Processor unit and method for predictive speed control of a motor vehicle

Abstract

The invention comprises a processor unit (1) for the predictive control of a motor vehicle (2), having a memory unit (3) containing an MPC cost function (J). This function includes a time term (Z) weighted by the factor (q1), which indicates the predicted travel time for a route in the prediction horizon, and an energy term (E) weighted by the factor (q2), based on stored energy (E sto (F whl, v)), which depends on the vehicle speed (v) and wheel force (F whl ). The processor unit (1) is capable of calculating a gradient (dE sto (F whl,v)) of the energy and approximate it using an approximation function. It periodically adjusts the coefficient vector to optimize an adapted MPC cost function (J), thereby achieving efficient speed control of the motor vehicle (2), regardless of the drivetrain variant. The invention also relates to a corresponding method and a motor vehicle.

Description

[0001] The invention relates to a processor unit and a method for predictive cruise control of a motor vehicle. Furthermore, the invention relates to a motor vehicle.

[0002] The driver and their driving style have a huge influence on energy consumption when operating a motor vehicle. Cruise control can only take route topology into account.

[0003] In order to keep energy consumption as low as possible when operating a motor vehicle, methods of model-based predictive control (MPC for short), based on a cost function, can be used for speed control, particularly in the area of ​​engine control in motor vehicles.

[0004] Such optimization-based controls require not only a lot of computing power and computing time but also precise knowledge of the vehicle and the powertrain, so that their use for online calculations is both problematic and requires a high level of development effort.

[0005] DE 10 2020 203 742 A1 discloses a processor unit for model-based predictive control of a motor vehicle, wherein the processor unit is configured to execute an MPC algorithm comprising a first solver module, a second solver module, a third solver module, a longitudinal dynamics model, and three cost functions, wherein a first cost function is assigned to the first solver module, a second cost function is assigned to the second solver module, and a third cost function is assigned to the third solver module, by executing the first solver module for a route section ahead, taking into account the longitudinal dynamics model, to calculate a speed trajectory minimizing the first cost function, according to which the motor vehicle is to move within a prediction horizon, and to calculate a battery state of charge profile minimizing the first cost function,which serves as an energy storage device for an electric machine of the motor vehicle, to calculate a trajectory of integer control variables minimizing the second cost function by executing the second solver module based on the speed trajectory, based on the battery charge state curve, and taking into account constraints, and to calculate a torque trajectory minimizing the third cost function by executing the third solver module for an initial section of the prediction horizon for the electric machine, an internal combustion engine, and a braking system of the motor vehicle, according to which torque trajectory the electric machine, the internal combustion engine, and the braking system are to provide torques within the prediction horizon.

[0006] Further prior art is disclosed in DE 10 2019 216 445 A1, DE 10 2020 202 803 A1, DE 10 2021 108 521 A1 and DE 10 2023 201 423 A1.

[0007] It is therefore an object of the invention to provide an improved processor unit and an improved method for predictive control of a motor vehicle. Furthermore, it is an object to provide a motor vehicle with such a processor unit and / or method.

[0008] The object is achieved by a processor unit having the features of claim 1 and a method having the features of claim 8 as well as a motor vehicle having the features of claim 16. Advantageous embodiments are the subject of the dependent claims.

[0009] The task is solved by a processor unit for predictive control of a motor vehicle, wherein a memory unit with an MPC (Model Predictive Control) cost function is provided, wherein the MPC cost function comprises a time term weighted with a first weighting factor, which contains a travel time predicted independently of the drive train variant of the motor vehicle, which the motor vehicle requires to cover an entire distance predicted within a prediction horizon, and wherein the MPC cost function comprises an energy term weighted with a second weighting factor, which is based on energy temporarily stored as a function of at least the vehicle speed and the wheel force, so that the energy term provided is independent of the drive train variant of the motor vehicle, wherein the processor unit is designed to determine a gradient of the temporarily stored energy and to use it for determining the energy term, and wherein the processor unit is designed to approximate the gradient by an approximation function with a coefficient vector and basis functions, and wherein the processor unit is designed to recursively adapt the coefficient vector at predetermined intervals during operation of the motor vehicle in order to determine an adapted MPC cost function, and wherein the processor unit is further configured to minimize the adapted MPC cost function to determine a speed control for the motor vehicle with respect to the prediction horizon as a function of the respectively adapted approximated gradient and the weighted time term, so that a minimization independent of the drive train variant of the motor vehicle is possible.

[0010] In the following, the temporarily stored energy is referred to as E sto (F whl ,v), the vehicle speed as v, the wheel force as F whl and the energy term is denoted as E and the time term as Z.

[0011] According to the invention, it was recognized that all described processor units for MPC-based planning in the prior art are based on a specific powertrain variant. This means that the respective processor unit is always tailored to the respective powertrain, such as for a purely internal combustion engine vehicle, a hybrid vehicle, or a purely electric vehicle, for example, for an electric vehicle with a single electric motor and a single transmission stage. Therefore, transferring the respective MPC cost function to a different powertrain variant is currently not possible.

[0012] The invention further recognized that developing and tuning a separate MPC cost function for each drive variant would result in unacceptably high application complexity. In particular, the problem to be solved becomes more complex due to the multitude of possibilities, such as hybrid vehicles with electric drive, fuel cells, purely electric drives with varying numbers of electric motors and transmissions, or pure combustion engines. Constantly new developments also necessitate constant updating. Furthermore, covering all variants is virtually impossible, especially with regard to time and the resulting development costs.

[0013] Furthermore, the invention further recognized that an artificially expanded formulation to a superset of variables that may be present in different powertrain variants is detrimental to the real-time capability of the processor unit. Furthermore, experiments have shown that these artificial expansions would lead to undesirable planning results in certain situations, which in turn would require situation-specific optimization constraints or optimization constraints, which are difficult to manage.

[0014] This is achieved by means of the processor unit according to the invention for predictive control of a motor vehicle. The processor unit according to the invention uses the energy term according to the invention to provide a generalized description of energy, which provides the current energy consumption in conjunction with time with respect to the prediction horizon in a uniform manner for a wide variety of drivetrain variants.

[0015] The MPC cost function contains, as a time term, a travel time weighted by a first weighting factor and predicted, in particular, according to a generally valid longitudinal dynamics model. This travel time is the time required by the vehicle to cover the entire distance predicted within the prediction horizon. The time term in the MPC cost function means that, depending on the choice of weighting factors, a low speed is not always considered optimal in terms of energy consumption, thus eliminating the problem that the resulting speed is always at the lower end of the permitted speed range.

[0016] In addition, the MPC cost function contains an energy term that is independent of a specific powertrain and is based on energy stored as a function of at least the vehicle speed and the wheel force, and thus specifies the energy required to drive the vehicle within the prediction horizon, independent of a specific energy form such as battery or fuel / gasoline.

[0017] According to the invention, the processor unit is designed to determine a gradient of the temporarily stored energy and to use it to determine the energy term and also to approximate the gradient by an approximation function with coefficient vector and basis functions.

[0018] Furthermore, the processor unit is configured to recursively adapt the coefficient vector at predetermined intervals during operation of the motor vehicle to determine an adapted MPC cost function. These predetermined intervals can be predefined temporal intervals or intervals based on external circumstances.

[0019] Furthermore, the processor unit is configured to minimize the adapted MPC cost function in order to determine a speed control for the motor vehicle with respect to the prediction horizon as a function of the respectively adapted approximated gradient and the weighted time term, so that a minimization independent of the drive train variant of the motor vehicle is possible.

[0020] The energy term is based on the total wheel force and the vehicle speed and describes the energy temporarily stored in the vehicle. The total wheel force can be expressed as positive for propulsion and negative for braking. The temporarily stored energy can, for example, be the chemical energy bound in gasoline in a combustion engine and / or the state of charge in a battery.

[0021] Energy consumption and travel time are preferably evaluated and weighted at the end of the horizon. This means that both the energy term and the time term are only relevant at the last waypoint of the horizon. Thus, the generalized energy term of the buffered energy is represented by the vehicle speed and wheel force, or by the gradient of the buffered energy as a function of the vehicle speed and total wheel force.

[0022] Furthermore, the gradient is approximated by an approximation function with a coefficient vector and basis functions. Furthermore, during operation of the vehicle, the coefficient vector is recursively adapted at specified intervals to determine an adapted MPC cost function.

[0023] In this case, the processor unit does not minimize the MPC cost function based on the gradient of the energy term itself, but uses an approximation function with a coefficient vector and basis functions, whereby the coefficient vector is adapted to changing conditions.

[0024] Thus, according to the invention, the coefficient vector is adapted during operation. This avoids the use of constant, inaccurate values ​​for the coefficients of the coefficient vector, which only inaccurately reflect the buffered energy or the gradient of the buffered energy as a function of the vehicle speed and total wheel force.

[0025] This allows for the continuous recursive adjustment of these coefficients, values ​​that are only vaguely known or change during operation, to the actual conditions. This makes it possible to use an initial, vague input parameterization / basic parameterization of the coefficients, saving computational effort and time. This simplifies and reduces the necessary computational effort in the processor unit.

[0026] Furthermore, the coefficients are continuously adjusted to current consumption and loss characteristics. This adaptation prevents speed planning based on outdated values. Furthermore, it is taken into account that the coefficients depend on the vehicle's service life, which a constant input / basic parameterization cannot account for.

[0027] The processor unit according to the invention can be used for all available drive trains, ie, for combustion engines, hybrid vehicles, electric vehicles with a different number of electric motors, fuel cell vehicles, etc. For implementation, the processor unit can be used in any motor vehicle and, for example, transmit the determined speed to a power electronics / control unit.

[0028] Furthermore, the energy term or gradient can be based on the vehicle speed and wheel force as well as on an engine temperature.

[0029] In a further development, the motor vehicle has a battery, where the energy term is based on the vehicle speed and wheel force, as well as the engine temperature and battery charge level. This allows the energy term to be more accurately represented in different situations.

[0030] In further development, the vehicle speed and thus the time term are calculated according to a longitudinal dynamics model of the vehicle. The longitudinal dynamics model can be calculated using: m=dvdt=Fwhl−Fslope−Froll−Fair where m is the vehicle mass and F slopeis the gradient resistance force, which describes a longitudinal component of the force of gravity acting on the motor vehicle when driving uphill or downhill and is dependent on the gradient of the road, F air the air resistance of the motor vehicle, where the air resistance depends on the speed of the motor vehicle; and F roll the rolling resistance force, which is an effect of the deformation of the tires when rolling and depends on the load on the wheels and thus on the gradient of the road and F whlrepresents the total wheel force. This representation allows the longitudinal dynamics model to be applied to any powertrain. Furthermore, current state variables can be measured, corresponding data can be recorded, and fed into the MPC cost function. For example, route data from an electronic map can be regularly updated for a prediction horizon in front of the vehicle.

[0031] Such a representation allows the longitudinal dynamics model to be applied to any powertrain. Furthermore, current state variables can be measured, corresponding data can be recorded, and fed into the MPC cost function J. For example, route data from an electronic map can be regularly updated for a prediction horizon in front of the vehicle.

[0032] The route data can include, for example, gradient information, curve information, and information about speed limits.

[0033] This general longitudinal dynamics model can thus incorporate vehicle parameters as well as knowledge of the road topography ahead, such as curves and gradients. Furthermore, knowledge of speed limits on the road ahead can also be incorporated into the longitudinal dynamics model.

[0034] However, the general longitudinal dynamics model is independent of a specific powertrain.

[0035] In a further embodiment, the MPC cost function has a comfort term weighted with a third weighting factor, which contains the predicted value of a wheel force change during the prediction horizon, wherein the processor unit is configured to generate a speed control for the motor vehicle by minimizing the adapted MPC cost function as a function of the time term, as a function of the adapted gradient and as a function of the comfort term.

[0036] To ensure a comfortable ride, a comfort term is added to the MPC cost function to ensure a smooth ride.

[0037] The wheel force F whl during the prediction horizon in the operating points of the vehicle by the following difference: Fwhl=Ftrc−Fbrk with F trcas the traction force exerted by the engine or engines on the wheels of the motor vehicle and F brk as braking force.

[0038] In a further embodiment, the processor unit is designed to adapt the coefficient vector if the residue between the gradient of the temporarily stored energy at a given time and the energy determined by the approximation function at the same time exceeds a predefined threshold value.

[0039] In particular, the processor unit is designed to form the residue by: (dEsto(k+1)−xT(k+1)a(k)) where dE sto (k + 1) represents the gradient of the stored energy at time k + 1 and x T (k + 1) represents a vector with basis functions at time k + 1 and a(k) represents the coefficient vector at a previous time k.

[0040] The adaptation does not necessarily have to be carried out at each new point in time, but can be situation-dependent or condition-dependent, for example only when the residue (dE sto (k + 1) - x T (k + 1)a(k)) exceeds a certain threshold value.

[0041] Alternatively or additionally, the processor unit can be designed to adapt the coefficient vector at predetermined time intervals. In a further embodiment, the processor unit is designed to calculate the gradient dE sto (F whl ,v) by a polynomial approximation with polynomials based on the speed v and the wheel force F whl to approximate.

[0042] This means that dE sto (F whl ,v) can be represented by: dEsto(Fwhl,v)≈a0+a1Fwhl+a2v+a3Fwhlv+a4Fwhl2+a5Fwhl2v+⋯=(1Fwhlv…)(a0a1a2⋮):=xTa

[0043] Alternatively, the processor unit can be designed to calculate the gradient dE sto (F whl ,v) by radial basis function networks with basis functions based on the speed v and the wheel force F whl to approximate.

[0044] Radial basis function networks with basis functions can be given by: dEsto(Fwhl,v)≈a0+a1φ1(Fwhl,v)+a2φ2(Fwhl,v)+⋯=(1φ1φ2…)(a0a1a2⋮):=xTa with Gaussian basis functions φi(Fwhl,v)=exp(−β‖(Fwhlv)−ci‖2) and a suitably chosen shape parameter β as well as suitably chosen support points c i Furthermore, the processor unit can be configured to adapt the coefficient vector using a recursive algorithm. The adaptation of the coefficient vector can be performed, in particular, using a recursive least squares method.

[0045] Furthermore, the problem is solved by a method for predictive control of a motor vehicle with the steps: - generating an MPC cost function, wherein the MPC cost function comprises a time term weighted with a first weighting factor, which contains a travel time predicted independently of the drive train variant of the motor vehicle, which the motor vehicle requires to cover a total distance predicted within a prediction horizon, and wherein the MPC cost function comprises an energy term weighted with a second weighting factor, which is based on energy temporarily stored as a function of at least the vehicle speed and the wheel force, so that the energy term provided is independent of the drive train variant of the motor vehicle, - Determining a gradient of the stored energy and using the gradient to determine the energy term, - Approximating the gradient by an approximation function with coefficient vector and basis functions, - Recursive adaptation of the coefficient vector during operation of the motor vehicle at specified intervals to determine an adapted MPC cost function, - Minimizing the adapted MPC cost function for determining a speed control for the motor vehicle with respect to the prediction horizon depending on the respectively adapted approximated gradient and the weighted time term, so that an independent minimization of the powertrain variant of the motor vehicle is possible.

[0046] The advantages and advantageous configurations of the processor unit can be transferred to the method.

[0047] In particular, the gradient can be approximated by a polynomial approximation with polynomials based on the velocity and wheel force, or by radial basis function networks with basis functions based on the velocity and wheel force. Other approximation methods are also possible.

[0048] Furthermore, the coefficient vector can be adapted using a recursive algorithm, in particular a recursive least squares method.

[0049] Furthermore, the object is achieved by a motor vehicle with a method as described above and / or a processor unit as described above, wherein the motor vehicle has a control unit for adjusting sensors and / or actuators, wherein the control unit is designed to receive the determined speed control for the motor vehicle and is designed to generate a trajectory based on the speed control and to implement it by means of the actuators and / or sensors.

[0050] In this case, trajectory can also be understood as just speed parameters for the implementation of the speed control or as a complete trajectory planning, for example with generation of the complete trajectory to be followed, for example for an autonomously driving motor vehicle.

[0051] Further features and advantages of the present invention will become apparent from the following description with reference to the accompanying figures. Fig. 1: a motor vehicle with a processor unit according to the invention, Fig. 2: an approximated gradient, Fig. 3: the approximated gradient at a later time, Fig. 4: the approximated gradient at a later time, Fig. 5: the method according to the invention.

[0052] Fig. Figure 1 shows a motor vehicle 2 with a processor unit 1 according to the invention. The motor vehicle 2 has a processor unit 1 for predictive control of a motor vehicle 2.

[0053] The processor unit 1 comprises a memory unit 4 and a detection unit, in particular a sensor system 3, for detecting status data relating to the motor vehicle 2 and, if applicable, route data ahead.

[0054] Furthermore, the motor vehicle 2 comprises a drive train, which can be designed, for example, as an electric machine that can be operated as a motor and as a generator, and an energy supply unit 5, which can be designed, for example, as a battery / gasoline tank / hydrogen tank for an internal combustion engine or hybrid drive and / or fuel cell.

[0055] The storage unit 4 has an MPC (Model Predictive Control) cost function J to be minimized.

[0056] The MPC (Model Predictive Control) cost function J has a general longitudinal dynamics model that is valid for all powertrains or for any motor vehicle 2. The general longitudinal dynamics model is formed by: dEkinds=Fwhl−Fslope−Froll−Fair where E kin describes the kinetic energy of the vehicle, F slope is the gradient resistance force, which describes a longitudinal component of the force of gravity acting on the motor vehicle 2 when driving uphill or downhill, and is dependent on the gradient of the road, F air the air resistance force of the motor vehicle 2, wherein the air resistance force depends on the speed of the motor vehicle 2; and F roll the rolling resistance force, which is an effect of the deformation of the tires when rolling and depends on the load on the wheels and thus on the gradient of the road and F whl represents the wheel force.

[0057] The time dependence is converted into path dependence as follows: dtds=1v Through this general description, current state variables can be measured, for example, by the sensor system 3 and / or received by the motor vehicle 2 and fed to the MPC cost function J.

[0058] For example, route data / surroundings data from an electronic map can be regularly updated for a prediction horizon, for example, a few hundred meters, in front of motor vehicle 2. The surroundings data can include, for example, gradient information, curve information, and information about speed limits. Furthermore, a curve curvature can be converted into a speed limit for motor vehicle 2 using a maximum permissible lateral acceleration. Furthermore, the location of motor vehicle 2 can be determined, in particular via a GPS signal for precise localization on the electronic map.

[0059] Using the longitudinal dynamics model, a time term Z weighted with a first weighting factor q1 is determined, which contains a travel time predicted independently of the drive train variant of the motor vehicle 2, which the motor vehicle 2 requires to cover the entire distance predicted within a prediction horizon.

[0060] The time term Z forms a first term of the MPC cost function J: Z=q1(1−λ)t(sN)with(λ)∈[0,1] where λ is the adjustment parameter for the trade-off between time and energy, here the energy consumption, and s N represents the waypoint at the end of the prediction horizon, and t is the time and t(s N ) thus the elapsed travel time at waypoint s N and q1 represents the weighting factor.

[0061] The MPC cost function J thus contains, as its first term, a travel time weighted with the first weighting factor q1 and predicted according to the longitudinal dynamics model, which vehicle 2 requires to cover the predicted distance. This means that, depending on the choice of weighting factors, a low speed is not always evaluated as optimal, thus eliminating the problem that the resulting speed is always at the lower end of the permitted speed range.

[0062] By using such a general longitudinal dynamics model, the time term Z is independent of the powertrain currently in use.

[0063] Furthermore, the MPC cost function J has an energy term E weighted with a second weighting factor, which is based on the temporarily stored energy E sto (F whl ,v) depending on at least the vehicle speed v and the wheel force F whlso that the energy term E provided is independent of the drive train variant of the motor vehicle: E=q2λEsto(sN)with(λ)∈[0,1] where λ is the tuning parameter for the time-energy trade-off and q2 is the weighting factor, and where the energy term E has a negative impact on the cost function J.

[0064] Furthermore, the MPC cost function J to be minimized has a third convenience term K: K=q3∑i=1N(dFwhl(si))2 so that the MPC cost function J = Z - E + K is formed from: J=q1(1−λ)t(sN)−q2λEsto(sN)+q3∑i=1N(dFwhl(si))2 where λ is the tuning parameter for the time-energy trade-off and q1, q2,q3 are positive weighting factors.

[0065] The cost function J therefore contains a further term as comfort term K. It should be noted that this term only depends on the wheel force change, ie the wheel force gradient (dFwhl (s i )) in the waypoints (s i ) and no longer depends on the longitudinal dynamics model.

[0066] The wheel force can be defined as F whl as the difference from the traction force F trc , which is exerted by the engine or engines on the wheels of the motor vehicle 2 and the braking force F brk be educated: Fwhl=Ftrc−Fbrk where F trc the traction force exerted by the engine or engines on the wheels of the motor vehicle and F brk the braking force is.

[0067] Furthermore, the energy term E as the second term depends only on the vehicle speed v and the wheel force F whl and no longer from the longitudinal dynamics model.

[0068] The energy term E is determined by the integral of the gradient dE sto (F whl ,v) is represented by the path, so that the MCP cost function J = Z - E + K is formed from: J=q1(1−λ)t(sN)−q2λEsto(Fwhl,ν)(sN)+q3∑i=1N(dFwhl(si))2 where λ is the tuning parameter for the time-energy trade-off and q1, q2,q3 are positive weighting factors.

[0069] Thus, the gradient of the energy term E used to minimize the MCP cost function J is expressed as follows: dEsto(Fwhl,ν) with F whl : wheel force, v: vehicle speed. where by converting time dependence into path dependence, dEstods=1νdEstodt=def1νdEsto(Fwhl,ν)

[0070] Furthermore, the gradient dEsto(Fwhl,ν) by an approximation function with coefficient vector a and basis functions x T These can be simple functions, for example polynomials in the speed v and the wheel force F whl so that: dEsto(Fwhl,ν)≈a0+a1Fwhl+a2ν+a3Fwhlν+a4Fwhl2+a5Fwhl2ν+⋯ =(1 Fwhl ν ⋯)(a0a1a2⋮):=xTa

[0071] Determination of the gradient dE sto (F whl ,v) is done in this case via the coefficients a i the polynomial approximation. These coefficients can be roughly initialized by an input parameterization / basic parameterization, for example, based on an engine map and on powertrain loss maps.

[0072] Fig. 2 shows the gradient dE approximated in this way sto (F whl ,v) as area for a drive train depending on the speed v in [m / s] and the wheel force F whl in [kN]. The gradient dE sto in [kW] as an approximated function (Approx_0) with an input parameterization / basic parameterization for the coefficient vector, ie at a time T=0 compared to the real gradients true_dE sto (Fwhl ,v).

[0073] Subsequently, at a time k, an adaptation of the coefficients a i made.

[0074] Here, dE sto (k) the currently estimated gradient of the energy stored in the motor vehicle 2, which is determined, for example, via the current fuel consumption and / or the current battery power at time k, and x T (k) = (1 F whl (k) v(k) ...) the current regressor in relation to the vehicle speed v(k) determined at the current time k and a current wheel force F estimated at the time k using engine and brake torques whl (k) and a(k) is the coefficient vector given at time k.

[0075] The adaptation of the coefficient vector a(k) at the transition to time k+1 can be achieved using a Recursive Least Squares method according to the formula a(k+1)=a(k)+P(k)x(k+1)ρ+xT(k+1)P(k)x(k+1)(dEsto(k+1)−xT(k+1)a(k)) where ρ ≤ 1 represents a suitably chosen forgetting factor and the update of the matrix P according to P(k+1)=1ρ(P(k)−P(k)x(k+1)ρ+xT(k+1)P(k)x(k+1)xT(k+1)P(k)) is carried out. The adaptation does not necessarily have to be carried out at each new point in time, but can be situation-dependent or condition-dependent, for example only when the residue (dE sto (k + 1) - x T (k + 1)α(k)) exceeds a certain threshold value.

[0076] This shows Fig. 3 the successive adaptation of the approximated gradient dE sto (F whl ,v) as area for a drive train depending on the speed v in [m / s] and the wheel force F whl in [kN]. The gradient dE stoin [kW] as an approximated function (Approx_1) with adapted coefficient vector, ie at a time T=160 s compared to the real gradients true_dE sto (F whl ,v). The determined values ​​dE sto (F whl ,v) are given as values ​​W.

[0077] Furthermore, the Fig. 4 the successive adaptation of the approximated gradient dE sto (F whl ,v) as area for a drive train depending on the speed v in [m / s] and the wheel force F whl in [kN]. The gradient dE sto in [kW] as an approximated function (Approx_2) with adapted coefficient vector, ie at a time T=360s compared to the real gradients true_dE sto (F whl ,v). The determined values ​​dE sto (F whl ,v) are given as values ​​W.

[0078] It can be seen that with progressive adaptation the initially only roughly matching area dE sto (F whl ,v) the actual area true_dE sto (F whl ,v) by continuously adjusting the coefficients. Such an adaptation makes it possible to start with a rough basic parameterization. Furthermore, it takes into account that the coefficients depend on the service life of the vehicle 2, which cannot be accounted for by a constant input parameterization / basic parameterization.

[0079] The processor unit 1 then minimizes the adapted cost function J for determining a speed control for the motor vehicle 2 with respect to the prediction horizon, regardless of the drive train variant of the motor vehicle 2.

[0080] Fig.Figure 5 shows a method according to the invention in another embodiment. In a first step S1, the MPC cost function J is first generated. The MPC cost function J has the time term Z weighted with a first weighting factor q1, which contains a travel time predicted independently of the drivetrain variant of the motor vehicle 2, which the motor vehicle 2 requires to cover the entire distance predicted within a prediction horizon. The weighted time term Z is configured as above.

[0081] Furthermore, the energy term E weighted with a second weighting factor q2 is generated, which is based on the energy term E depending on at least the vehicle speed v and the wheel force F whl stored energy E sto (F whl ,v) so that the energy term (E) provided is independent of the drive train variant of the motor vehicle (2),

[0082] In this case, the gradient dE sto (F whl ,v) the stored energy is determined and used to determine the energy term E sto (F whl ,v). Furthermore, the MCP cost function J can also include the convenience term K.

[0083] In a second step S2, the gradient dE sto (F whl ,v) is approximated by an approximation function with coefficient vector and basis functions.

[0084] The gradient dE sto (F whl ,v) by radial basis function networks with basis functions based on the speed v and the wheel force F whl approximated.

[0085] Radial basis function networks with basis functions can be given by: dEsto(Fwhl,ν)≈a0+a1φ1(Fwhl,ν)+a2φ2(Fwhl,ν)+⋯= (1 φ1 φ2 ⋯)(a0a1a2⋮):=xTa with Gaussian basis functions φi(Fwhl,ν)=exp(−β‖(Fwhlν)−ci‖2) and a suitably chosen shape parameter β as well as suitably chosen support points c i . In a third step S3, the coefficient vector a is adapted during operation of the motor vehicle 2 to determine an adapted MPC cost function J using a recursive least squares method.

[0086] In a fourth step S4, the adapted MPC cost function J for determining a speed control for the motor vehicle 2 is minimized with respect to the prediction horizon, independent of the drive train variant of the motor vehicle 2.

[0087] In a fifth step S5, the speed control is transmitted to a control unit, which generates a trajectory based on the speed control and implements it by means of the actuators and / or sensors 3. List of reference symbols 1 processor unit 2 motor vehicles 3 Sensor system 4 storage unit 5 Energy supply unit J MPC cost function Z time term K Comfort term J Cost function v Vehicle speed F whl Wheel force dF whl Wheel force change or gradient E sto (F whl ,v) stored energy dE sto (F whl ,v) Gradient E energy term

Claims

[1] Processor unit (1) for predictive control of a motor vehicle (2), characterized by , that a storage unit (3) with an MPC (Model Predictive Control) cost function (J) is provided, wherein the MPC cost function (J) comprises a time term (Z) weighted with a first weighting factor (q1), which contains a travel time predicted independently of the drive train variant of the motor vehicle (2) that the motor vehicle (2) requires to cover an entire distance predicted within a prediction horizon, and wherein the MPC cost function (J) comprises an energy term (E) weighted with a second weighting factor (q2) which is based on a function of at least the vehicle speed (v) and the wheel force (F whl ) temporarily stored energy (E sto (F whl,v)) so that the energy term (E) provided is independent of the drive train variant of the motor vehicle (2), wherein the processor unit (1) is designed to Gradients (dE sto (F whl ,v)) of the stored energy (E sto (F whl ,v)) and to use it for determining the energy term (E), and wherein the processor unit (1) is designed to determine the gradient (dE sto (F whl ,v)) by an approximation function with coefficient vector and basis functions, and wherein the processor unit (1) is designed to adapt the coefficient vector recursively at predetermined intervals during operation of the motor vehicle (2) in order to determine an adapted MPC cost function (J), and wherein the processor unit (1) is designed to minimize the adapted MPC cost function (J) in order to determine a speed control for the motor vehicle (2) with respect to the prediction horizon as a function of the respectively adapted approximated gradient (dE sto (F whl ,v)) and the weighted time term (Z), so that a minimization of the cost function (J) is possible regardless of the drive train variant of the motor vehicle (2). [2] Processor unit (1) according to claim 1, characterized by that the processor unit (1) is designed to adapt the coefficient vector when the residue between the actual gradient (dE sto (k + 1)) of the stored energy (E sto (F whl ,v)) at a given time (k + 1) and the value determined by the approximation function at the same time exceeds a predefined threshold. [3] Processor unit (1) according to claim 2, characterized by that the processor unit (1) is designed to form the residue by: (dEsto(k+1)−xT(k+1)a(k)) where (dE sto (k + 1)) the actual gradient of the stored energy (E sto (k + 1)) at time k + 1 and x T (k + 1) represents the vector with basis functions at time k + 1 and a(k) represents the coefficient vector at a previous time k. [4] Processor unit (1) according to claim 1, characterized by that the processor unit (1) is designed to adapt the coefficient vector at predetermined time intervals. [5] Processor unit (1) according to one of the preceding claims, characterized by that the processor unit (1) is designed to determine the gradient (dE sto (F whl ,v)) by a polynomial approximation with polynomials based on the speed (v) and the wheel force (Fwhl ) to approximate. [6] Processor unit (1) according to one of the preceding claims 1 to 4, characterized by that the processor unit (1) is designed to determine the gradient (dE sto (F whl, v)) by radial basis function networks with basis functions based on the speed (v) and the wheel force (F whl ) to approximate. [7] Processor unit (1) according to one of the preceding claims, characterized by that the processor unit (1) is designed to adapt the coefficient vector with a recursive algorithm. [8] Method for predictive control of a motor vehicle (2) characterized by the steps: - generating an MPC cost function (J), wherein the MPC cost function (J) comprises a time term (Z) weighted with a first weighting factor (q1), which contains a travel time predicted independently of the drive train variant of the motor vehicle (2) which the motor vehicle (2) requires to cover an entire distance predicted within a prediction horizon, and wherein the MPC cost function (J) comprises an energy term (E) weighted with a second weighting factor (q2), which is based on an energy term dependent on at least the vehicle speed (v) and the wheel force (F whl ) temporarily stored energy (E sto (F whl ,v)) so that the energy term (E) provided is independent of the drive train variant of the motor vehicle (2), - Determination of a gradient (dE sto (F whl ,v)) of the stored energy and using the gradient (dE sto (F whl,v)) for the determination of the energy term E - Approximating the gradient (dE sto (F whl ,v)) by an approximation function with coefficient vector and basis functions, - Recursively adapting the coefficient vector during operation of the motor vehicle (2) at predetermined intervals to determine an adapted MPC cost function, - minimizing the adapted MPC cost function (J) for determining a speed control for the motor vehicle (2) with respect to the prediction horizon as a function of the respectively adapted approximated gradient and the weighted time term (Z), so that a minimization of the cost function (J) is possible independently of the drive train variant of the motor vehicle (2). [9] Method (1) according to claim 8, characterized by that an adaptation of the coefficient vector is carried out when the residue between the gradient (dE sto(k + 1)) of the temporarily stored energy at a given time and the value determined by the approximation function at the same time exceeds a predefined threshold. [10] Method according to claim 9, characterized by that the residue is formed by: (dEsto(k+1)−xT(k+1)a(k)) where dE sto (k + 1) represents the gradient of the stored energy at time k + 1 and x T (k + 1) represents the vector with basis functions at time k + 1 and a(k) represents the coefficient vector at a previous time k. [11] Method according to claims 8 to 10, characterized by that an adaptation of the coefficient vector is carried out at predetermined time intervals. [12] Method according to one of the preceding claims 8 to 11, characterized by that the gradient dE sto (F whl,v) by a polynomial approximation with polynomials based on the speed (v) and the wheel force (F whl ) is approximated. [13] Method according to one of the preceding claims 8 to 12, characterized by that the gradient dE sto (F whl ,v) by radial basis function networks with basis functions based on the speed (v) and the wheel force (F whl ) is approximated. [14] Method according to one of the preceding claims 8 to 13, characterized by that the coefficient vector is adapted using a recursive algorithm. [15] Method according to claim 14, characterized by that the recursive algorithm is developed as a Recursive Least Squares method. [16] Motor vehicle (2) with a method according to one of the preceding claims 8 to 15 and / or a processor unit (1) according to one of the preceding claims 1 to 7, wherein the motor vehicle (2) has a control unit for setting sensors and / or actuators, wherein the control unit is designed to receive the determined speed control for the motor vehicle (2) and is designed to generate a trajectory based on the speed control and to implement it by means of the actuators and / or sensors.

Images