You are on page 1of 23

Multiobjective Optimization Applied to

Parameter-Varying Control Problems

Alesi Augusto, Richard Andrade

November 29, 2019


Table of contents

1. Introduction and Objectives

2. Problem Statement

3. Candidate Solution Treatment

4. Conclusions

November 29, 2019 1/16


Introduction and Objectives
Introduction

• Three degrees of
freedom.
• Two actuated wheels.
• Nonlinear Model.
• Linearization through
a desired trajectory
(Time-Varying Linear
Model)

November 29, 2019 Introduction and Objectives 2/16


Objectives

• To yield a reference trajectory;


• To estimate some Pareto-Optimal fronts (POFs) using MPC;
• To yield individual trajectories;
• To decide what control law should be used.

November 29, 2019 Introduction and Objectives 3/16


Problem Statement
Control Strategy

Model Predictive Control

∑{
N−1 }
min J = x̃a (K)′ Qˆ
ˆ ˆ ′ Rũ(K)
x̃a (K) + ũ(K) ˆ x̃a (N)′ Lˆ
+ˆ x̃a (N)

K=1

Where the compact form is given by



min J = b
x Wy b b ′ Wu u
x+u b.
b
u

subject to:

b
x = Pû + Gx̃a (K),
bmin ≤ u
u b≤u
bmax ,
b
xmin ≤ b
x≤b
xmax .

November 29, 2019 Problem Statement 4/16


Multiobjective Problem

For ωr = 0,
min J = α1 J1 + α2 J2
b
u

subject to:

b
xi = P(Ai )û + Gx̃a (K), ∀i = 1, 2,
bmin ≤ u
u b≤u
bmax ,
b
xmin ≤ b
xi ≤ b
xmax , ∀i = 1, 2.

November 29, 2019 Problem Statement 5/16


Pareto-Optimal Front Estimation

Figure 1: Illustrative example to yield each POF.

November 29, 2019 Problem Statement 6/16


Results

Figure 2: Estimated Pareto Figure 3: Normalized Pareto


Fronts. Front Estimates.

November 29, 2019 Problem Statement 7/16


Candidate Solution Treatment
Epsilon-Dominance Approach
Objective
Diversify the points in the front, and reduce the number of them.

Epsilon
Dominance
Algorithm

November 29, 2019 Candidate Solution Treatment 8/16


Utopian Distance Approach

November 29, 2019 Candidate Solution Treatment 9/16


Minimum Distance between the Utopian Point and the Estimated
Front.

Distance α1 α2 Distance α1 α2
0.3965 0.13 0.87 0.4661 0.58 0.42
0.3156 0.22 0.78 0.4952 0.86 0.14
0.4622 0.30 0.70 0.5585 0.97 0.03
0.4604 0.50 0.50 0.5134 0.71 0.29
0.5204 0.71 0.29 0.4652 0.49 0.51
0.5318 0.84 0.16 0.4474 0.28 0.72
0.5164 0.61 0.39 0.3158 0.22 0.78
0.4920 0.44 0.56 0.3965 0.13 0.87
0.4216 0.27 0.73 0.3156 0.22 0.78
0.3166 0.22 0.78 0.4622 0.30 0.70
0.3964 0.13 0.87 0.4604 0.50 0.50
0.3157 0.22 0.78 0.5204 0.71 0.29
0.4759 0.37 0.63 - - -

November 29, 2019 Candidate Solution Treatment 10/16


Performance Criteria

Mean square error (MSE) of the states:


Ns
1 ∑
MSE = x̃2a,i (j), i = 1, 2,
Ns
j=1

where “Ns ” is the number of samples.


Total control value (TCV) of the torques applied into the robot:
Ns

TCV = |ũi (j)| , i = 1, 2.
j=1

November 29, 2019 Candidate Solution Treatment 11/16


Performance Criteria

November 29, 2019 Candidate Solution Treatment 12/16


Decision-Making
AHP
Objective
To define priorities (W) based on pairwise comparisons.

Table 1: Importance Relation between Criteria.

TCVτ l TCVτ r MSEv MSEω


TCVτ l 1 5 3 3
TCVτ r 1/5 1 4 4
MSEv 1/3 1/4 1 3
MSEω 1/3 1/4 1/3 1

Table 2: Priority Vector.

TCVτl TCVτr MSEv MSEω


w 0.3679 0.3679 0.1686 0.0956
November 29, 2019 Decision-Making 13/16
Promethee

Objective
To define preferences based on directed graphs.


3
Pj (ai , ak ) = G[cj (ai ) − cj (ak )], wj Pj (ai , ak )


 0, if x ≤ 0 P(ai , ak ) =
j=1
,
G(x) = x/p, if 0 < x ≤ p ∑
3

 1, wj
if x > p, j=1
p ≜ max{cj (a)} − min{cj (a)}. ϕ(ai ) ≜ ϕ+ (ai ) − ϕ− (ai ).

November 29, 2019 Decision-Making 14/16


Results

Table 3: Best solutions.

Points α1 α2 TCVτl TCVτr MSEv MSEω


1/11/21 0.10 0.90 148.33 180.15 0.0136 0.0045

November 29, 2019 Decision-Making 15/16


Conclusions
Conclusions

• A methodology to control a nonlinear system was proposed


based in MPC:
1. Reference trajectory;
2. POF estimation;
3. Minimum distance criterion;
4. Individual trajectories;
5. Performance indexes;
6. Decision-making.
• The weighted-sum approach was used to treat the biobjective
problem;
• At the end, some equivalent solutions were provided.

November 29, 2019 Conclusions 16/16


Questions?

November 29, 2019 Conclusions 16/16

You might also like