lower and upper bounds respectively. Generally our objectivein linear MPC is minimization of cost function of the form
J
=
x
t
Qx
+
u
t
Ru
(10)Majority of industrial MPC applications use linear empiricalmodels, therefore most of MPC products and optimizationalgorithms are based on this model type.
B. Optimization Methods
Main challenge in MPC is to ﬁnd fastest ways of optimization as time required for solving onoptimization is very limited. Thus we need realtime optimal solution. Sometimes weﬁnd a trade off by looking for a suboptimal solution which isless complex. Constraints are linear. Exact solution methodsare well studied for linear MPC. The optimal solution relieson a linear dynamic model of the process, respects all inputand output constraints, and minimizes a performance ﬁgure.This is usually expressed as a quadratic or a linear criterion,so that the resulting optimization problem can be cast as aquadratic program (QP) or linear program (LP), respectively,for which a rich variety of efﬁcient activeset and interiorpoint solvers are available[10].
1) Linear Programming (LP):
A few authors have investigated MPC optimization based on linear programming [6],[7], [8]. If we have objective function of the form
minJ
=
min

Px
N

∞
+
N
−
1
k
=0

Qx
k

∞
+

Ru
k

∞
(11)subject to constraints
Gz
≤
W
+
Sx
(
t
)
then MPC law can be deﬁned by the solution of a LinearProgram[9]. Schechter[9] proved that this is true for anysum of convex piecewise afﬁne costs.
2) Algebraic Method:
If we consider the following objective function:
minJ
=
x
t
+
N
y

t
Px
t
+
N
y

t
+
N
y
−
1
k
=0
x
t
+
k

t
Qx
t
+
k

t
+
u
t
+
k
Ru
t
+
k
(12)subject to constraints
y
min
≤
y
t
≤
y
max
, k
= 1
,
2
,...,N
c
u
min
≤
u
t
≤
u
max
, k
= 0
,
1
,...,N
c
and system dynamics
x
t
+
k
+1
=
Ax
t
+
k

t
+
Bu
t
+
k
, k
≥
0
,y
t
+
k

t
+
Cx
t
+
k

t
, k
≥
0
,u
t
+
k
=
Kx
t
+
k

t
, N
u
≤
k
≤
N
y
,
where matrices
Q
=
Q
≥
0
,
R
=
R
≥
0
and
P
≥
0
.
N
u
,N
y
,N
c
are input horizon, output horizon and constrainthorizon respectively such that
N
y
≥
N
u
and
N
c
≤
N
y
−
1
. We solve this problem (12) repeatedly at eact time
t
for current measurement
x
t
and predicted state variable,
x
t
+1

t
,...,x
t
+
k

t
at time steps
t
+ 1
,...,t
+
k
and corresponding optimal control actions
U
∗
=
{
u
∗
t
,...,u
∗
t
+
k
−
1
}
isobtained. The ﬁrst predicted input is applied to the system asﬁrst control action i.e.,
u
t
=
u
∗
t
. This procedure is repeatedat time
t
+ 1
based on new state
x
t
+1
.The the tuning cost function matrix
P
and state feedback gain
K
is generally used to guarantee closed loop stability of ststem (12). Algebraic solution of this system depends uponﬁnding the values of
P
and
Q
matrices.
P
is found by thesolution of discrete Lyapunov equation
P
=
A
PA
+
Q.
Assuming the problem is unconstrained, inﬁnite horizoni.e.,
N
c
=
N
u
=
N
y
=
∞
we can ﬁnd the state feedback gain
K
by solving the Algebraic Ricatti equation:
K
=
−
(
R
+
B
PB
)
−
1
B
PA,P
= (
A
+
BK
)
P
(
A
+
BK
) +
K
RK
+
Q
Solution of Lyapunov and Algebraic Ricatti equations isthe most popular method to ﬁnd the values of
K
and
P
matrices[13], [14] thus solving the problem algebraically.
3) Quadratic Programming (QP):
Rawlins and Morari[10], [11], [12] proved that linear MPC can be posed asQuadratic Programming (QP) problem. If we incorporate thefollowing realtion
x
t
+
k

t
=
A
k
x
t
+
k
−
1
j
=0
A
j
Bu
t
+
k
−
1
kj
into system represented by set of equations (12) then it givesus the following quadratic programming (QP) optimizationproblem [18]
J
∗
(
x
t
) =
min
12
U
HU
+
x
t
FU
+12
x
t
Yx
(
t
)
(13)subject to constraints
Gz
≤
W
+
Sx
(
t
)
where
U
[
u
t
,...,u
t
+
Nu
−
1
]
∈
R
s
,
and
s
mN
u
,
isa vector of optimization variables,
H
=
H
0
,
and
H,F,Y,G,W,E
matrices are obtained from state constraintmatrix
S
and input matrix
R
. MPC is applied by solvingQP problem (13) repeatedly at each time
t
≥
0
for
x
t
, thecurrent state value .Despite the fact that efﬁcient QP solversare available to solve , computing the input
u
t
online mayrequire signiﬁcant computational effort [15].
4) Multi Parametric Quadratic Programming (mpQP):
In MPC our goal is to reduce online optimization timebecause system operates in realtime. These days, substantialresearch is being carried out to ﬁend more efﬁcient optimization algorithms. Bemporad et. al. [13], [16], [17] solvedthe problem (12) by multiparametric quadratic programming