Safety Controller Synthesis for Switched Systems using Multiscale Symbolic Models

Similar documents
Safety Controller Synthesis for Switched Systems using Multiscale Symbolic Models

A Symbolic Approach to Control via Approximate Bisimulations

Symbolic Control of Incrementally Stable Systems

Hybrid Control and Switched Systems. Lecture #2 How to describe a hybrid system? Formal models for hybrid system

Second Lecture: Basics of model-checking for finite and timed systems

Designing finite automata II

Model Reduction of Finite State Machines by Contraction

The size of subsequence automaton

Minimal DFA. minimal DFA for L starting from any other

Symbolic models for unstable nonlinear control systems

NFAs continued, Closure Properties of Regular Languages

Formal Languages and Automata

Assignment 1 Automata, Languages, and Computability. 1 Finite State Automata and Regular Languages

NFAs continued, Closure Properties of Regular Languages

CS415 Compilers. Lexical Analysis and. These slides are based on slides copyrighted by Keith Cooper, Ken Kennedy & Linda Torczon at Rice University

Chapter 2 Finite Automata

dx dt dy = G(t, x, y), dt where the functions are defined on I Ω, and are locally Lipschitz w.r.t. variable (x, y) Ω.

Random subgroups of a free group

Fundamentals of Computer Science

Convert the NFA into DFA

Lecture 09: Myhill-Nerode Theorem

Speech Recognition Lecture 2: Finite Automata and Finite-State Transducers. Mehryar Mohri Courant Institute and Google Research

CS 267: Automated Verification. Lecture 8: Automata Theoretic Model Checking. Instructor: Tevfik Bultan

Learning Moore Machines from Input-Output Traces

11.1 Finite Automata. CS125 Lecture 11 Fall Motivation: TMs without a tape: maybe we can at least fully understand such a simple model?

Controlling a population of identical NFA

CIRCULAR COLOURING THE PLANE

Refined interfaces for compositional verification

A tutorial on sequential functions

Formal languages, automata, and theory of computation

Reinforcement learning II

Exercises Chapter 1. Exercise 1.1. Let Σ be an alphabet. Prove wv = w + v for all strings w and v.

Java II Finite Automata I

More general families of infinite graphs

Inductive and statistical learning of formal grammars

CMPSCI 250: Introduction to Computation. Lecture #31: What DFA s Can and Can t Do David Mix Barrington 9 April 2014

First Midterm Examination

THE INTERVAL LATTICE BOLTZMANN METHOD FOR TRANSIENT HEAT TRANSFER IN A SILICON THIN FILM

Nondeterminism and Nodeterministic Automata

1. For each of the following theorems, give a two or three sentence sketch of how the proof goes or why it is not true.

Exercise 5.5: Large-scale log-normal fading

Topological, Automata-Theoretic and Logical Characterization of Finitary Languages

Regular expressions, Finite Automata, transition graphs are all the same!!

Regular Expressions (RE) Regular Expressions (RE) Regular Expressions (RE) Regular Expressions (RE) Kleene-*

Coalgebra, Lecture 15: Equations for Deterministic Automata

On Determinisation of History-Deterministic Automata.

Regular Languages and Applications

Non Deterministic Automata. Linz: Nondeterministic Finite Accepters, page 51

80 CHAPTER 2. DFA S, NFA S, REGULAR LANGUAGES. 2.6 Finite State Automata With Output: Transducers

Grammar. Languages. Content 5/10/16. Automata and Languages. Regular Languages. Regular Languages

Formal Language and Automata Theory (CS21004)

1 The Lagrange interpolation formula

CS 373, Spring Solutions to Mock midterm 1 (Based on first midterm in CS 273, Fall 2008.)

How to simulate Turing machines by invertible one-dimensional cellular automata

Non-Deterministic Finite Automata. Fall 2018 Costas Busch - RPI 1

Section: Other Models of Turing Machines. Definition: Two automata are equivalent if they accept the same language.

Chapter Five: Nondeterministic Finite Automata. Formal Language, chapter 5, slide 1

Introduction to Electronic Circuits. DC Circuit Analysis: Transient Response of RC Circuits

The Cayley-Hamilton Theorem For Finite Automata. Radu Grosu SUNY at Stony Brook

CS 301. Lecture 04 Regular Expressions. Stephen Checkoway. January 29, 2018

Fast Frequent Free Tree Mining in Graph Databases

5.1 How do we Measure Distance Traveled given Velocity? Student Notes

Speech Recognition Lecture 2: Finite Automata and Finite-State Transducers

A discrete chaotic multimodel based on 2D Hénon maps

Flexible Beam. Objectives

Design and Analysis of Distributed Interacting Systems

12.1 Nondeterminism Nondeterministic Finite Automata. a a b ε. CS125 Lecture 12 Fall 2016

Section 4.8. D v(t j 1 ) t. (4.8.1) j=1

Non-Deterministic Finite Automata

Centrum voor Wiskunde en Informatica REPORTRAPPORT. Supervisory control for nondeterministic systems

Worked out examples Finite Automata

First Midterm Examination

Space Curves. Recall the parametric equations of a curve in xy-plane and compare them with parametric equations of a curve in space.

Compiler Design. Fall Lexical Analysis. Sample Exercises and Solutions. Prof. Pedro C. Diniz

DFA minimisation using the Myhill-Nerode theorem

The University of Nottingham SCHOOL OF COMPUTER SCIENCE A LEVEL 2 MODULE, SPRING SEMESTER LANGUAGES AND COMPUTATION ANSWERS

On Stability of Nonlinear Slowly Time-Varying and Switched Systems

Review of Gaussian Quadrature method

Sufficient condition on noise correlations for scalable quantum computing

1 Nondeterministic Finite Automata

Chapter 4. Additional Variational Concepts

Homework Assignment 6 Solution Set

Module 6 Value Iteration. CS 886 Sequential Decision Making and Reinforcement Learning University of Waterloo

Complementing Büchi Automata with a Subset-tuple Construction

Overview HC9. Parsing: Top-Down & LL(1) Context-Free Grammars (1) Introduction. CFGs (3) Context-Free Grammars (2) Vertalerbouw HC 9: Ch.

CS S-12 Turing Machine Modifications 1. When we added a stack to NFA to get a PDA, we increased computational power

Approximate Time-Optimal Control via Approximate Alternating Simulations

Math 270A: Numerical Linear Algebra

Tutorial Automata and formal Languages

Software Engineering using Formal Methods

Physics 2135 Exam 1 February 14, 2017

Lecture 08: Feb. 08, 2019

Lecture 9: LTL and Büchi Automata

Application of Exact Discretization for Logistic Differential Equations to the Design of a Discrete-Time State-Observer

CHAPTER 1 Regular Languages. Contents. definitions, examples, designing, regular operations. Non-deterministic Finite Automata (NFA)

CSCI 340: Computational Models. Kleene s Theorem. Department of Computer Science

Foundations of XML Types: Tree Automata

Section 6.1 Definite Integral

Suppose we want to find the area under the parabola and above the x axis, between the lines x = 2 and x = -2.

Myhill-Nerode Theorem

Transcription:

Sfety Controller Synthesis for Switched Systems using Multiscle Symolic Models Antoine Girrd Lortoire des Signux et Systèmes Gif sur Yvette, Frnce Workshop on switching dynmics & verifiction Pris, Jnury 28-29, 2016 A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 1 / 33

Introduction Controller synthesis for clss of continuous-time switched systems Incrementlly stle systems: the influence of initil condition symptoticlly vnishes. Sfety specifiction: controlled invrince. Approch sed on the use of symolic models Discrete (time nd spce) pproximtion of the switched system. Approch sed on uniform discretiztion of time nd spce. [Girrd, Pol nd Tud, 2010] Distnce etween trjectories of incrementlly stle switched system nd of symolic model is uniformly ounded, nd cn e mde ritrrily smll. Sfety controller synthesis using symolic models vi lgorithmic discrete controller synthesis. A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 2 / 33

Motivtion Limittions of the symolic control pproch Sptil nd time resolution must e chosen crefully to chieve given precision: fst switching requires fine sptil resolution; Uniform sptil discretiztion: excessive computtion time nd memory consumption. Overcome this prolem with multiscle symolic models Use of multiscle discretiztions of time nd spce Incrementl explortion of symolic models during controller synthesis: The finer scles explored only if sfety cnnot e ensured t corser level. A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 3 / 33

Outline 1 Incrementlly stle switched systems 2 Multiscle symolic models 3 Sfety controller synthesis using multiscle symolic models 4 Computtionl experiments A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 4 / 33

Switched systems Definition A switched system is tuple Σ = (R n, P, P, F ), where R n is the stte spce; P = {1,..., m} is the finite set of modes; P is suset of S(R + 0, P), the set of functions from R+ 0 to P with finite numer of discontinuities on every ounded intervl of R + 0 ; F = {f 1,..., f m } is collection of smooth vector fields indexed y P. For switching signl p P, initil stte x R n, x(., x, p) is the trjectory of Σ, solution of: ẋ(t) = f p(t) (x(t)), x(0) = x. S τd (R + 0, P) is the set of switching signls p with minimum dwell-time τ d R + : discontinuities of p re seprted y t lest τ d. A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 5 / 33

Incrementl stility Definition Σ is incrementlly glolly uniformly symptoticlly stle (δ-guas) if there exists KL function β such tht for ll x 1, x 2 R n, p P, t R + 0 : x(t, x 1, p) x(t, x 2, p) β( x 1 x 2, t). x(t, x 2, p) x(t, x 1, p) t A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 6 / 33

Lypunov chrcteriztion Definition V p : R n R n R + 0, p P re multiple δ-guas Lypunov functions for Σ if there exist κ, µ R + with µ 1, K functions α, α, such tht for ll x 1, x 2 R n, p, p P: α( x 1 x 2 ) V p (x 1, x 2 ) α( x 1 x 2 ); V p x 1 (x 1, x 2 )f p (x 1 ) + Vp x 2 (x 1, x 2 )f p (x 2 ) κv p (x 1, x 2 ); V p (x 1, x 2 ) µv p (x 1, x 2 ). Theorem Let τ d R +, Σ = (R n, P, P, F ) with P S τd (R + 0, P) dmitting multiple δ-guas Lypunov functions. If τ d > log µ κ, then Σ is δ-guas. A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 7 / 33

Additionl ssumption In the following, we will ssume tht there exists K function γ such tht for ll x 1, x 2, x 3 R n V p (x 1, x 2 ) V p (x 1, x 3 ) γ( x 2 x 3 ), p P; This is not restrictive if V p re smooth nd we work on ounded suset of R n. A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 8 / 33

Outline 1 Incrementlly stle switched systems 2 Multiscle symolic models 3 Sfety controller synthesis using multiscle symolic models 4 Computtionl experiments A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 9 / 33

Trnsition systems Definition A trnsition system is tuple T = (X, U, Y,, X 0 ) where X, U, Y, X 0 re the sets of sttes, inputs, outputs nd initil sttes; X U X Y is trnsition reltion. T is metric if Y is equipped with metric d, symolic if X nd U re finite or countle sets. (x, u, x, y) is denoted (x, y) (x, u); u U is enled t x X, denoted u en(x), if (x, u) ; If en(x) =, then x is locking, otherwise it is non-locking; T is deterministic if for ll x X nd u en(x), (x, u) = 1. A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 10 / 33

Trjectories A trjectory of T is finite or infinite sequence of trnsitions σ = (x 0, u 0, y 0 )(x 1, u 1, y 1 )(x 2, u 2, y 2 )... where (x i+1, y i ) (x i, u i ), for ll i 0. It is: initilized if x 0 X 0 ; mximl if it is infinite or it is finite nd ends in locking stte. x X is rechle if there exists n initilized trjectory reching x. T is non-locking if ll initilized mximl trjectories re infinite or equivlently if ll rechle sttes re non-locking. A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 11 / 33

Approximte isimultion Definition Let T i = (X i, U, Y, i, X 0 i ), with i = 1, 2 e metric trnsition systems with the sme sets of inputs U nd outputs Y equipped with the metric d. Let ε R + 0, R X 1 X 2 is n ε-pproximte isimultion reltion etween T 1 nd T 2 if for ll (x 1, x 2 ) R, u U: (x 1, y 1) 1 (x 1, u), (x 2, y 2) 2 (x 2, u), d(y 1, y 2 ) ε nd (x 1, x 2 ) R; (x 2, y 2) 2 (x 2, u), (x 1, y 1) 1 (x 1, u), d(y 1, y 2 ) ε nd (x 1, x 2 ) R. T 1 nd T 2 re ε-pproximtely isimilr, denoted T 1 ε T 2, if X 0 1 R 1 (X 0 2 ) nd X 0 2 R(X 0 1 ). A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 12 / 33

Switched systems s trnsition systems Let Σ τd = (R n, P, P, F ) e switched system with P = S τd (R + 0, P). We consider controllers tht cn select: 1 mode p P; 2 durtion θ Θ N τ during which the mode remins ctive where Θ N τ = {θ s = 2 s τ s = 0,..., N}. where τ R +, N N re time smpling nd scle prmeters. We ssume τ d = θ Nd for some N d {0,..., N}, then Θ N d τ = {θ s Θ N τ θ s τ d }. Let C(I, R n ) denote the set of continuous functions from I to R n. A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 13 / 33

Switched systems s trnsition systems Let T N τ (Σ τd ) = (X, U, Y,, X 0 ) where: X = R n P, z = (x, p) X consists of continuous stte x nd n ctive mode p. U = P Θ N τ, u = (p, θ s ) U consists of mode p nd durtion θ s. Y = s=n s=0 C([0, θ s], R n ) is set of continuous functions, equipped with the metric: { d(y, y y y ) = if θ s = θ s + if θ s θ s X 0 = R n P. A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 14 / 33

Switched systems s trnsition systems For z = (x, p) X, z = (x, p ) X, u = ( p, θ s ) U, y Y, (z, u, z, y) ( p, θ s ) {p} Θ N τ (P \ {p}) Θ N d τ x = x(θ s, x, p) nd p = p. y = x θs (., x, p) y = x θs (., x, p) x = x(θ s, x, p) x T N τ (Σ τd ) is deterministic nd metric. A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 15 / 33

Computtion of the symolic model We pproximte R n y sequence of emedded multiscle lttices { [R n ] 2 s η = q R n 2 s+1 } η q[i] = k i, k i Z, i = 1,..., n n where η R + is stte spce smpling prmeter. We ssocite multiscle quntizer Q s η : R n [R n ] 2 s η such tht Q s η(x) = q q[i] 2 s η n x[i] < q[i] + 2 s η n, i = 1,..., n. Let X s η = [R n ] 2 s η P, then X 0 η X 1 η X N η. We define the symolic model s T N τ,η(σ τd ) = (X N η, U, Y, η, X 0 η ). A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 16 / 33

Computtion of the symolic model For r = (q, p) X, r = (q, p ) X, u = ( p, θ s ) U, y Y, ( p, θ s ) {p} Θ N (r, u, r τ (P \ {p}) Θ N d τ, y) q = Q s η(x(θ s, q, p)) nd p = p. y = x θs (., q, p) q = Q 1 η(x(θ 1, q, p)) q y = x θ1 (., q, p)) A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 17 / 33

Computtion of the symolic model For r = (q, p) X, r = (q, p ) X, u = ( p, θ s ) U, y Y, ( p, θ s ) {p} Θ N (r, u, r τ (P \ {p}) Θ N d τ, y) q = Q s η(x(θ s, q, p)) nd p = p. y = x θs (., q, p) q = Q 0 η(x(θ 0, q, p)) q y = x θ0 (., q, p)) A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 17 / 33

Approximtion result T N τ (Σ τd ) is symolic, deterministic nd metric. Theorem Let Σ τd dmit multiple δ-guas Lypunov functions V p, p P. Consider prmeters τ, η R +, N N, nd precision ε R +. If τ d > log µ κ nd { s=nd (( ) )] 1µ η min min [2 s γ 1 s=0 e κθs α(ε), [ ( s=n 1 e min 2 s γ 1 κθ s ( )} 1µ α(ε) )], α 1 s=0 µ α(ε) then T N τ (Σ τd ) ε T N τ,η(σ τd ). A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 18 / 33

Outline 1 Incrementlly stle switched systems 2 Multiscle symolic models 3 Sfety controller synthesis using multiscle symolic models 4 Computtionl experiments A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 19 / 33

Sfety specifiction Let T = (X, U, Y,, X 0 ) e symolic, deterministic trnsition system where Y C([0, θ y ], R n ). θ y R + Let S R n e suset of sfe sttes. We define the trnsition system T S = (X, U, Y, S, X 0 ) where for x, x X, u U, y Y, u en(x); (x, y) S (x, u) (x, y) = (x, u); t [0, θ y ], y(t) S. T S is symolic nd deterministic. Remrk: sfety is defined on continuous-time outputs. A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 20 / 33

Sfety controller Definition A sfety controller for T S = (X, U, Y, S, X 0 ) is reltion C X U such tht for ll x X : C(x) en(x); if C(x), then u C(x), C(x ) with S (x, u) = (x, y). We denote the domin of C s dom(c) = {x X C(x) }. The controlled trnsition system is T S/C = (X, U, Y, S/C, X 0 C ) where X 0 C = X 0 dom(c) nd for x, x X, u U, y Y, (x, y) S/C (x, u) { u C(x); (x, y) = S (x, u). T S/C is symolic, deterministic nd non-locking. A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 21 / 33

Mximl sfety controller Lemm There exists unique mximl sfety controller C X U such tht for ll sfety controllers C, C C. Definition A stte x X is sfety controllle if nd only if x dom(c ). The set of sfety controllle sttes is denoted cont(t S ). Computtion of C requires complete explortion of T S. A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 22 / 33

Lzy sfety synthesis Lzy sfety synthesis: trde-off etween mximlity nd efficiency. Give priority to inputs with longer durtion, which led to sttes on corser grids. Compute the symolic model on the fly. Finer scles re explored (computed) only if sfety cnnot e ensured t the corser scles. Let us define priority reltion on inputs: totl preorder U U The ssocited equivlence nd strict wek order reltions re u u u u nd u u; u u u u nd u u. For multiscle symolic models where U = P Θ N τ : (p, θ s ) (p, θ s) θ s θ s; (p, θ s ) (p, θ s) θ s = θ s; (p, θ s ) (p, θ s) θ s < θ s. A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 23 / 33

Mximl lzy sfety controller Definition A mximl lzy sfety (MLS) controller for T S = (X, U, Y, S, X 0 ) is sfety controller C X U such tht: ll sfety controllle initil sttes re in dom(c): X 0 cont(t S ) dom(c); ll sttes x dom(c) re rechle in T S/C ; for ll sttes x dom(c): 1 if u C(x), then u en(x) with u u, (x, y) = S (x, u ), u C(z) x cont(t S ); 2 if u C(x), then u en(x) with u u, (x, y) = S (x, u ), x / cont(t S ). A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 24 / 33

Mximl lzy sfety controller Theorem There exists unique MLS controller for T S. MLS controller synthesis: X 0 = {x 1, x 2, x 3 },. x 1 x 2 x 3 x 1 x 2 x 3 x 4 A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 25 / 33

Mximl lzy sfety controller Theorem There exists unique MLS controller for T S. MLS controller synthesis: X 0 = {x 1, x 2, x 3 },. x 1 x 2 x 3 x 1 x 2 x 3 x 4 A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 25 / 33

Mximl lzy sfety controller Theorem There exists unique MLS controller for T S. MLS controller synthesis: X 0 = {x 1, x 2, x 3 },. x 1 x 2 x 3 x 1 x 2 x 3 x 4 A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 25 / 33

Mximl lzy sfety controller Theorem There exists unique MLS controller for T S. MLS controller synthesis: X 0 = {x 1, x 2, x 3 },. x 1 x 2 x 3 x 1 x 2 x 3 x 4 A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 25 / 33

Mximl lzy sfety controller Theorem There exists unique MLS controller for T S. MLS controller synthesis: X 0 = {x 1, x 2, x 3 },. x 1 x 2 x 3 x 1 x 2 x 3 x 4 A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 25 / 33

Mximl lzy sfety controller Theorem There exists unique MLS controller for T S. MLS controller synthesis: X 0 = {x 1, x 2, x 3 },. x 1 x 2 x 3 x 1 x 2 x 3 x 4 A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 25 / 33

Mximl lzy sfety controller Theorem There exists unique MLS controller for T S. MLS controller synthesis: X 0 = {x 1, x 2, x 3 },. x 1 x 2 x 3 x 1 x 2 x 3 x 4 A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 25 / 33

Mximl lzy sfety controller Theorem There exists unique MLS controller for T S. MLS controller synthesis: X 0 = {x 1, x 2, x 3 },. x 1 x 2 x 3 x 1 x 2 x 3 x 4 A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 25 / 33

Mximl lzy sfety controller Theorem There exists unique MLS controller for T S. MLS controller synthesis: X 0 = {x 1, x 2, x 3 },. x 1 x 2 x 3 x 1 x 2 x 3 x 4 A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 25 / 33

Mximl lzy sfety controller Theorem There exists unique MLS controller for T S. MLS controller synthesis: X 0 = {x 1, x 2, x 3 },. x 1 x 2 x 3 x 1 x 2 x 3 x 4 A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 25 / 33

Outline 1 Incrementlly stle switched systems 2 Multiscle symolic models 3 Sfety controller synthesis using multiscle symolic models 4 Computtionl experiments A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 26 / 33

Switched system with dwell-time We consider the switched system: ẋ(t) = A p(t) x(t) + p(t), p(t) {1, 2}, with A 1 = [ ] 0.25 1 2 0.25, A2 = [ ] 0.25 2 1 0.25, 1 = [ ] 0.25 2, 2 = [ 0.25 1 ]. The switched system dmits multiple δ-guas Lypunov functions nd is incrementlly stle for switching signls with minimum dwell-time τ d = 2. Multiscle strction with prmeters τ = 4, η = 8 100, N = 3 2 Uniform strction with prmeters τ = 1 2, η = 1 100 2 = precision ε = 0.4. Sfe set: S = [ 6, 6] [ 4, 4] \ [ 1.5, 1.5] [ 1, 1]. A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 27 / 33

Switched system with dwell-time Controller synthesis: Uniform symolic model Multiscle symolic model Time 160s 7.3s Size (10 3 ) 5228 33 Durtions 0.5 (100%) 4 (26%) 2 (54%) 1 (11% 0.5 (9%) A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 28 / 33

Switched system with dwell-time MLS Controller: Mode 1 is ctive Mode 2 is ctive A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 29 / 33

Switched system with dwell-time Controlled switched system: 4 3 2 1 0-1 -2-3 -4-6 -5-4 -3-2 -1 0 1 2 3 4 5 6 A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 30 / 33

Circulr n-room uilding We consider the system: Ṫ i (t) = α(t i+1 (t) + T i 1 (t) 2T i (t)) +β(t e T i (t)) + γ(t h T i (t))u i (t) where: T i (t) is the temperture of room i, 1 i n, T 0 (t) = T n (t) nd T n+1 (t) = T 1 (t). u i (t) = 1 if room i is heted, u i (t) = 0 otherwise nd n i=1 u i(t) 1. n-dimensionl switched system with n + 1 modes dmits common Lypunov function nd is incrementlly stle. Multiscle strction with prmeters τ = 80, η = 0.28, N = 4 = Precision ε = 0.4. Sfe set: S = [19, 21.5] n. A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 31 / 33

Circulr n-room uilding Controller synthesis: Multiscle symolic models n = 3 n = 4 n = 5 Time 0.2s 6s 312s Size (10 3 ) 2 45 1 077 Durtions 40 (1%) 20 (25%) 20 (6%) 20 (37%) 10 (73%) 10 (92%) 10 (62%) 5 (2%) 5 (2%) Computtionl complexity increses with dimension: Stte nd input spce re lrger. The control prolem is lso intrinsiclly more complex in higher dimension ecuse of the constrint: n u i (t) 1. i=1 A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 32 / 33

Conclusions Multiscle pproximtely isimilr symolic models for incrementlly stle switched systems: Bsed on multiscle smpling of time nd spce; Allow significnt complexity reduction for controller synthesis. Multiscle sfety controller synthesis: Bsed on the notion of mximl lzy sfety controller; Prtil explortion of the symolic strctions; Cn e extended to more generl sfety properties, e.g. specified y hyrid utomton. Future work: MLS controller synthesis lgorithm for non-deterministic systems; Consider other types of specifictions, e.g. rechility: mximl lzy rechility controller my not e unique. Girrd, Gössler nd Mouelhi, Sfety controller synthesis for incrementlly stle switched systems using multiscle symolic models. IEEE TAC, 2016. A. Girrd (L2S-CNRS) Synthesis using multiscle symolic models 33 / 33