThe minimum and maximum instantaneous charge and discharge power of the battery pack is mainly restricted by the terminal voltage of the charge and discharge; p_e,minAnd P_e,maxIs the minimum and maximum output power of the engine; delta P_e,minAnd Δ P_e,maxIs the maximum allowable ramp down and ramp up rate of the engine.

Further, for the first-order RC model corresponding to the first-stage prediction domain, P is the state variable because the polarization voltage is the state variable_b,minAnd P_b,maxThe corresponding peak power constraint can be directly converted into a voltage constraint, which is calculated as:

V_low≤V_oc(soc(k))-i(k)R₀(soc(k))-v₁(k)≤V_high

for the pure internal resistance model corresponding to the second stage prediction domain, P_b,minAnd P_b,maxThe calculation is as follows:

wherein, V_lowAnd V_highThe discharge cut-off voltage and the charge cut-off voltage of the battery pack are respectively; v. of₁(n) is the polarization voltage state of the last time step of the first phase.

Advantageous effects

The invention provides a hybrid electric vehicle energy management control method based on model predictive control, which divides the whole prediction domain into two continuous stages in time domain: within the first-stage prediction domain, applying a reduced-order first-stage RC model to obtain a more accurate power constraint, and adaptively determining a duration of the first-stage prediction domain from the estimated RC model time constant; in the second stage prediction domain, a simple pure internal resistance model is applied to improve the calculation speed so as to improve the capability of meeting the battery power constraint on the premise of ensuring the calculation efficiency. Compared with the traditional method based on the pure internal resistance model, the scheme can more reasonably process the battery power constraint on the premise of not increasing the calculation burden, thereby obtaining better fuel economy than the traditional method under the condition of sparse control grid division, and further improving the fuel economy of the hybrid electric vehicle when being applied to a real vehicle.

Drawings

FIG. 1 is a flow chart of a two-stage dual-model predictive control method for energy management of a hybrid electric vehicle according to an embodiment of the present invention;

FIG. 2 is a schematic diagram of two battery equivalent circuit models, a pure internal resistance model and a first-order RC model, according to an embodiment of the present invention;

FIG. 3 is a diagram illustrating that the feasible region discrete points at each time step in the first-stage prediction domain can be approximately equivalent to a straight line according to an embodiment of the present invention;

fig. 4 is a schematic diagram of an implementation of a hybrid electric vehicle energy management control method according to an embodiment of the present invention.

FIG. 5 is a graph showing the relative percentage of fuel consumption for a two-stage dual-model method and a pure internal resistance model method obtained from simulation testing

Detailed Description

The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

For the hybrid electric vehicle energy management method based on model predictive control, because the calculation is simple, a pure internal resistance model is usually adopted, and the circuit schematic diagram is shown in fig. 2(a), the model only has one state variable of SoC, the calculation efficiency is high, but the processing of battery power constraint is rough. The circuit diagram of the first-order RC model is shown in fig. 2(b), and the RC link takes into account the polarization voltage of the battery, so that the first-order RC model has more advantages in processing the power capability of the battery, but also increases the dimension of the system state space and increases the calculation cost.

In order to better process the power constraint of the battery on the premise of ensuring the computational efficiency, the invention provides a two-stage dual-model energy management control scheme embedded in model predictive control. The main difference with the existing scheme is that the entire prediction domain is divided into two consecutive stages in the time domain: within a first phase of the prediction domain, applying a first order RC model to more accurately compute the power constraint boundaries of the battery; in the second stage of the prediction domain, a simple pure internal resistance model is applied to ensure the calculation efficiency; meanwhile, the equivalent order reduction processing within the error allowable range is carried out on the first-order RC model so as to further reduce the calculation cost.

The energy management control scheme provided by the invention aims to improve the fuel economy of the whole vehicle, so that the cost function of the optimal energy management problem is set as the accumulated fuel consumption in a prediction domain, and meanwhile, in order to realize electric quantity maintenance, an SoC track deviation punishment needs to be added. The total cost function is then expressed as:

wherein eta is_e(P_e(k) ) is an optimal efficiency curve for the hybrid system; q_lhvIs the low heat value of the fuel oil; zeta is a SoC track deviation penalty coefficient; SoC _ ref (k) is a SoC reference trace; n is the time length of the prediction domain, k is the time step, in this embodiment, the length of each time step is 1 second, and the time length of the prediction domain is N seconds.

For the pure internal resistance model, SoC is the only state variable, and the state transition equation is as follows:

for the first-order RC model, the state variables are SoC and the polarization voltage v₁：

soc(k+1)＝soc(k)-i(k)/Q_nom (3.1)

Wherein R is₁And τ₁The polarization internal resistance and the polarization time constant, tau, of the first-order RC model circuit in FIG. 2(b) are shown₁＝R₁C₁，C₁Is a polarization capacitor; the instantaneous current i (k) may be based on a control input P_e(k) Calculated by the following formula:

at the same time, the following state and control constraints due to physical constraints and safety considerations must be satisfied throughout the prediction domain:

soc_lb≤soc(k)≤soc_hb (5.1)

P_e,min≤P_e(k)≤P_e,max (5.2)

ΔP_e,min≤P_e(k+1)-P_e(k)≤ΔP_e,max (5.3)

P_b,min≤P_b(k)≤P_b,max (5.4)

wherein, soc_lbAnd soc_hbIs the minimum and maximum values of SoC; p_b,minAnd P_b,maxThe minimum and maximum instantaneous charge and discharge power of the battery pack is mainly restricted by the terminal voltage of the charge and discharge; p_e,minAnd P_e,maxIs the minimum and maximum output power of the engine; delta P_e,minAnd Δ P_e,maxIs the maximum allowable ramp down and ramp up rate of the engine.

For the first-order RC model corresponding to the first-stage prediction domain, P is the state variable because the polarization voltage is the state variable_b,minAnd P_b,maxThe represented peak power constraint can be directly translated into a voltage constraint, calculated as:

V_low≤V_oc(soc(k))-i(k)R₀(soc(k))-v₁(k)≤V_high (6.1)

for the pure internal resistance model corresponding to the second stage prediction domain, P_b,minAnd P_b,maxThe calculation is as follows:

wherein, V_lowAnd V_highThe discharge cut-off voltage and the charge cut-off voltage of the battery pack are respectively; v. of₁(n) is the polarization voltage state of the last time step of the first phase. To further reduce the amount of computation, it is necessary to pair feasible fieldsAnd calculating the state space, and performing order reduction processing on the first-order RC model. Two boundary points [ soc ] at each time step in the first-stage prediction domain are determined as shown in FIG. 3_min(k+1),v_1,max(k+1)]And [ soc)_max(k+1),v_1,min(k+1)]：

After determining the boundary points of the feasible region at each time step, the equation of the approximate straight line of the feasible region in fig. 3 is calculated:

to control the reduced order error, the first stage predicts the domain duration N₁Determined by the formula, N₁Maximum n to satisfy the following:

wherein, delta_relThe relative tolerance is preset, which is preset to 20% in the embodiment; e is a natural constant. The first phase duration N corresponding to different polarization time constants can be obtained₁The look-up table may be stored in practice and is operable to adaptively determine the duration of the first stage prediction domain based on the estimated polarization time constant of the first stage RC model.

Determining a feasible region boundary soc for each time step in the second stage prediction domain by_min(k) And soc_max(k)，k＝N₁+1,N₁+2,…,N，

The solution according to the invention is further illustrated below by means of several examples.

Example 1

As shown in fig. 1, the embodiment provides a two-stage dual-model predictive control method for energy management of a hybrid electric vehicle, including:

s01: acquiring estimated SoC state and polarization voltage state v of current battery pack₁And obtaining the internal resistance R of the battery pack₀Internal polarization resistance R₁Polarization time constant τ₁And a prediction domain length N; wherein, the SoC state and the polarization voltage state v of the current battery pack₁Estimated by a state observer, the polarization time constant tau₁＝R₁C₁，C₁Is a polarization capacitance.

S02: dividing a prediction domain into two continuous stages, constructing a first-stage RC model aiming at the first-stage prediction domain, and constructing a pure internal resistance model aiming at the second-stage prediction domain; based on first-order RC model and polarization time constant tau₁Obtaining the first stage prediction domain length N₁Further obtain the second stage prediction domain length N₂(ii) a First stage prediction field length N₁By equation (9) and polarization time constant τ₁The length N of the prediction domain in the second stage is obtained by solving₂＝N-N₁。

S03: respectively acquiring a feasible domain boundary of each time step in the first-stage prediction domain and the second-stage prediction domain; the method specifically comprises the following steps: calculating the feasible region boundary [ soc ] of each time step in the first-stage prediction domain according to the formulas (7.1) and (7.2)_min(k),v_1,max(k)]And [ soc)_max(k),v_1,min(k)]，k＝1,2,…,N₁Calculating the feasible region boundary soc of each time step in the second-stage prediction domain according to the formula (10)_min(k) And soc_max(k)，k＝N₁+1,N₁+2,…,N。

S04: respectively dispersing the state space based on the feasible domain boundary of each time step in the first-stage prediction domain and the second-stage prediction domain, and expressing the state point in the feasible domain of each time step obtained after dispersion as x_i(k)。

The method specifically comprises the following steps: discretizing the state space according to the feasible domain boundary of each time step in the first-stage prediction domain and the second-stage prediction domain obtained in the step S03; for the first stage prediction domain, firstly dispersing the state variable SoC into (k＝1,2,…,N₁) Another state variable polarization voltage v₁Based onAnd formula (8) is directly discretized intoFor the second stage prediction domain, dispersing the state variable SoC into (k＝N₁+1,N₁+2,…,N)。

S05: acquiring all state points x from an initial state to a time step N_i(N) optimal feasible path, screening the path with optimal global cost and obtaining the optimal state point x corresponding to the path at the time step N^*(N) and optimal control input P transitioning from time step N-1 to time step N_e ^*(N-1) and reversely recurrently obtaining the optimal control input P for transferring from the initial state to the time step 1_e ^*(0)。

For better presentation, the following description is givenThe feasible domains obtained after the dispersion of each time step in the first-stage prediction domain and the second-stage prediction domain are uniformly expressed as state vectors(k-1, 2, …, N), feasible domainThe state points in are denoted x_i(k) In that respect The method specifically comprises the following steps:

s051: uniformly dispersing the controlled variable in the constraint range of the controlled variable intoIs calculated atUnder the action of the action, the ith state point x can be transferred from the time step k to the step k +1_iState set of (k +1)

In the formula, g () represents the inverse function of the state transition equation of the state variable SoC (i.e., formulas (2), (3.1));

s052: calculating from initial state to state point x_iCost of feasible path of (k +1)

In the formula (I), the compound is shown in the specification,to representLast time step state vectorTo the state point x_iThe cost of the transfer of (k +1),state vector representing initial state to last time stepThe optimal cost of (2); whereinAndcalculated by the following formula:

s053: screening from initial state to state point x_iOptimal cost of feasible path of (k +1)And a corresponding optimal control input for transitioning from the state at time step k to time step k +1

S054: repeating the above steps S051 to S053, and gradually recurring from the time step 1 to the time step N to obtainTo all state points x from initial state to time step N_i(N) optimal feasible path, screening the path with optimal global cost and obtaining the state point x corresponding to the path at the time step N^*(N) and optimal control input P transitioning from time step N-1 to time step N_e ^*(N-1) and reversely recurrently obtaining the optimal control input P for transferring from the initial state to the time step 1_e ^*(0)。

S06: with P_e ^*(0) Performing power distribution control as a target output power of the engine; the above steps S01 to S06 are repeated with scrolling in time steps. The implementation of the above control process is schematically shown in fig. 4.

In order to verify the control performance of the energy management control method provided by the invention, a power system of a series hybrid school bus is taken as an object, a Matlab/Simulink/Stateflow is used for building a MiL test platform, the method and a single Rint model (namely a pure internal resistance model) method are simulated under 6 city and suburban driving conditions respectively, the fuel consumption values of the two obtained methods are shown in the following table (Dual and Rint in the table respectively refer to the method provided by the invention and the Rint model method), and the relative percentage of the fuel consumption of the two methods is shown in FIG. 5. Since the final SoC values after each simulation are not completely equal (Δ socf ≠ 0), for fair comparison of the fuel consumption differences, the fuel consumption values in the table have been corrected according to the final SoC value differences. Because the difference of the SoC final value is small, the error brought by the correction does not influence the overall result of the fuel consumption comparison. In addition, in order to verify the control performance of the method under different vehicle speed prediction accuracies, the simulation test also sets two situations of accurate speed prediction and inaccurate speed prediction (constant speed in a prediction domain).

TABLE 1 fuel consumption table for two methods under different working conditions

As can be seen from fig. 5, the proposed method achieves lower fuel consumption when the discrete state points are sparse, and the fuel consumption difference between the two methods is 1.81% under the condition of inaccurate vehicle speed prediction. This shows that the proposed method is more robust to inaccuracies in the vehicle speed prediction. In addition, although the proposed method has no advantage in fuel consumption compared to the Rint model method when the discrete state points are encrypted to 200, the algorithm takes most of the sampling time (1 second) to complete the calculation in view of the single-step calculation time of the algorithm listed in the table, and such a high density of discrete state points obviously does not meet the requirement of real-time performance. Compared with a Rint model method, the method has higher engineering application value due to the fact that the proposed method has advantages under sparse discrete state points and real-time performance is also needed by the vehicle-mounted controller.

To sum up, the invention provides a two-stage and two-model predictive control method for energy management of a hybrid electric vehicle, which divides the whole predictive domain into two continuous stages in the time domain: in the first stage prediction domain, a first-order RC model is applied to obtain more accurate power constraint; in the second stage prediction domain, a low-order Rint model is applied to ensure the calculation efficiency; meanwhile, the equivalent order reduction processing within the error allowable range is carried out on the first-order RC model so as to further reduce the calculation cost; the capability of meeting the battery power constraint can be improved on the premise of ensuring the calculation efficiency. Compared with the traditional Rint model-based method, the scheme can more reasonably process the battery power constraint on the premise of not increasing the calculation burden, thereby obtaining better fuel economy than the traditional method under the condition of sparse control grid division, and further improving the fuel economy of the hybrid electric vehicle when being applied to a real vehicle.

The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.