FR3129497A1 - DYNAMIC SYSTEM CONTROL SYSTEM FOR MOTOR VEHICLES, METHOD AND PROGRAM BASED ON SUCH CONTROL SYSTEM - Google Patents

Abstract

The invention relates to a control system for a physical system (SP) of a motor vehicle, the control system comprising - a predictive control means (FFN) of the model type, - an on-board model (M) configured to obtain data from prediction of states of the physical system (SP),the predictive control means (FFN) being configured to determine an optimal control signal () and a reference signal () on the basis of predicted states () of the on-board model ( M),- a buffer block (BF) configured to buffer the control signal () and the reference signal (),- a robust feedback regulator (FBR) receiving said buffered reference signal () and a sensor measurement signal at the output of the physical system (SP). The invention also relates to a method and a program based on such a system. Figure 4

Description

DYNAMIC SYSTEM CONTROL SYSTEM FOR MOTOR VEHICLES, METHOD AND PROGRAM BASED ON SUCH CONTROL SYSTEM

The invention relates to the field of systems for controlling dynamic systems for motor vehicles, in particular for predictive type controls.

The invention finds a particularly advantageous application in the context of controlling the air loop of a heat engine, and is also envisaged for controlling the air loop of the cathode of a fuel cell.

The genericity of the proposed solution allows its application to a large number of robust optimal control problems encountered in the automotive field.

A control solution of the prior art is based on a direct chain control (or " feedforward " in English) by an inverse model MI, of the type presented in , for actuators of a physical system SP.

This type of control law considers the nominal model of a controlled system and uses it to calculate a control so that . Unfortunately, this control law cannot handle the parametric uncertainties of the system. In addition, it is not sensitive to measurement noise. It is often accompanied by a feedback regulator developed in other solutions.

Another control solution of the prior art is based on a direct chain control FF and retroactive FB (or " feedback " in English) of the actuators of the system, of the type presented in .

The FF direct chain control allows you to format a command necessary to achieve an objective or benchmark . A feedback regulator allows to drive the output of the system towards the reference taking into account parametric uncertainties as well as measurement noise .

Another control solution of the prior art is based on a predictive control based on a model, of the MPC type (“ Linear Model Predictive Control ” in English) called linear or non-linear (NMPC – “ Nonlinear Model Predictive Control”). in English), of the type presented in . In this figure, O is an optimizer, SE is an embedded simulation module integrating a model of the system MS.

This predictive control implies an optimal online control law. Its operating principle is as follows:

It uses the model of the MS system to calculate the evolution of its outputs and states; an EE state estimator filters measurement noise and makes it possible to estimate the current states of the system; at each time step, an optimization determines the control minimizing a cost function FC over a finite time horizon; constraints on commands and/or states are taken into account; and a first element E1 of the predicted command is then applied to the system.

Unfortunately, the solutions of the prior art are not fully satisfactory, in particular for the following reasons:

The parametric dispersions of the system are not taken into account during the design and/or operation of the regulators, their effects are noticed late after construction.

Linear control laws do not make it possible to explicitly take into account constraints on the control and/or on the states.

MPC controllers require knowledge of the state of the system and therefore of state estimators that are sensitive to measurement noise and parametric uncertainties of the system.

Predictive controllers implemented in real time require significant computing time to optimize control, especially for nonlinear and/or multivariable systems.

The invention aims to overcome the problems of the prior art, and in particular to propose a solution taking into account the parametric dispersions of the system, the constraints on the control and/or on the states, measurement noises and parametric uncertainties of the system.

To achieve this objective, the invention proposes a control system for a motor vehicle physical system, the control system comprising
- a means of predictive control based on a model,
- an embedded model configured to obtain state prediction data from the physical system,
the predictive control means being configured to determine an optimal control signal and a reference signal based on predicted states of the on-board model,
- a buffer block at the output of said predictive control means, configured to buffer the control signal and the reference signal,
a robust feedback regulator at the output of the buffer block, and upstream of the physical system, the robust feedback regulator receiving said buffered reference signal which it compares to the sensor measurement signal at the output of the physical system.

Advantageously, a state estimator is not necessary, the predictive control means using the predicted states of the on-board model that it integrates.

In addition, the performance of the control law is not very dependent on the optimization time used by the predictive control means thanks to the buffer block.

In addition, the predictive control means can integrate the knowledge of the future reference signals thanks to the on-board model. The effect of the parametric dispersions of the system is taken into account and countered by the robust feedback regulator.

According to a variant, the robust feedback regulator additionally receives a noise signal. This also allows the measurement noise to be taken into account in the regulation.

According to a variant, the control system further comprises a control and reference optimization means on the basis of the formulas:

,

,

with a command vector ,

and a setpoint vector ,

considering

a time , Or is a sampling period and is a positive integer;
a prediction horizon , Or is a number of samples composing this horizon;
a calculation time , with , Or is a minimum integer number of periods necessary for optimization.

This makes it possible to have a precise control optimization value.

According to a variant, the buffer block comprises storage means configured to store, during a current optimization of the predictive control means, the commands and references obtained at a previous optimization.

This makes it possible to take advantage of a previous optimization for a following command.

According to a variant, the buffer block further comprises a selector means configured to successively apply the commands and references coming from the storage means.

This makes it possible to apply optimized commands.

The invention further relates to a control method for a motor vehicle physical system, the control method comprising the following steps:
- determining, by model-type predictive control, an optimal control signal and a reference signal on the basis of predicted states of an on-board model,
- buffer the control signal and the reference signal,
- implementing robust feedback regulation based on said buffered reference signal compared to a sensor measurement signal at the output of the physical system.

According to a variant, the control method further comprises a control and reference optimization step based on the formulas:

;

,

with a command vector ,

and a setpoint vector ,

considering

a time , Or is a sampling period and is a positive integer;
a prediction horizon , Or is a number of samples composing this horizon;
a calculation time , with , Or is a minimum integer number of periods necessary for satisfactory optimization.

According to a variant, the buffering step comprises a storage step for storing, during a current optimization of the step of determining by predictive control, the commands and references obtained at a previous optimization.

According to a variant, the buffering step comprises a step of filtering by selector to successively apply the commands and references resulting from the storage step.

Another object of the invention relates to a computer program comprising program code instructions for the execution of the steps of a control method according to the invention, when said program is running on a computer.

The invention further relates to a motor vehicle comprising a control system according to the invention.

The invention will be further detailed by the description of non-limiting embodiments, and on the basis of the appended figures illustrating variants of the invention, among which:

schematically illustrates a control solution in direct chain by inverse model of the prior art;

schematically illustrates a direct FF and retroactive chain control solution of the prior art;

schematically illustrates a prior art linear or non-linear model predictive control solution.

schematically illustrates a control system according to a preferred variant of the invention;

schematically illustrates a means of storing a buffer block of a control system according to the preferred variant of the invention; And

schematically illustrates a selector of a buffer block of a control system according to the preferred variant of the invention.

A control system may include a control algorithm.

The common problem of model-based predictive control is its sensitivity to parametric dispersions of the controlled system, measurement noise, and the computation time it requires. Unlike the conventional nonlinear NMPC predictive control scheme producing a control signal that is both predictive (" feedforward ") and retroactive (" feedback "), the control scheme of the invention proposes to use an NMPC algorithm to generate a predictive type command and associate it with a robust FBR feedback linear regulator. The NMPC algorithm determines an optimal command to be applied to the physical system SP as well as the nominal trajectory that its output must follow.

The robustness of the regulation is ensured by the feedback regulator FBR whose output is added to that of the NMPC command. This robust FBR regulator makes it possible to react to the effect of parametric dispersions of the system as well as unmeasured disturbances. The measurement noise as well as the parametric dispersions must be taken into account during the synthesis of the FBR feedback regulator. This synthesis can be of the type , -synthesis, CRONE, QFT, etc.

The proposed control algorithm takes into account the computation time of the NMPC optimization. The command and reference signals are updated when each new optimization completes. During an optimization, the control and reference signals resulting from the previous optimization are applied.

The advantage of the invention lies in particular in the robustness with respect to parametric uncertainties and noise; the optimization calculation time taken into account; taking time constraints into account when determining the optimal NMPC predictive control; taking into account the frequency constraints during the synthesis of the feedback regulator FBR; the dynamic inversion of the nominal nonlinear model of the system controlled using the NMPC command; possible anticipation taking into account future setpoint and disturbance signals.

The proposed invention consists in associating an NMPC algorithm generating a predictive control and a robust linear regulator FBR generating a feedback control.

The retroactive regulator is based on:
- an uncertain model of the linear behavior of the system to be controlled. It makes it possible to take into account potential operating points, nonlinear behavior, parametric variations;
- frequency constraints relating to the performance of the closed loop relating to the setpoint signals, disturbances, commands and outputs;
- off-line optimization of a constrained cost function to tune the controller parameters;
- a reference signal compatible with the achievable performance of the system to be controlled.

The NMPC algorithm is based on:
- an embedded model M (non-linear, linear with varying parameters, etc.) finely characterizing the nominal behavior of the system to be controlled;
- a cost function (quadratic, linear, non-linear) allowing performance optimization;
- constraints relating to the command, the states and the outputs to be controlled;
- a time step allowing a good discretization of the embedded model and in conformity with the capacities of the computer;
- a prediction horizon allowing a satisfactory calculation of the cost function, in particular covering the duration of the transient phenomena to be controlled;
- an optimizer allowing online optimization of the predictive control with reasonable calculation times.

The association of the NMPC algorithm and the robust linear regulator is permitted thanks to the logic described below:
- either time , Or is the sampling period and is a positive integer;
- either the prediction horizon of the NMPC algorithm , Or is the number of samples composing this horizon;
- either the calculation time of the NMPC algorithm with , Or is the minimum integer number of periods necessary for optimization.

The results of the optimization triggered at a moment are the control vector and the setpoint vector .

To ensure the continuity of the order, during an optimization in progress, the first commands and applied reference signals that were obtained during the previous optimization are: And .

This operation is made possible thanks to various sub-parts and in particular to a buffer block composed of the following two blocks:
- a storage space S1, illustrated in , which allows during the current NMPC optimization launched at instant to keep orders and references obtained by the previous NMPC optimization;
- an S2 selector, illustrated in , which successively applies to the controlled system the commands and references from the storage block S1.

The operation of these two blocks is carried out with .

Figure 6 corresponds to And . The dotted line of the graph, i.e. corresponds to the start of the current optimization.

At moments And the control and reference samples used are the result of the previous optimization (launched at instant ) at the same time. The solid vertical line of the graph indicates the end of the current optimization.

In order for the constraint on the predictive control signal used by the NMPC optimization to be adapted, the level of the control actually applied to the system (which also depends on the feedback control) is regularly transmitted to the NMPC algorithm.

The invention further relates to a computer program comprising program code instructions for the execution of the steps of a control method as described above, when said program is running on a computer. In particular, the invention can be integrated into a motor vehicle control computer.

The invention also relates to a motor vehicle comprising a control system as described previously.

The advantage of the invention in use lies in particular in an optimization of the reference trajectories for a gain in performance; maintaining performance thanks to robustness with respect to parametric dispersions; mastery of sensitivity to measurement noise; a determination of a generic methodology adapted to the constraints of a large class of systems to be controlled (nonlinear, multivariable, with dispersed parameters, etc.); a real-time implementation that is compatible with the performance of serial automotive computers.

Claims (10)

A control system for a motor vehicle physical system (SP), the control system comprising
- a means of predictive control (FFN) based on model,
- an embedded model (M) configured to obtain state prediction data from the physical system (SP),
the predictive control means (FFN) being configured to determine a control signal ( ) and a reference signal ( ) optimal based on predicted states ( ) of the on-board model (M),
- a buffer block (BF) at the output of said predictive control means (FFN), configured to buffer the control signal ( ) and the reference signal ( ),
- a robust feedback regulator (FBR) at the output of the buffer block (BF), and upstream physical system (SP), the robust feedback regulator (FBR) receiving said reference signal ( ) buffered which it compares to the sensor measurement signal output from the physical system (SP). A control system according to claim 1, further comprising means for controlling and reference optimization based on the formulas:
,
,
with a command vector ,
and a setpoint vector ,
considering
a time , Or is a sampling period and is a positive integer;
a prediction horizon , Or is a number of samples composing this horizon;
a calculation time , with , Or is a minimum integer number of periods necessary for optimization. Control system according to any one of Claims 1 to 2, characterized in that the buffer block (BF) comprises a storage means (S1) configured to store, during a current optimization of the predictive control means (FFN) , the orders ( ) and references ( ) obtained at a previous optimization. Control system according to Claim 3, characterized in that the buffer block (BF) further comprises selector means (S2) configured to successively apply the commands and references coming from the storage means (S1). Control method for a motor vehicle physical system (SP), the control method comprising the following steps:
- determining, by type-to-model predictive control, a control signal ( ) and a reference signal ( ) optimal based on predicted states ( ) of an embedded model (M),
- buffer the control signal ( ) and the reference signal ( ),
- implementing robust feedback regulation based on said reference signal ( ) buffered compared to a sensor measurement signal output from the physical system (SP). Control method according to claim 5, further comprising a control and reference optimization step based on the formulas:
;
,
with a command vector ,
and a setpoint vector ,
considering
a time , Or is a sampling period and is a positive integer;
a prediction horizon , Or is a number of samples composing this horizon;
a calculation time , with , Or is a minimum integer number of periods necessary for satisfactory optimization. Control method according to any one of Claims 5 to 6, characterized in that the buffering step comprises a storage step for storing, during a current optimization of the determination step by predictive control, the orders ( ) and references ( ) obtained at a previous optimization. Control method according to Claim 7, characterized in that the buffering step comprises a step of filtering by selector (S2) to successively apply the commands and references resulting from the storage step. A computer program comprising program code instructions for performing the steps of a control method according to any one of claims 5 to 8, when said program is running on a computer. Motor vehicle comprising a control system according to any one of Claims 1 to 4.