Hierarchical real-time multi-target coordination control strategy based on MPC

Abstract

The invention discloses a hierarchical real-time multi-target coordination control strategy based on MPC, and aims to solve the problems that the real-time performance is difficult to guarantee in the optimal control of the existing hybrid power vehicle, and the fuel economy, the dynamic property and the comfort are difficult to balance. According to the strategy, the upper layer is a vehicle speed prediction layer and predicts and corrects the vehicle speed by fully considering driving characteristics, the middle layer is an SOC track planning layer, after the optimal SOC track is solved through DP according to historical data, the mapping relation between working condition information including the predicted vehicle speed obtained by the upper layer and SOC changes is established based on a neural network, and the lower layer is an MPC real-time multi-target coordination control layer. A multi-target nonlinear optimization problem is solved on the basis of a direct multi-target shooting method and a sequential quadratic programming method by taking the SOC track obtained in the middle layer as a reference, so that the calculation time of a control strategy optimization problem is shortened and the real-time performance of the strategy is improved while multi-target tradeoff of dynamic property, economical efficiency and comfort is fully considered.

Description

Technical Field

[0001] The present invention relates to the technical field of power transmission and control of hybrid vehicles, and specifically to a hierarchical real-time multi-objective coordinated control strategy based on MPC. Background Art

[0002] With the development of hybrid power theory and application technology, some scholars have proposed that the MPC algorithm can use the information in the prediction time domain to replace the prior global driving information to realize the prediction of the driver's required torque, and solve the problems that the existing rule-based control strategy for hybrid vehicles has poor applicability, and the setting of threshold values mostly depends on empirical values, resulting in the inability to fully exert the advantages of energy conservation and emission reduction; although the MPC algorithm can realize the optimal control of the vehicle energy management strategy under unknown driving conditions, it is necessary to solve the optimal control problem in the prediction time domain at each stage, which requires high hardware computing power and the real-time performance of the existing algorithm is poor; secondly, the consideration of driving performance such as comfort in the optimization process is lacking in addition to fuel economy.

[0003] Some existing patents, such as the invention patent with the patent number CN 113788007 B, propose a hierarchical real-time energy management method and system. After obtaining the future speed sequence by using the speed prediction network, the predicted speed sequence, the current vehicle SOC state, and the working condition information are segmented as input indexes, and the control action amount is interpolated on the control map trained based on the Q learning algorithm, and then the control action amount is used as the input amount to input into the plug-in hybrid vehicle energy management model to obtain the SOC reference sequence as the control reference of the lower-layer MPC controller, so as to seek the optimal trade-off between the speed prediction accuracy and the model prediction control effect, and realize the optimal real-time energy management. Although this invention realizes the prediction of the future speed sequence through the speed prediction network, only two variables, namely vehicle speed and vehicle acceleration, are considered in the driving samples, and the driving characteristics are not fully reflected. In addition, this invention uses the interpolation method to obtain the control action amount on the control map, and then calculates the SOC reference sequence through the energy management model. The SOC reference sequence calculated by this method loses a certain accuracy and is not the global optimal sequence. Coupled with the fact that the control map cannot be updated online or offline, the finally obtained control sequence is only a local optimal solution or a sub-optimal solution, and the multi-objective trade-off problem in the actual application process of the energy management strategy is not considered. Summary of the Invention

[0004] The main technical problem to be solved by the present invention is to overcome the existing technical difficulties and establish a hierarchical real-time multi-objective coordinated control strategy based on MPC. The upper-layer vehicle speed prediction layer fully considers driving characteristics to predict and correct the vehicle speed. After the middle-layer SOC trajectory planning layer uses DP to solve the optimal SOC trajectory based on historical data, a mapping relationship between the operating conditions information including the predicted vehicle speed obtained from the upper layer and the SOC change is established based on a neural network. The lower-layer MPC real-time multi-objective coordinated control layer uses the SOC trajectory obtained from the middle layer as a reference, and solves the multi-objective non-linear optimization problem based on the direct multiple shooting method and the sequential quadratic programming method. While fully considering the trade-off among power performance, economy and comfort, the calculation time of the control strategy optimization problem is shortened, and the real-time performance of the strategy is improved.

[0005] To solve the above technical problems, the present invention is implemented by adopting the following technical solutions:

[0006] A hierarchical real-time multi-objective coordinated control strategy based on MPC, comprising the following steps:

[0007] The first step: Establish a vehicle speed prediction model considering driving characteristics, select five input variables, including the accelerator pedal opening x1 and its change rate x2 of the driver, the brake pedal opening x3 and its change rate x4, and the historical vehicle speed x5. The output variable is the predicted vehicle speed v. Establish a vehicle speed prediction model based on the BP neural network, select the hyperbolic tangent S-shaped function as the activation function, and select p groups of training input data X and training output data Y to train the model. The predicted vehicle speed obtained by training is shown in Equation (1):

[0008]

[0009] In the formula, f BPNN is the trained vehicle speed prediction model, X is the model input signal, v real is the actual vehicle speed at time k, a max is the acceleration correction coefficient;

[0010] The second step: Establish a global SOC reference trajectory planning model. With the goal of optimal fuel economy, use the DP algorithm to calculate the global optimal SOC trajectory under a whole cycle of operating conditions as shown in Equation (2):

[0011] dSOC(t + 1) = x(t + 1) - x(t) = model(x(t), u(t), cyc(t)) - x(t) (2)

[0012] In the formula, x(t) is the state variable SOC, u(t) is the control variable, model is the vehicle model, and cyc(t) is the cycle of operating conditions.

[0013] Select the Gaussian function as the radial basis function, and establish a reference SOC model based on the RBF neural network, as shown in Equation (3):

[0014]

[0015] where f dSOC is the established dSOC reference model, N in is the model input signal, SOC0 is the current SOC, SOC t is the terminal SOC, D total is the total travel distance, V k is the current vehicle speed, is the predicted vehicle speed at time k+N p is the model target output signal, dSOC tar is the current SOC change rate calculated by Equation (2), k is the predicted SOC change rate at time k+N calculated by Equation (2); p ;

[0016] Step 3: Establish a real-time multi-objective coordinated control strategy based on MPC, with power performance, fuel economy, and comfort as the objectives. The objective function is Equation (4), the state variables and control variables are Equation (5), the constraint conditions are Inequality (6), the reference SOC is Equation (7), and the predicted vehicle speed is Equation (8):

[0017]

[0018] where k a is the current time, N p is the prediction horizon length, w1, w2, and w3 are the weight factors corresponding to power performance, fuel economy, and comfort in the objective function, be is the fuel consumption rate, I b is the battery current, Q b is the battery charge, f gear is the shift penalty matrix, m f is the engine fuel consumption, T eng is the engine torque, n eng is the engine speed, SOC ref is the reference SOC, SOC int (t) is the initial SOC at time t.

[0019] In the control strategy, the process of the real-time multi-objective coordinated control strategy based on MPC in the third step includes:

[0020] S1: Starting from the vehicle speed prediction layer, according to the driving task characteristics and the vehicle speed prediction model with offline weight update based on historical data, calculate the predicted vehicle speed in the prediction horizon through Formula (8) Enter step S2;

[0021] S2: In the SOC trajectory planning layer, according to the predicted vehicle speed and the global SOC reference trajectory planning model updated offline based on historical data, calculate the reference SOC of the prediction horizon through formula (7) ref (k a +1,…,k a +N p ), and enter step S3;

[0022] S3: Based on the objective function formula (4) and the equality constraint condition (5) and inequality constraint condition (6), within each control horizon N c , use the direct multiple shooting method and sequential quadratic programming algorithm to solve the multi-objective nonlinear optimization problem. First, transform the short-term multi-objective optimization problem within the prediction horizon into an NLP problem, and then solve the NLP problem to obtain the optimal control sequence U * at the current moment within the control horizon. Take the first set of control quantities of the control sequence and apply them to the current moment, and enter step S4;

[0023] S4: The control variable U(k a ) at the current moment acts on the vehicle model, and calculate the state X(k a +1) at the next moment through formula (2). If the current moment is less than the total time T of the driving cycle, then use X(k a +1) as the state X(k a ) at the current moment and feedback it to the control strategy model, and return to step S1. Otherwise, save the calculation data for output.

[0024] Compared with the prior art, the advantages of the present invention are:

[0025] 1. The hierarchical real-time multi-objective coordinated control strategy based on MPC described in the present invention fully considers driving characteristics and establishes a vehicle speed prediction model based on BPNN, reducing the dependence of the control strategy on prior global driving information. At the same time, the network weights are updated offline according to historical information, further improving the prediction accuracy of the model;

[0026] 2. The hierarchical real-time multi-objective coordinated control strategy based on MPC described in the present invention establishes a global SOC reference trajectory planning model, uses historical data to solve the approximate optimal SOC trajectory by DP, and then establishes a non-linear mapping relationship between the driving condition information and dSOC through RBF neural network. On the basis of ensuring the model accuracy, the calculation time is significantly shortened, and the requirement for hardware computing power is reduced;

[0027] 3. The hierarchical real-time multi-objective coordinated control strategy based on MPC according to the present invention establishes a real-time multi-objective coordinated control architecture based on MPC. While fully considering the trade-off among power performance, economy, and comfort, it significantly reduces the computational complexity of the control strategy optimization problem and improves the real-time performance of the strategy. Description of the Drawings

[0028] The present invention will be further described below in conjunction with the drawings and embodiments:

[0029] Figure 1 It is a flowchart of a hierarchical real-time multi-objective coordinated control strategy based on MPC according to the present invention. Detailed Embodiment

[0030] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the specific embodiments and with reference to the drawings. It should be understood that these descriptions are exemplary and are not intended to limit the scope of the present invention. In addition, in the following description, the descriptions of well-known structures and technologies are omitted to avoid unnecessarily confusing the concepts of the present invention.

[0031] Taking a domestic power-split hybrid bus of a certain model as an example, the present invention will be further described below in conjunction with the drawings.

[0032] Refer to Figure 1 , the present invention provides a hierarchical real-time multi-objective coordinated control strategy based on MPC, which specifically includes the following steps:

[0033] First step: Referring to the parameters in Table 1, a vehicle model of a power-split hybrid bus is established based on the Matlab / Simulink platform, including an engine model, a motor MG1 model, a motor MG2 model, a battery model, a planetary gear set model, a driver model, and a vehicle longitudinal dynamics model. The construction process of the vehicle model of the power-split hybrid vehicle adopted by the present invention belongs to conventional technology and will not be elaborated here.

[0034] Table 1 Vehicle Parameter Table

[0035]

[0036] Step 2: Establish a vehicle speed prediction model considering driving characteristics. Select five input variables, including the accelerator pedal opening x1 and its change rate x2, the brake pedal opening x3 and its change rate x4, and the historical vehicle speed x5. The output variable is the predicted vehicle speed v. Select the hyperbolic tangent S-shaped function as the activation function, and establish a vehicle speed prediction model based on the BP neural network. Select p groups of training input data X and training output data Y to train the model. The training process is carried out on the Matlab / Simulink platform. Compare the error between the actual output value and the ideal value of the model, and adjust the number of hidden neurons, learning rate, and error gradient of the neural network according to the training results until the prediction error is within ±5 km / h to obtain the trained vehicle speed prediction model. Referring to Equation (1), in particular, for abnormal prediction data, set an acceleration correction coefficient for correction:

[0037]

[0038] where f BPNN is the trained vehicle speed prediction model, X is the model input signal, v real is the actual vehicle speed at time k, a max is the acceleration correction coefficient.

[0039] Step 3: Establish a global SOC reference trajectory planning model. With the goal of optimal fuel economy, use the DP algorithm to calculate the global optimal SOC trajectory under a whole cycle condition as shown in Equation (2):

[0040] dSOC(t + 1) = x(t + 1) - x(t) = model(x(t), u(t), cyc(t)) - x(t) (2)

[0041] where x(t) is the state variable SOC, u(t) is the control variable, model is the vehicle model, and cyc(t) is the cycle condition.

[0042] Select the Gaussian function as the radial basis function, establish a reference SOC model based on the RBF neural network. After setting the number of hidden layer neurons, error precision, and expansion constant, train the RBF neural network to meet the error precision target value to obtain the reference SOC trajectory model. The reference SOC calculated by the model is shown in Equation (3):

[0043]

[0044] where f dSOC is the established dSOC reference model, N in is the model input signal, SOC0 is the current SOC, SOC t is the terminal SOC, D total is the total travel distance, V kis the current vehicle speed, is the predicted time k+N p at the predicted vehicle speed, N tar is the model target output signal, dSOC k is the current SOC change rate calculated by equation (2), is the SOC change rate at the predicted time k+N calculated by equation (2) p at.

[0045] Step 4: Establish a real-time multi-objective coordinated control strategy based on MPC, with power performance, fuel economy, and comfort as the objectives. The objective function is equation (4), the state variables and control variables are equation (5), the constraint conditions are inequality (6), the reference SOC is equation (7), and the predicted vehicle speed is equation (8):

[0046]

[0047]

[0048] where k a is the current time, N p is the prediction horizon length, w1, w2, and w3 are the weight factors corresponding to power performance, fuel economy, and comfort in the objective function, be is the fuel consumption rate, I b is the battery current, Q b is the battery charge, f gear is the shift penalty matrix, m f is the engine fuel consumption, T eng is the engine torque, n eng is the engine speed, SOC ref is the reference SOC, SOC int (t) is the initial SOC at time t.

[0049] In the specific implementation manner, the process of the real-time multi-objective coordinated control strategy based on MPC in the fourth step includes:

[0050] S1: Starting from the vehicle speed prediction layer, according to the driving task characteristics and the vehicle speed prediction model with offline weight update based on historical data, calculate the predicted vehicle speed in the prediction horizon through formula (8) Enter step S2;

[0051] S2: In the SOC trajectory planning layer, according to the predicted vehicle speed and the global SOC reference trajectory planning model with offline weight update based on historical data, calculate the reference SOC in the prediction horizon through formula (7) ref (k a +1,…,k a +N p) and proceed to step S3;

[0052] S3: Based on the objective function formula (4) and the equality constraint condition (5) and the inequality constraint condition (6), within each control time domain N c , the direct multiple shooting method and the sequential quadratic programming algorithm are used to solve the multi-objective nonlinear optimization problem. First, the short-term multi-objective optimization problem within the prediction time domain is transformed into an NLP problem, and then the NLP problem is solved to obtain the optimal control sequence U within the control time domain at the current moment * . Take the first set of control quantities of the control sequence and apply them to the current moment, and proceed to step S4;

[0053] S4: The control variable U(k a ) acts on the vehicle model, and the state X(k a + 1) at the next moment is calculated through formula (2). If the current moment is less than the total time T of the driving cycle, then X(k a + 1) is re-used as the state X(k a ) at the current moment and fed back to the control strategy model, and return to step S1. Otherwise, save the calculation data for output.

[0054] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A hierarchical real-time multi-objective coordinated control strategy based on MPC, characterized in that: The following steps are involved: Step 1: Establish a vehicle speed prediction model that takes driving characteristics into consideration. Select five input variables, including the driver's accelerator pedal opening x1 and change rate x2, brake pedal opening x3 and change rate x4, and historical vehicle speed x5. The output variable is the predicted vehicle speed v. The vehicle speed prediction model is established based on the BP neural network. The hyperbolic tangent S-shaped function is selected as the excitation function. Select p groups of training input data X and training output data Y to train the model. The predicted vehicle speed obtained by training is shown in equation (1): In the formula, f BPNN is the trained vehicle speed prediction model, X is the model input signal, and v real is the actual vehicle speed at time k, a max is the acceleration correction factor; Step 2: Establish a global SOC reference trajectory planning model, take the fuel economy optimization as the goal, and use the DP algorithm to calculate the global optimal SOC trajectory under the entire cycle condition as shown in equation (2): dSOC(t+1)=x(t+1)-x(t)=model(x(t),u(t),cyc(t))-x(t) (2) In the formula, x(t) is the state variable SOC, u(t) is the control variable, model is the vehicle model, cyc(t) is the cycle condition, The Gaussian function is selected as the radial basis function, and the reference SOC model is established based on the RBF neural network, as shown in equation (3): In the formula, f dSOC For the established dSOC reference model, N in is the model input signal, SOC0 is the current SOC, SOC t For terminal SOC, D total is the total travel distance, V k is the current vehicle speed, is time k+N p Predicted vehicle speed at tar is the model target output signal, dSOC k is the current SOC change rate calculated by equation (2), is the predicted time k+N calculated by equation (2) p SOC change rate at Step 3: Establish a real-time multi-objective coordinated control strategy based on MPC, with power, fuel economy and comfort as the goals. The objective function is equation (4), the state variables and control variables are equation (5), the constraints are inequalities (6), the reference SOC is equation (7), and the predicted vehicle speed is equation (8): In the formula, k a is the current moment, N p is the prediction time domain length, w1, w2, w3 are the weight factors corresponding to power, fuel economy and comfort in the objective function, be is the fuel consumption rate, I b is the battery current, Q b is the battery capacity, f gear is the gear shift penalty matrix, m f is the engine fuel consumption, T eng is the engine torque, n eng is the engine speed, SOC ref For reference SOC, SOC int (t) is the initial SOC at time t.

2. A hierarchical real-time multi-objective coordinated control strategy based on MPC according to claim 1, characterized in that: The third step of the MPC-based real-time multi-objective coordinated control strategy process includes: S1: Starting from the vehicle speed prediction layer, according to the driving task characteristics and the vehicle speed prediction model updated with offline weights based on historical data, the predicted vehicle speed in the prediction time domain is calculated by formula (8): Go to step S2; S2: In the SOC trajectory planning layer, according to the predicted vehicle speed The global SOC reference trajectory planning model based on historical data and offline weight update is used to calculate the reference SOC in the prediction time domain through formula (7): ref (k a +1,…,k a +N p ), proceed to step S3; S3: Based on the objective function formula (4) and the equality constraint (5) and inequality constraint (6), in each control time domain N c In this paper, the direct multiple-targeting method and sequential quadratic programming algorithm are used to solve the multi-objective nonlinear optimization problem. First, the short-term multi-objective optimization problem in the prediction time domain is transformed into an NLP problem. Then, the NLP problem is solved to obtain the optimal control sequence U in the control time domain at the current moment. * , take the first set of control quantities of the control sequence and apply it to the current moment, and go to step S4; S4: The current control variable U(k a ) acts on the vehicle model, and the next state X(k a +1), if the current time is less than the total cycle time T, then X(k a +1) as the current state X(k a ) is fed back to the control strategy model and returns to step S1, otherwise the calculated data is saved for output.

Images