Method for controlling a switched electronic circuit

Abstract

To this end, the invention relates to a method for controlling a switched electronic circuit connecting an electrical voltage source u to a load R and forming a system having an output y and a plurality of operating modes i, at least some of which thus can be activated by following a switching rule o. According to the invention, the following steps are provided: —measuring state variables x of the system having equilibrium values xr; —introducing at least one parameter p into the system, representing a measuring error of an electrical unknown of the operation of the circuit; —estimating the parameter from a state monitor and applying a Lyapunov function thereto; —deducing therefrom the equilibrium values of the system in order to obtain the control rule.

Description

TECHNICAL FIELD OF THE INVENTION

The present invention relates to the field of electronics and more particularly to controlling switched electronic circuits.

TECHNOLOGICAL BACKGROUND

The category of switched electronic circuits includes power converters. These devices find very widespread use in domestic and industrial applications for adjusting the power delivered by a power source to a load. A model flyback DC-to-DC converter is illustrated in FIG. 1. This converter comprises a primary circuit and a secondary circuit. The primary circuit comprises a voltage source that is connected in series with the primary winding of a transformer and a switch S. An inductor L is added to model the transformer. The secondary circuit comprises a secondary winding of the transformer that is connected by a diode D to a resistive load R. A capacitor C is connected in parallel with the secondary winding and the diode. A resistor R_esr is added in series with the capacitor C to model losses due to the capacitor C. The transistor may be controlled, unlike the diode. The input voltage source V_e is an ideal voltage source, i.e. without any series resistance and with unlimited power. The inductor of the primary L is also ideal, i.e. without any series resistance or magnetic losses. The diode and the transistor are ideal (the voltage in conduction mode is zero). The transformer is also assumed to be ideal.

The converter may operate in three modes:

• • Mode 1: the inductor L collects energy from the voltage source V. while the capacitor C supplies the load R with power.
  • Mode 2: the energy collected in the inductor L flows into the transformer, charges the capacitor C and supplies the load R with power.
  • Mode 3: the capacitor C supplies the load R with power.

If only Modes 1 and 2 are active during the cycle, the converter operates in CCM (continuous-conduction mode). If Mode 3 is active, the system works in DCM (discontinuous-conduction mode), i.e. the energy stored in the inductor in Mode 1 is completely transferred to the capacitor in Mode 2.

It is known practice to control these converters on the basis of a linear time-invariant (LTI) system model. These methods result in local stability. They are generally implemented with two loops, a fast current loop and a voltage loop.

The miniaturization of electronic components has made it possible to develop less expensive systems. Additionally, it is advantageous to work at low current levels so as to decrease losses. In this context, the switching frequency of the converter plays a fundamental role: the higher the frequency to which the switching frequency may be increased, the greater the extent to which the electronic parts may be miniaturized and losses decreased.

However this miniaturization presents robustness problems: the short response time as a result of the smaller components make the system more sensitive to variations.

It has been envisaged to take the explicit hybrid nature of these converters into account in the design of stabilization control laws which are based in particular on a switched affine system model.

One example of a switched affine system model applicable to such a flyback DC-to-DC converter and the corresponding operating conditions are described below. The notation will be as follows: the set composed of N first integers is denoted by K={1, . . . , N} and if the simplex of dimension N−1 is Λ_N:={λε^N|∀iε, λ_i>0, Σ_i=1^Nλ_i=1}. The convex combination of a set of matrices A=(A₁, . . . , A_N) is denoted by A(λ)=Σ_i=1^Nλ₁A_i, where λεΛ_N. The identity matrix is denoted by I for all dimensions.

Such a converter is a switched system that can be modeled in the form of a switched state space as follows:{dot over (x)}=A_ox+B_ou  (1)where x(t)εⁿ is the state and u(t)ε^m is the input that is assumed to be fixed (u(t)=V_o), wherein said input cannot be used as a control variable. The control command is formed by the switching law σ: ⁿ→K which indicates the active mode at any given time. The state vector x^T=[i_L, v_C] is composed of i_L (current through the inductor L) and of v_C (voltage across the terminals of the capacitor C). With reference to FIG. 1 and by using Kirchhoff's laws, the sets A={A1, . . . , A_N} and β={B1, . . . , B^N}, where N=3, are given by:

$\begin{array}{cc}\begin{array}{c}{A}_1=\left[\begin{array}{cc}0& 0\\ 0& \frac{-1}{\mathrm{RCK}}\end{array}\right],A_2=\left[\begin{array}{cc}\frac{n^2{R}_{\mathrm{esr}}}{\mathrm{LK}}& \frac{{\mathrm{nR}}_{\mathrm{esr}}}{\mathrm{RLK}}-\frac{n}{L}\\ \frac{\mathrm{CK}}{}& \frac{-}{\mathrm{RCK}}\end{array}\right]\\ A_3=\left[\begin{array}{cc}0& 0\\ 0& \frac{-1}{\mathrm{RCK}}\end{array}\right]\end{array}& \left(2\right)\\ B_1=\left[\begin{array}{c}\frac{1}{L}\\ 0\end{array}\right]B_2=\left[\begin{array}{c}0\\ 0\end{array}\right]B_3=\left[\begin{array}{c}0\\ 0\end{array}\right]& \left(3\right)\end{array}$ where

$K=\left(1+\frac{\mathrm{Resr}}{R}\right).$ Hereinafter, what is considered is the CCM configuration (only Modes 1 and 2 may be activated, hence N=2). The control problem, which is a switched system control problem, may also be considered by using, for the system defined in (1), a “convexified” representation given by:{dot over (x)}=A(λ)x+B(λ)u  (4)where λεΛ_N. Specifically, applying the density theorem guarantees that all of the trajectories of the convexified system defined in (4) may be reached by the switched system defined by equation (1) by means of sufficiently fast switching. This means that it is possible to study the convexified system and apply the results obtained to the switched system. Since the matrix A(λ) is a Hurwitz matrix, it is possible to design stabilization strategies even though not all of the A_i are stable. In this case, the equilibrium point of the system defined by equation (4), denoted by x_e, is found in the set X_e such that, for fixed u, it is defined as:

$\begin{array}{cc}{X}_e=\left\{x\in {ℝ}^N|x=-A^{-1}\left(\lambda \right)B\left(\lambda \right)u,\lambda \in \Lambda ,A\left(\lambda \right)\mathrm{Hurwitz}\right\}& \left(5\right)\end{array}$

The stabilization strategies of this control law are based on an assumed perfect knowledge of all of the operating parameters of the converter, including the input voltage and the load.

By assuming that for every λεΛ, such that A(λ) is Hurwitz, there is at least one index i, such that λ_i≠0 and A_i is Hurwitz (Hypothesis 1), and by stating that the estimate {tilde over (x)} of the state x has a value of {tilde over (x)}=x−x^r where x^r is the equilibrium point of this state, the Lyapunov function V is defined by:V(x):={tilde over (x)}^rP{tilde over (x)}  (6)where P=P^T>O, P being the Lyapunov matrix, the solution to Lyapunov equation (7) isA_i^TP+PA_i+α_iP≥0i=1, . . . ,N  (7)where

$\alpha \left(\lambda \right)=\sum _{i=1}^N{\lambda }_i{\alpha }_i,\{\begin{array}{cc}{\alpha }_i>0& \mathrm{if}A\left(\lambda \right)\mathrm{is}\mathrm{Hurwitz}\\ {\alpha }_i=0& \mathrm{else}\end{array}$ According to a first proposition (Proposition 1), what is considered is the switched system (1) and an equilibrium point x^r∈X_e where u=V_e, λ=λ^r and A(λ^r) Hurwitz: for P=P^T>0 solution to inequality (7), the equilibrium point x^r is asymptotically stable overall with the following feedback law:

$\begin{array}{cc}{\sigma }^{*}\left(x\right)=\arg \left[\underset{i\in K}{\min }\left[{\tilde{x}}^{T}P\left(A_{i}x+B_{i}u\right)\right]\right]& \left(8\right)\end{array}$

This is proven using a Lyapunov function

V(x,xr)=(x,xr)={tilde over (x)}^TP{tilde over (x)}. The derivative of V is denoted, along the trajectories d=A(λ)x+B(λ)u, by:{dot over (v)}(x,x^r;λ)=2{tilde over (x)}^TP(A(λ)x+B(λ)u)

Defining the Lyapunov matrix P gives:{dot over (V)}(x,x^r;λ)=2{dot over (x)}^TP(a(λ){tilde over (x)}+A(λ)x^r+B(λ)u)≥−α(λ)x^TP{dot over (x)}+B(λ)(u)For λ=λ^r, the term A(λ)x^r+B(λ)u=0 and so:

$\begin{array}{cc}\begin{array}{c}\stackrel{.}{V}\left(x,x^r,{\lambda }^r\right)=\sum _{i=1}^N{\lambda }_i^{r}2{\tilde{x}}^{T}P\left(A_{i}x+B_{i}u\right)\\ \le -\alpha \left({\lambda }^r\right){\tilde{x}}^{T}P\tilde{x}\end{array}& \left(9\right)\end{array}$ Since αλ^r=Σ_i=1^Nλ₁^rα_i cannot be zero due to Hypothesis 1, it can be concluded that the choice of switching rule (8) leads to:V(x,x^r;σ*)=2{tilde over (x)}^{{dot over (T)}}P(A_σ*x+B_σ*u)≥−σ(λ^r){tilde over (x)}^TP{tilde over (x)}>0,if{tilde over (x)}≠0  (10)

In practice, the control law (8) must be sampled and it has been shown that this sampling results in a static error, which is dependent on the sampling time T. With infinite switching frequency it is possible to decrease this static error to zero, but this is not realistic in a real-time application. Moreover, the method uses a Lyapunov function V which is dependent on the equilibrium point x^r. However, for power converters such as mentioned above, the equilibrium point x^r is dependent on the load R and on the input voltage u, and if there are variations or uncertainties in these parameters, the robustness of stabilization of the equilibrium point must be analyzed.

OBJECT OF THE INVENTION

An object of the invention is to provide a stable control law for switched systems that takes unknown constant parameters into account. For example, in the case of power converters, the invention provides a stable control law even in the absence of knowledge of the input voltage and/or the output load.

DISCLOSURE OF THE INVENTION

To this end, the invention relates to a method for controlling a switched electronic circuit connecting a voltage source u to a load R and forming a system having an output y and a plurality of operating modes i, at least some of which can therefore be activated by following a switching law a. According to the invention, the following steps are provided:

• • measuring state variables x of the system having equilibrium values x^r;
  • introducing at least one parameter p into the system, which parameter is representative of a measurement error for an electrical operating unknown of the circuit;
  • estimating the parameter on the basis of a state observer;
  • deducing therefrom the equilibrium values of the system that are needed for control-law convergence.

More specifically, the method preferably comprises the steps of:

• • measuring state variables x of the system having equilibrium values x^r;
  • associating at least one error parameter p with at least one electrical operating unknown of the system so as to represent the system in the form

$\hspace{1em}\{\begin{array}{c}\stackrel{.}{x}=A_{\sigma }x+B_{\sigma }u+G_{\sigma }p\\ \stackrel{.}{p}=0\end{array}$

• • A_σ being the state matrix, B_σ the input matrix and G_σ the parameter matrix in mode i controlled by the switching law σ;
  • estimating the values of the parameter p on the basis of a state observer of type

$\hspace{1em}\{\begin{array}{c}\stackrel{.}{\hat{z}}={\stackrel{\_}{A}}_{\sigma }\hat{z}+{\stackrel{\_}{B}}_{\sigma }u+L_{\sigma }\left(y-\hat{y}\right)\\ \hat{y}=\stackrel{\_}{C}\hat{z}\end{array}$

• •  where z is a state vector such that z^T=[x^T,p^T], {tilde over (C)} is an observation matrix such that {tilde over (C)}=[1 0], and L_σ is a gain matrix calculated by solving a linear matrix inequality (LMI) such that the error e with time between the state vector z and its estimate 2 tends exponentially toward zero;
  • deducing therefrom estimated equilibrium values {circumflex over (x)}^r of the state variables x;
  • the switching law taking the following form

${\sigma }^{*}\left(\hat{x},{\hat{x}}^r,\hat{p}\right)=\arg \left[\underset{i}{\min }\left[w^{T}P\left(A_i\hat{x}+B_{i}u+G_i\hat{p}\right)\right]\right]$

• • where w=({circumflex over (x)}−{circumflex over (x)}^r) and P is the solution to the Lyapunov equation.

Thus, the proposed control law depends on estimating at least one parameter added to the model of the system to represent losses and variations in quantities that cannot be measured or in any case known exactly.

Other features and advantages of the invention will become apparent upon reading the following description of one particular nonlimiting implementation of the invention.

BRIEF DESCRIPTION OF THE DRAWINGS

Reference will be made to the appended drawings, in which:

FIG. 1 is a simplified diagram of a flyback converter circuit;

FIG. 2 is a schematic representation of the parameter p₁;

FIG. 3 is a schematic representation of the parameter p₂;

FIG. 4 shows the incorporation of the parameters p₁ and p₂ into this circuit;

FIG. 5 shows curves illustrating a stepwise variation in R and V_e with the switching-and-observing control method of the invention.

DETAILED DISCLOSURE OF THE INVENTION

The invention is described here in terms of application to a flyback DC-to-DC converter. However, it goes without saying that the invention is applicable to any switched system.

As mentioned above, in power converters it is common for there to be no input voltage measurement and no output load measurement available. The invention envisages taking into account two parameters that are representative of the missing input voltage measurement and of the missing output load measurement. These parameters are here related to the resistance R and to the voltage u, such as illustrated in FIGS. 2 and 3. The voltage parameter p₂ makes it possible to identify the error in the input voltage, as explained by equations (12) (in which x₁ represents the uncontrolled state variables), and the current parameter p, makes it possible to identify the error in the output load, as explained by equations (11).

$\begin{array}{cc}i=\frac{v}{R}=p_1+\frac{r}{R}\hat{R}=\frac{1}{\frac{1}{R^{*}}+\frac{p_1}{v}}& \left(11\right)\end{array}$

$\begin{array}{cc}{\stackrel{.}{x}}_1=\frac{v_L}{L}=\frac{u^{*}-p_2}{L}\stackrel{.}{u}=u^{*}-p_2& \left(12\right)\end{array}$

For ease of understanding, these two parameters, given by equations (11) and (12), are added virtually to the converter, as shown in FIG. 4.

A perfect estimate of these parameters gives key information on the steady-state point which is unknown, a priori, in the absence of R and u measurements.

Similarly, the switched affine system describing the converter of FIG. 4 is given by:

$\begin{array}{cc}\{\begin{array}{c}\stackrel{.}{x}=A_{\sigma }x+B_{\sigma }u+G_{\sigma }p\\ \stackrel{.}{p}=0\end{array}& \left(13\right)\end{array}$ where p is an unknown-parameter vector, A is the state matrix, B is the input matrix and G is the parameter matrix.

To estimate this vector, use is first made of an augmented state vector z^T=[x^T, p^T] to rewrite the system defined by the equations (13) as:

$\begin{array}{cc}\{\begin{array}{c}\stackrel{.}{z}={\tilde{A}}_{\sigma }z+{\tilde{B}}_{\sigma }u\\ y=\tilde{C}z\end{array}& \left(14\right)\end{array}$ where à is the augmented state matrix and {tilde over (B)} is the augmented input matrix, which are defined as

${\tilde{A}}_1=\left[\begin{array}{cc}{A}_t& G_t\\ 0& 0\end{array}\right]{\tilde{B}}_1=\left[\begin{array}{c}{B}_t\\ 0\end{array}\right]$ $G_1=\left[\begin{array}{cc}0& \frac{-1}{L}\\ \frac{-1}{\mathrm{CK}}& 0\end{array}\right]G_2=\left[\begin{array}{cc}\frac{{\mathrm{nR}}_{\mathrm{esr}}}{\mathrm{LK}}& \frac{-1}{L}\\ \frac{-1}{\mathrm{CK}}& 0\end{array}\right],G_3=\left[\begin{array}{cc}0& 0\\ \frac{-1}{\mathrm{CK}}& 0\end{array}\right]$

Without losing generality, {tilde over (C)}=[C 0] is used for the output matrix. In the present case, C=I but the approach resulting in the invention is also applicable when the state x is not available for measurement. The set of equilibrium points of the system defined by formula (14) is then:Z_e={z∈R^N|{hacek over (A)}(λ)z +{hacek over (B)}(λ)u=0}  (15)It is assumed that z^r∈Z_e and λ=λ^r, such thatÃ(λ^r)z^r+{tilde over (B)}(λ^r)V_e=0  (16)

An augmented state observer is used to retrieve the unknown parameters. The integer-order observer is defined by:

$\begin{array}{cc}\{\begin{array}{c}\stackrel{.}{\hat{z}}={\tilde{A}}_{\sigma }\hat{z}+{\tilde{B}}_{\sigma }u+L_{\sigma }\left(y-\stackrel{.}{y}\right)\\ \stackrel{.}{y}=\tilde{C}\stackrel{.}{z}\end{array}& \left(17\right)\end{array}$

By inserting the error e between the vector z and its estimate {circumflex over (z)} such that e(t)=z(t)−{circumflex over (z)}(t), its dynamic behavior is given by:ċ−({dot over (A)}_a−L_aC)e  (19)

The gains L_i must be designed to satisfy

lim_t→∞[e(t)]=0, and to ensure that this is achievable using the following conventional result.

According to a second proposition (Proposition 2): if there is a positive definite matrix P_obs=P_obs>0, matrices R_i and scalars γ_i>0 satisfying, for every i∈K:Â_t^TP_obs+P_obsÃ_t−C^TR_t^T−R_tC+γ_tP_obs<0  (20)then the switched systems meeting equation (19) are, overall, exponentially stable, with a decrease that is at least equal to γ=_1ϵK^minγ_i.

This is proven by analyzing the Lyapunov function V(e)=e^TP_obs^e and the well-known variable change R_i=P_obsL_i.

If equation (20) allows a solution with P_obs=P_obs^T>0, this means that V=e^TP_obs^e is a Lyapunov function and that lim_t→∞[e(t)]=0 for any switching rule.

This makes it possible to construct a stabilization switching control law that is based on a parameter and state estimate by modifying switching law (8) to guarantee overall asymptotic stabilization.

The main difficulty is associated with determining the operating point based on an estimate {circumflex over (p)} of p. To explain how to get there, the following definitions and hypotheses are required.

Definition 1: for fixed ξ≥0, the set Λ_ref(ξ)={λεΛ: A(λ+v) is Hurwitz for every vεβ(0,ξ)}.

Definition 2: for fixed ξ≥0, the set of accessible equilibrium points is defined as:

$X_{\mathrm{ref}}\left(\xi \right)=\left\{x^r\in {ℝ}^N:x^r=-{A\left(\lambda \right)}^{-1}\left[B\left(\lambda \right)G\left(\lambda \right)\right]\left[\begin{array}{c}u\\ p\end{array}\right],\left(p,\lambda \right)\in {ℝ}^{n_p}\times A_{\mathrm{ref}}\left(\xi \right)\right\}$

x₂ is used as the controlled portion of the state variables and x₁ the remaining state variables, such that x^T=(x₁^T,X₂^T). Without losing generality, it is assumed that x₁=C_x1^x and x₂=C_x2^x. The function φ in C^∞ is considered as:φ(λ,x₁,x₂,u,p)=A(λ)x+B(λ)u+G(λ)p  (21)For λ∈Λ_ref(ξ), meaning that A(Λ) is Hurwitz and, consequently, invertible, there is, for every fixed u and p, a single x, such thatφ(Λ,x₁,x₂,u,p)=0Hypothesis 2: for every x₂^r∈X_ref(ξ)|x₂,u,p, there is a single class function C¹0=(0₁,0₂) such thatx₁^T=ϕ₁(x₂^r,u,p)∈X_ref(ξ)|_x1  (22)Λ^r=ϕ₂(x₂^r,u,p)∈Λ_ref(ξ)  (23)which satisfies the following relationship:

$\begin{array}{c}\psi \left(\lambda ,x_1,x_2,u,p\right)=\psi \left({\lambda }^r,x_1^r,x_2^r,u,p\right)\\ =0\end{array}$ Moreover, it is assumed that there is a constant η(x^r,u,ξ)>0 such that

$\begin{array}{cc}{\frac{\partial {\varphi }_1\left(x_2^r,u,·\right)}{\partial p}}_{\infty }\le \eta \left(x_2^r,u,\xi \right)& \left(24\right)\end{array}$

The parameterization effected by equations (22) and (23) plays an important role in the proposed approach, since it makes it possible to estimate the reference point based on an estimate {circumflex over (p)} of the parameters p and a reference (or equilibrium value) x₂^r for the controlled portion x₂ of the state variables. This leads to{circumflex over (x)}_l^r=φ₁(x₂^r,u,{circumflex over (p)})  (25){circumflex over (Λ)}^r=φ₂(x₂^r,u,{circumflex over (p)})  (26)From this, it is assumed that:

${\hat{x}}^r=\left[\begin{array}{c}{\hat{x}}_1^r\\ x_2^r\end{array}\right].$ The dynamic behavior of {circumflex over (x)}₁^r, for fixed x₂^r and u, is:

$\begin{array}{cc}{\stackrel{.}{\hat{x}}}_1^r=\frac{\partial {\varphi }_1}{\partial p}\stackrel{.}{\hat{p}}& \mathrm{ }\left(27\right)\\ =\frac{\partial {\varphi }_1}{\partial p}\left(\stackrel{.}{\hat{p}}-\stackrel{.}{p}\right)& \left(28\right)\end{array}$ {dot over (x)} ₂ ^r=0  (29)

By assuming that p=0 and by replacing p with its expression (18), what is obtained is:

$\begin{array}{cc}{\stackrel{.}{\hat{x}}}_1^r=\frac{\partial {\varphi }_1}{\partial p}\underset{\_}{L}\left({\lambda }^r\right)\tilde{C}e& \left(30\right)\end{array}$ {dot over (x)} ₂ ^r=0  (31)

This leads to the following theorem.

Theorem 1: by considering the switched affine system with unknown parameters such as defined in (13), by writing x^rT=(x₁^rT,x₂^rT) with x₁^r=θ₁(x₂^r,u,p) and by assuming that Proposition 1 and Proposition 2 are satisfied, what is obtained, according to Hypothesis 1 and Hypothesis 2 and for every x₂^r∈X_ref(λ)|_x2 where ξ>0, is the feedback law defined by:

$\begin{array}{cc}{\sigma }^{*}\left(\stackrel{.}{x},{\stackrel{.}{x}}^r,\stackrel{.}{p}\right)=\arg \left[\underset{i}{\min }\left[w^{T}P\left(A_i\hat{x}+B_{i}u+G_i\hat{p}\right)\right]\right]& \left(32\right)\end{array}$ where w=({circumflex over (x)}−{circumflex over (x)}^r).

This feedback law guarantees overall asymptotic stability of the equilibrium point x^r as can be proven below.

It is taken from Proposition 2 that the observation error converges toward zero for any switching rule and, from Proposition 1, the Lyapunov matrix P and the scalars α_i. For {circumflex over (λ)}^e∈Λ_ref(ξ), equation (7) gives:2ω^TPA({circumflex over (Λ)}^r)ω+α({circumflex over (Λ)})ω^TPω≤0  (33)where α({circumflex over (Λ)}^r)=Σ_i=1^N{circumflex over (Λ)}_i^rα.

With A({circumflex over (Λ)}^r){circumflex over (x)}^r+B({circumflex over (Λ)}^r)u+G({circumflex over (Λ)}^r){circumflex over (p)}=0 and {circumflex over (Λ)}_r defined in equations (25) and (26), giving:2ω^TP(A({circumflex over (x)}^r){circumflex over (x)}+B({circumflex over (Λ)}^r)u+G({circumflex over (Λ)}^r){circumflex over (p)})≤−α({circumflex over (Λ)}^r)ω^TPωΣ_i=1^N{circumflex over (Λ)}_i²ω^TP(A_i{dot over (x)}+B_iu+G_i{circumflex over (p)})≤−α({circumflex over (Λ)}^r)ω^TPω

Since α({circumflex over (Λ)}^r)=Σ_i=1^N={circumflex over (Λ)}_i^rα_i cannot be zero according to Hypothesis 1, it is possible to conclude that control law (32) leads to:2ω^TP(A_σ+{circumflex over (x)}B_σ·u+G_{σ·{circumflex over (P)})≤−α({circumflex over (λ)}}^r)ω^T Pω  (34)<0, if ω≠0  (35)

By taking the Lyapunov function V defined in equation (6) and by using ({circumflex over (x)},{circumflex over (x)}^r,Λ) instead of x,x^r, what is obtained is:

$\stackrel{.}{V}\left(\hat{x},{\hat{x}}^r,\lambda \right)=2{\omega }^{T}P\left(A\left(\lambda \right)\hat{x}+B\left(\lambda \right)u+G\left(\lambda \right)\hat{p}+\left(\stackrel{\_}{L}\left(\lambda \right)-\frac{\partial {\varphi }_1}{\partial p}\underset{\_}{L}\left(\lambda \right)\right)\tilde{C}e\right)$

If the following term is added to each side of inequality (34)

$2{\omega }^{T}P\left({\stackrel{\_}{L}}_{{\sigma }^{*}}-\frac{\partial {\varphi }_1}{\partial p}{\underset{\_}{L}}_{{\sigma }^{*}}\right)\tilde{C}e$ what is obtained is:

$\begin{array}{cc}\stackrel{.}{V}\left(\hat{x},{\hat{x}}^r;{\sigma }^{*}\right)\le -\alpha \left({\hat{\lambda }}^r\right){\omega }^{T}P\omega +2{\omega }^{T}P\left({\stackrel{\_}{L}}_{{\sigma }^{*}}-\frac{\partial {\varphi }_1}{\partial p}{\underset{\_}{L}}_{{\sigma }^{*}}\right)\tilde{C}e& \left(36\right)\end{array}$

Since e converges exponentially toward 0 for any switching rule and therefore, in particular, for the switching rules resulting from control law (32), where γ=min_iγ_i, there is K>0,∥e(t)∥≤Ke^−γt∥e₀∥Hence:

$\begin{array}{cc}\stackrel{.}{V}\left(\hat{x},{\hat{x}}^r,{\sigma }^{*}\right)\le -\alpha \left({\hat{\lambda }}^r\right){\omega }^{T}P\omega +2{\beta }_{\max }\omega \times \left({\stackrel{\_}{L}}_{{\sigma }^{*}}+{\frac{\partial {\varphi }_1\left(x_2^r,u,·\right)}{\partial p}}_{\infty }{\underset{\_}{L}}_{{\sigma }^{*}}\right)\times \tilde{C}{\mathrm{Ke}}^{-\pi }{ϵ}_0& \left(37\right)\end{array}$ where β_max is the maximum eigenvalue of P.A condition sufficient to guarantee

$\stackrel{.}{V}\left(\hat{x},{\hat{x}}^r;{\sigma }^{*}\right)\le -\frac{\alpha \left({\hat{\lambda }}^r\right)}{2}{\omega }^{T}P\omega <0$ is

$\omega >\underset{i}{\max }4\frac{{\beta }_{\max }}{{\alpha }_{\min }{\beta }_{\min }}\left({\stackrel{\_}{L}}_i+{\frac{\partial {\varphi }_1\left(x_2^r,u,·\right)}{\partial p}}_{\infty }{\underset{\_}{L}}_i\right)\times \tilde{C}K{e}^{-\pi }{\varepsilon }_0$ where β_min is the minimum eigenvalue of P and

${\alpha }_{\min }=\underset{\lambda \in {\Lambda }_{\mathrm{ref}}\left(\xi \right)}{\min }\alpha \left(\lambda \right)$ It should be noted that α_min>0 for ξ>0 according to Hypothesis 1 and by defining Λ_ref(ξ).

Since there is η(x₂^r,u,ξ)>0, such that

${\frac{\partial {\varphi }_1\left(x_2^r,u,·\right)}{\partial p}}_{\infty }\le \eta \left(x_2^r,u,\xi \right),$ it can be concluded that there is a constant M(x₂^r,u,ξ)>0, such that if∥ω∥>(x₂^r,u,ξ)e^−γt∥e₀∥then,

$\stackrel{.}{V}\left(\hat{x},{\hat{x}}^r;{\sigma }^{*}\right)\le -\frac{\alpha \left({\hat{\lambda }}^r\right)}{2}{\omega }^{T}P\omega <0$ Since it is possible to state that:∀e₀,∀e,∃t₂, such that Me^−γt∥e_O∥<ε,∀t<t₂,there is necessarily

$\stackrel{.}{V}\left(\hat{x},{\hat{x}}^r;{\sigma }^{*}\right)\le -\frac{{\alpha }_{\min }}{2}{w}^{T}Pw<0$ for any w, such that ∥w∥>∈. This means that for any w(0), there is t₃, such that:∥w(t)∥≤∈for every t>t₃.It can therefore be concluded that w converges asymptotically toward 0. Since e also converges toward zero, there is:

$\hat{x}\to x\to {\hat{x}}^r=\left[\begin{array}{c}{\hat{x}}_1^r\\ x_2^r\end{array}\right]$ where x₁^R=O₁(x₂^r,r,p), since {circumflex over (p)}→p.

The conditions in Hypothesis 2 may be relaxed as follows.

A third hypothesis (Hypothesis 3) is then formulated: it is assumed that the application θ=(θ₁,θ₂) in Hypothesis 2 is no longer valid for every p but only for any p in a closed convex domain denoted by P(x₂,u,ξ) dependent on u,x_u and ξ. It is also assumed that the nominal value of parameter p is included in P(x₂^r,u,ξ) and Λ=x2(x₂^r,u,p)∈Λ_ref (ξ).

According to a third proposition (Proposition 3): according to Hypothesis 3, Theorem 1 remains true if the operating point ({circumflex over (x)}₁^r,{circumflex over (Λ)}^r) is such that{circumflex over (x)}₁^r=ϕ₁(x₂^r,u,Π{circumflex over (p)})  (38){circumflex over (x)}^r=ϕ₂(x₂^r,u,Π{circumflex over (p)})  (39)where Π{circumflex over (p)} is the orthogonal projection of {circumflex over (p)} on P(X₂^r,u,ξ). This can be proven in the following way. Since P(x₂^r,u,ξ) is closed and convex, the orthogonal projection Π{circumflex over (p)} of {circumflex over (p)} on P(x₂^r,u,ξ) is continuous. The function θ is considered and the projection Π is used, such that {circumflex over (p)}→θ_i(x₂^r,u,Π{circumflex over (p)}), i=1,2 for fixed x₂^r and u. Since {circumflex over (Λ)}=θ₂ (x₂^r u,Π{circumflex over (p)})∈Λ_ref(ξ), it is seen that

$\begin{array}{cc}\begin{array}{c}{\hat{x}}_1^r={\varphi }_1\left(x_2^r,u,\prod \hat{p}\right)\\ =-C_{x1}{A\left({\hat{\lambda }}^r\right)}^{-1}\left[B\left({\hat{\lambda }}^r\right)G\left({\hat{\lambda }}^r\right)\right]\left[\begin{array}{c}u\\ \prod \hat{p}\end{array}\right]\end{array}& \left(40\right)\end{array}$ which is well defined as A({circumflex over (Λ)}^r), is invertible.

The projection 1 makes it possible to calculate a reference point ({circumflex over (x)}₁^r, x₂^r) for any t. Of course, the operating point {circumflex over (x)}^r,({circumflex over (x)}₁^r,x₂^r) calculated in this way with a projection of p is not an equilibrium point of the system when {circumflex over (p)}∈P(x₂^r,u,ξ).

Since {circumflex over (p)}∈P(x₂^r,u,ξ), the convergence of the error e toward zero for any switching rule and the continuity of θ guarantee that there is a time t₁(e_o), such that {circumflex over (Λ)}^r=θ₂(x₂^r,u,{circumflex over (p)})∈Λ_ref(ξ) and {circumflex over (p)}∈P(x₂^r,u,ξ) for every t>t₁, as long as P is sufficiently close to p. For every t>t₁, proof is reached by following steps similar to those taken to prove Theorem 1.

Thus, to drive a flyback converter such as described, it is sufficient to combine the converter with a unit for controlling the switch which is programmed to implement the control law of Theorem 1.

A numerical example will now be described.

A flyback converter described by equation (1) is chosen, this depending on six parameters having the following nominal values:u=28V,R=75Ω,L=200μH,C=2.6μF,R_esr=0.02Ω and n=2.

The target equilibrium point is written as x₂^r=15V. It is assumed that the converter is working with said nominal values and, having robustness as the primary objective, it is also assumed that x₁ and x₂ are measured. It is also assumed that the load R and the input voltage V_e may change and that these parameters are not measured.

As hypothesis, what is considered is a piecewise constant function with variations of between 15Ω and 150Ω for the load and between 44 V and 28 V for the input voltage.

It is verified that this system satisfies Hypothesis 3. For a given value x₂^r, and by calculating {circumflex over (p)} on the basis of (17), what is obtained is Λ^r:

$\begin{array}{cc}\{\begin{array}{c}{\lambda }_1^r=\frac{\left(R_{\mathrm{esr}}+R_o\right)\left(p_2+{\mathrm{nx}}_2^r\right)}{\left(R_{\mathrm{esr}}+R_o\right)V_e+{\mathrm{nR}}_o{x}_2^r-{\mathrm{nR}}_{\mathrm{esr}}{R}_o{p}_1}\\ {\lambda }_2^r=1-{\lambda }_1^r\end{array}& \left(41\right)\end{array}$

For a given value ξ>0, the set Λ_ref(ξ) is given by

$\hspace{1em}\{\begin{array}{c}{\lambda }_1\in [\begin{array}{cccc}0& 1& -& \xi \end{array}[\\ {\lambda }_2=1-{\lambda }_1\end{array}$ since only A₁ is not Hurwitz. It is then straightforward to verify, on the basis of equations (41), that if P(x₂^r,u,ξ) is defined by:−p_imax≤p_l<p_lmax<p_llim −nx₂^r≤p2≤α(ξ)p1+b(ξ)where

$a\left(\xi \right)=-\left(1-\xi \right)\frac{{\mathrm{nR}}_{\mathrm{esr}}R}{R_{\mathrm{esr}}+R},b\left(\xi \right)=\left(1-\xi \right)V_e-{\mathrm{nR}}_{\mathrm{esr}}{x}_2^r$ and

$p_{1_{\lim }}=\frac{\left(R_e\mathrm{sr}+R\right)V_e+{\mathrm{nRx}}_2^r}{{\mathrm{nR}}_{\mathrm{esr}}R},$ then Λ₁^r∈Λ_ref(ξ) and hence x₁^r is well defined for the set P. Moreover, since the set P is compact, inequality (24) is satisfied, as is, consequently, Hypothesis 3 for the application under consideration. In practice, p_1max=10³, which value is very far from the admissible physical values for p.

LMI conditions (7) are solved to obtain the matrix P with α₁=0, α₂=35 and ξ=0,1:

$\begin{array}{cc}P=\left(\begin{array}{cc}74291& 36\\ 36& 993\end{array}\right)& \left(43\right)\end{array}$

LMI conditions (20) are also solved to obtain the gains L_i for the observer with γ₁=γ₂=3,3×10⁴. In this application, a gain of the current observer is obtained for each mode:

$\begin{array}{cc}{L}_i=\left(\begin{array}{cc}362& 19\\ 19& 1569\\ 243& -3153\\ -3153& -243\end{array}\right),i\in 𝕂& \left(44\right)\end{array}$

By applying the robust and adaptive control of Theorem 1, the results shown in FIG. 5 are obtained. To see the behavior of the system in a closed loop, in the case of all of the parameters being perfectly known, the simulation scenario begins with the nominal values x₂. The variations in R and U are introduced after 1 ms. FIG. 5 shows a stepwise variation in R and V_e with the switching-and-observing control method. Curves 1 and 2 show x₁,{circumflex over (x)}₁^r and x₂,x₂^r, respectively, while curve 3 shows the estimation parameters p₁ and p₂ and curve 4 shows the variations in U and R. Times are given in ms.

It is apparent in these curves that parameter p converges toward a new value each time R or V_e changes, and that the controlled output, x₂, exhibits robustness faced with variations in R and V_e. The control law clearly rejects the disruption in x2 and the current x1 is well adjusted to the variations.

It is worth noting here that the control method according to the invention is not based on a linear approximation of the system. The control method is implemented directly on the model of the switched system by taking non-linearities into account, which makes it possible to guarantee the overall asymptotic stability of the corrector.

Moreover, stability is demonstrated using the observer in the control loop, which is unlike the general practice of the prior art.

Generally, moreover, the control method is developed under the assumption that the observer has converged. However, it is proven here that the observer does not need to have already converged to guarantee overall asymptotic stability.

The control method according to the invention is thus particularly advantageous.

As seen above, the control method makes it possible to obtain a stable control law even in the absence of exact knowledge of all of the operating parameters of the switched electronic circuit.

Here, neither the input voltage nor the output load is measured: the observers used allow these parameters to be taken into account.

Control that is self-adaptive with respect to variations in operating parameters of the switched electronic circuit is thus obtained without it being necessary to measure said operating parameters. The switched electronic circuit is simplified and its cost decreased by removing the difficulties related to obtaining accurate measurements of the operating parameters.

It should be noted that the control law is stable over a wider operating range than the stability range obtained with the control laws of the prior art (outside linear domain).

Moreover, this stability, obtained by implementing the control method according to the invention, can be proven. Being able to prove stability is a great asset in applications requiring certification work, and in particular in aerospace applications. Without such a proof of stability, even if the circuit appears to operate perfectly well, it is effectively impossible to guarantee stability over the entire operating range, and hence it is impossible to certify and to authorize a device implementing such a circuit for flight.

Of course, the invention is not limited to the embodiments described but includes all variants that come within the scope of the invention as defined by the claims.

Claims

1. A method for controlling a switched electronic circuit connecting a voltage source u to a load R and forming a system having an output y and a plurality of operating modes i, at least some of which can therefore be activated by following a switching law σ, characterized in that the method comprises the steps of: measuring state variables x of the system having equilibrium values xr;associating at least one error parameter p with at least one electrical operating unknown of the system so as to represent the system in the form

2. The method as claimed in claim 1, wherein the switched electronic circuit is a converter comprising a primary circuit and a secondary circuit, the primary circuit comprising the voltage source that is connected in series with a primary winding of a transformer and with a switch, an inductor L being connected in parallel with the primary winding and being under a voltage vL, the secondary circuit comprising a secondary winding of the transformer that is connected by a diode to a resistive load R, the method comprising the step of defining a current parameter p1 to represent the error in the output load such that