Explicit model predictive control of gas-liquid separation ... · evaluation. The approximate...
Transcript of Explicit model predictive control of gas-liquid separation ... · evaluation. The approximate...
Revie
w C
opy
1
Explicit model predictive control of gas-liquid
separation plant via orthogonal search tree
partitioning
Alexandra Grancharova*, 1, 2, Tor A. Johansen*, 3, Juš Kocijan†, ‡, 4
* Department of Engineering Cybernetics, Norwegian University of Science and Technology
7491 Trondheim, Norway
† Department of Systems and Control, Jozef Stefan Institute, Jamova 39
1000 Ljubljana, Slovenia
‡ Nova Gorica Polytechnic, Vipavska 13, 5000 Nova Gorica, Slovenia
1 Present address: Institute of Control and System Research, Bulgarian Academy of Sciences, Acad. G. Bonchev str., Bl.2, P.O.Box 79, Sofia 1113, Bulgaria 2 Corresponding author: [email protected] 3 [email protected] 4 [email protected]
1 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
2
Abstract
Solutions to constrained linear model predictive control (MPC) problems can be pre-computed
off-line in an explicit form as a piecewise linear (PWL) state feedback defined on a polyhedral partition
of the state space. This admits implementation at high sampling frequencies in real-time systems with
high reliability and low software complexity. Recently, algorithms that determine an approximate
explicit PWL state feedback solution by imposing an orthogonal search tree structure on the partition,
have been developed, and it has been shown that they may offer computational advantages. This paper
considers the application of an approximate approach to the design of an explicit model predictive
controller for a two-input two-output laboratory gas-liquid separation plant, including experimental
evaluation. The approximate explicit MPC controller achieves performance close to that of the
conventional MPC, but requires only a fraction of the real-time computational machinery, thus leading to
fast and reliable computations.
Keywords: Constrained linear model predictive control; Multi-parametric quadratic programming; Gas-
liquid separator
2 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
3
1. Introduction.
Model predictive control (MPC) has become the accepted methodology to solve complex control
problems related to process industries (Garcia et al., 1987; Biegler & Rawlings, 1991; Mayne et al.,
2000). It allows the design of multi-input multi-output (MIMO) control systems that minimize a
quadratic performance index in the presence of input and output constraints imposed on the system.
Recently, several methods for explicit solution of MPC problems have been developed. The main
motivation behind explicit MPC is that an explicit state feedback law avoids the need for real-time
optimization, and is therefore potentially useful for applications where MPC has not traditionally been
used. In addition to embedded system applications, this technology is also suitable for certain small-scale
process control applications. In particular, low-level control of fairly simple unit processes is typically
implemented using conventional PI/PID control rather than MPC. Some reasons are the need for high
reliability, fast processing and low software complexity, as these are typically executed in a highly
reliable computer environment. On the other hand, explicit MPC is ideally suited for such problems as it
offers the benefits of constrained MIMO control with low complexity and high reliability for such small-
scale problems.
In (Bemporad et al., 2000; Pistikopoulos et al., 2000) it was recognized that the constrained
linear MPC problem can be posed as a multi-parametric quadratic program (mp-QP), when the state is
viewed as a parameter to the problem. It was shown that the solution (the control input) has an explicit
representation as a piecewise linear (PWL) state feedback on a polyhedral partition of the state space, see
also (Bemporad et al., 2002; Johansen et al., 2002; Seron et al., 2000; Tøndel et al., 2003), and they
develop an mp-QP algorithm to compute this function.
Recently, in (Johansen & Grancharova, 2002; Johansen & Grancharova, 2003; Grancharova &
Johansen, 2002), algorithms that determine an approximate explicit PWL state feedback solution by
imposing an orthogonal search tree structure on the partition, have been developed. It has been shown
that they may lead to more efficient real-time computations. The present paper considers the application
of these algorithms to the design of an explicit model predictive controller for a laboratory gas-liquid
separation plant, including experimental evaluation. A preliminary version of this paper is given in
(Grancharova et al., 2003).
3 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
4
2. Approximate approach to explicit MPC.
2.1. Explicit MPC and exact mp-QP.
Formulating a linear MPC problem as an mp-QP is briefly described below, see (Bemporad et al., 2000;
Pistikopoulos et al., 2000) for further details. Consider the linear system:
)()()()()1(
ttttt
CxyBuAxx
=+=+
(1)
where nt ℜ∈)(x , mt ℜ∈)(u , and pt ℜ∈)(y are the state, input and output variable. Also, nn×ℜ∈A ,
mn×ℜ∈B , np×ℜ∈C and (A, B) is a controllable pair. It is assumed that a full measurement of the state
)(tx is available at the current time t. Then, for the current )(tx , MPC solves the optimization problem:
{ }))(,(min))((
1,...,
* tJtVNtt
xUxuuU −+≡
= (2)
subject to:
0,
0,)(
1,...,1,0,
,...,1,
||
||1
|
maxmin
max|min
≥=
≥+=
=−=≤≤
=≤≤
++
++++
+
+
kk
tNk
Nk
tkttkt
kttkttkt
tt
kt
tkt
CxyBuAxx
xxuuuyyy
(3)
with the cost function given by:
[ ] tNttNtN
kktkttkttkttJ ||
T1
0
T||
T))(,( ++−
=++++ ++= ∑ PxxRuuQxxxU (4)
and symmetric 0>R , 0≥Q , 0>P . The column vector sNtt ℜ∈≡ −+
TT1
T ),...,( uuU , mNs = , is the
optimization vector and the final cost matrix P may be taken as the solution of the discrete-time algebraic
Riccati equation (Ogata, 1995):
QPBARPBBPBAPAAP ++−= − TT1TTT )()( (5)
With the assumption that no constraints are active for Nk ≥ this corresponds to an infinite horizon LQ
criterion, and the MPC is stabilizing (Chmielewski & Manousiouthakis, 1996).
It is shown in (Bemporad et al., 2000) that by substituting:
∑−
=−−++ +=
1
01| )(
k
jjkt
jktkt t BuAxAx (6)
in the optimization problem defined by (2), (3) and (4), this can be rewritten in the form:
4 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
5
)(tosubject
)(21min)()(
21))(( TTT*
t
ttttV
ExWGU
FUxHUUYxxxU
+≤
++=
(7)
Here, 0T >= HH , and H, F, Y, G, W, E are easily obtained from Q, R, and (2) – (6) (since only the
optimizer U is needed, the term involving Y is usually removed from (7)). Further, by defining
)(T1 txFHUz −+≡ , sℜ∈z , the optimization problem (7) is transformed into the following equivalent
problem (Bemporad et al., 2000):
SxWGz
Hzzxz
+≤
=
tosubject21min)( T*
zV(8)
where ( )xFFHYxxx T1T**
21)()( −−−= VVz and T1FGHES −+≡ . The vector x is the current state,
which can be treated as a vector of parameters. The number of inequalities is denoted q and the number
of free variables is mNs = . Then ss×ℜ∈H , sq×ℜ∈G , 1×ℜ∈ qW , nq×ℜ∈S . It has been shown in
(Bemporad et al., 2000; Pistikopoulos et al., 2000) that the optimization problem (8) is a multi-
parametric quadratic program (mp-QP) and its solution can be found in an explicit form )(xzz ** = as a
PWL function of x defined over a polyhedral partition of the parameter space. Algorithms for iteratively
constructing a polyhedral partition of the state space and computing the PWL solution are given in
(Bemporad et al., 2000; Pistikopoulos et al., 2000; Bemporad et al., 2002; Johansen et al., 2002; Seron et
al., 2000; Tøndel et al., 2003).
2.2. Approximate mp-QP algorithm.
In (Johansen & Grancharova, 2002; Johansen & Grancharova, 2003), an algorithm that determines an
approximate explicit PWL state feedback solution of possible lower complexity is developed. The idea is
to require that the state space partition is represented as a binary search tree (quad-tree (Bentley, 1975),
cf. Fig. 1), i.e. to consist of orthogonal hypercubes organized in a hierarchical data-structure. This allows
extremely fast real-time search. When searching the tree, only n scalar comparisons are required at each
level.
In this paper, an improved version of the approximate mp-QP algorithm (Johansen &
Grancharova, 2002; Johansen & Grancharova, 2003) is used which is based on a k - d tree partition of the
5 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
6
state space (Fig. 2) as a more flexible and powerful alternative to the generalized quad-tree (Fig. 1). With
the k - d tree (Bentley, 1975), a hyper-rectangle is split into two equal parts and thus only one scalar
comparison is required at each level when searching the tree. Also, the k - d tree allows the incorporation
of heuristic rules that split the hyper-rectangle at the axis along which the change of error is maximal
(before splitting).
The approximate mp-QP algorithm is based on the cost function approximation error defined as:
)()(ˆ)( * xxx zz VV −=ε (9)
where )(ˆ)(ˆ21)(ˆ T xzHxzx 00=zV and )()(
21)( T* xHzxzx **=zV are respectively the sub-optimal and the
optimal costs, and )(ˆ xz0 and )(xz* are the sub-optimal and the optimal PWL solutions. The main idea
of the algorithm is to compute the solution of the problem (8) at the n2 vertices of a considered hyper-
rectangle 0X by solving up to n2 QPs. Based on these solutions, a feasible local linear approximation
)(ˆ xz0 to the PWL optimal solution )(xz* , valid in the whole hyper-rectangle 0X , is computed by using
the following result (Bemporad & Filippi, 2001):
Lemma 1. Consider the bounded polyhedron fXX ⊆0 with vertices { }M21 vvv ,...,, (here fX is the
feasible set: })3(satisfying)({ Ux ∃ℜ∈= nf tX ). If 0K and 0g solve the QP:
( ) ( )∑=
−−−−M
i 1
T
,)()(min 0i0i
*0i0i
*
gKgvKvzHgvKvz
00
(10)
subject to:
( ) { }Mi ,...,2,1, ∈+≤+ WSvgvKG i0i0 (11)
then the least squares approximation 000 gxKxz +=)(ˆ is feasible for the mp-QP (8) for all 0X∈x .
If the maximal approximation error 0ε in the hyper-rectangle 0X is smaller than some
prescribed tolerance 0>ε , no further refinement of 0X is needed. Otherwise, 0X is partitioned into two
hyper-rectangles and the procedure described above is repeated for each of these. The upper bound 0ε of
the approximation error is determined by using the method proposed in (Johansen & Grancharova, 2003).
In order to reduce the complexity of the partition, the heuristic rule described in (Grancharova &
Johansen, 2002) is applied when splitting the hyper-rectangle 0X . The rule attempts to split the hyper-
6 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
7
rectangle at the axis along which the change of the approximation error is maximal (before splitting),
because it is reasonable to hope this is how the largest reduction of the error can be made. It has been
shown in (Grancharova & Johansen, 2002) that the use of such heuristics reduces the complexity of the
partition significantly.
The complexity is further reduced by implementing control input trajectory parameterization
with less degrees of freedom in order to reduce the dimension of the optimization problem (Tøndel &
Johansen, 2002). The most common approach is to pre-determine the time-instants at which the control
input iu is allowed to change (input blocking).
The following approximate mp-QP algorithm is taken from (Johansen & Grancharova, 2003):
Algorithm 1 (approximate mp-QP)
Step 1. Initialize the partition to the whole hyper-rectangle, i.e. { }XP = . Mark the hyper-rectangle X as
unexplored.
Step 2. Select any unexplored hyper-rectangle PX ∈0 . If no such hyper-rectangle exists, go to step 8.
Step 3. Compute the solution to the QP (8) for x fixed to each of the n2 vertices of the hyper-rectangle
0X . If all QPs have a feasible solution, go to step 5. Otherwise, go to step 4.
Step 4. Compute the size of 0X using some metric. If it is smaller than some given tolerance, mark 0X
infeasible and explored. Go to step 2. Otherwise, go to step 7.
Step 5. Compute an affine state feedback 0z using Lemma 1, as an approximation to be used in 0X . If
no feasible solution was found, go to step 7.
Step 6. Compute the error bound 0ε in 0X . If εε ≤0 , mark 0X as explored and feasible and go to step
2.
Step 7. Split the hyper-rectangle 0X into two hyper-rectangles 1X and 2X by applying the heuristic
splitting rule from (Grancharova & Johansen, 2002). Mark them unexplored, remove 0X from P, add
1X and 2X to P, and go to step 2.
Step 8. If necessary, split the hyper-rectangles containing the origin such that 0)( =xz* is optimal
everywhere in these hyper-rectangles. Terminate.
7 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
8
This algorithm will terminate with a PWL function that is an approximation to the PWL exact
solution and is defined on an inner approximation fX of the set fXX ∩ . The set fX is represented as
a union of hyper-rectangles.
3. Model of the gas-liquid separation plant.
We consider a sub-process within a semi-industrial installation which is used for reduction of NOx in
effluent gases and technological waste water treatment by means of neutralisation with CO2 contained in
flue gases (Vrančić et al., 1995). The role of the separation unit is to capture flue gases under low
pressure from effluent channels by means of water flow and to carry them over under high enough
pressure to the downstream (neutralisation) stage. The separation unit is shown in Fig. 3 (Vrančić et al.,
1995). The flue gases coming from the effluent channels are “pooled” by the water flow into the water
circulation pipe through the injector I1. The water flow is generated by the pump P1 (water ring). The
speed of the pump is kept constant. The pump feeds the mixture of water and gas into the separator R1
where gas is separated from water. Hence the accumulated gas in R1 forms a sort of “gas cushion” with
increased internal pressure. Owing to this pressure, flue gas is blown out from R1 into the next
neutralisation unit. On the other side the “cushion” forces water to circulate back to the reservoir R2. The
quantity of water in the circuit is constant. If for some reason additional water is needed, the water supply
path through the valve V5 is utilised.
The complete process model can be described by the following expressions (Vrančić et al., 1995)
(the meaning of the variables is given in the Appendix):
Valve 1V : )]bar/(s[1.461.75 111011
11 lRKK vvV
−− ⋅=⋅= (12)
<>
=otherwise
0011
1
1
1
1
rrifrif
v (13)
Valve 2V : ])bar/(s[66.75742.0 112022
22 lRKK vvV
−− ⋅=⋅= (14)
<>
=otherwise
00
2
2
max22max2
2
rrif
rrifrv (15)
8625.0max2 =r (16)
8 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
9
The speed limit of valves positions:
<>
=otherwise
minmin
maxmax
rvrifvvrifv
v�
������
� (17)
]s[66.0 1max
−=v� (18)
]s[33.0 1min
−−=v� (19)
Air flow through valve 1V : ]/s[111 lpK=Φ (20)
Water flow through valve 2V : ( ) ]/s[21122 lhhKpK RW −+=Φ (21)
Water flow to the separator R1: ]/s[1644.0 lw =Φ (22)
Air flow to the separator R1: ]/s[615.146.6 1110 lppairairair ⋅−=⋅+= ΦΦΦ (23)
The change of water level in R1 is:
( ) ]m/s[12
1
1Fw K
Sdtdh
ΦΦ −= (24)
The change of water level in R2 is:
( ) ]m/s[12
2
2Fw K
Sdtdh
ΦΦ −= (25)
The change of air pressure inside R1 is:
( ) ( )( )[ ] ]bar/s[121010
1FwFair KppKp
Vdtdp
ΦΦΦΦ −++−= (26)
The air volume inside R1 is:
( ) ( ) ]m[25.2 311111 hShhSV R −=−= (27)
A linearized model (Vrančić et al., 1995) can be obtained from the existing non-linear model:
∆∆
+
∆∆
=
∆∆
2
1
1
1
1
1
vv
hp
hp
cc BA��
(28)
where 1p∆ and 1h∆ denote the deviations of separator gas pressure 1p and liquid level 1h from the
steady-state values ( sppp 111 −=∆ , shhh 111 −=∆ ), and 1v∆ and 2v∆ are respectively the deviations of
the two valves positions 1v and 2v ( svvv 111 −=∆ , svvv 222 −=∆ ). The elements of matrices cA and cB
are computed as follows (Vrančić et al., 1995):
9 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
10
( )
+−
−=
sV
ss
s
sair
s
F
pK
ppp
Kp
VKa
2
210
1
11011 22
Φ (29)
( )
+−=
sV
wss
s
F
pKK
ppVK
a2
21012
2(30)
−=
sV
sF
pK
SK
a2
2
121
2(31)
−=
sV
wsF
pKK
SK
a2
2
122
2(32)
( )111011 ln Vsss
F RpKpVKb −= (33)
( ) ( )2221012 ln VsVsss
F RpKppVKb +−= (34)
021 =b (35)
( )2221
22 ln VsVsF RpK
SKb −= (36)
where sK1 , sK2 and sV denote the steady-state values of 1K , 2K and V , respectively. The steady-state
pressure on the valve 2V is marked as sVp 2 and is equal to:
( ) ]bar[2112 RwsV hhKpp −+= (37)
By using the above equations, a linear continuous-time model is obtained that corresponds to the
following steady state:
7462.0,4152.0,]m[4.1,]bar[5.0 2111 ==== ssss vvhp (38)
From the continuous-time model, a linear discrete-time model corresponding to sampling interval
s1=sT is obtained by applying the exact discretization using zero-order hold (the control inputs are
assumed piecewise constant over the sampling period) (Ogata, 1995). The state and control matrices of
the discrete-time model are:
=
0.99990.0006-0.0001-0.9719
A ,
=
0.0023-00.0041-0.0832-
B (39)
10 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
11
The state variables are ]bar[11 px ∆= and ]m[12 hx ∆= , and the control variables are 11 vu ∆= and
22 vu ∆= . The following input and rate constraints are imposed on the valve positions 1v and 2v :
10 1 ≤≤ v , 8625.00 2 ≤≤ v (40)
66.033.0 1 ≤≤− v� , 66.033.0 2 ≤≤− v� (41)
which by taking into account the steady state values (38) are represented as the following constraints on
the control inputs 1u and 2u :
0.5848)(0.4152- 1 ≤+≤ ktu , 1-N0,1,...,k,0.1163)(0.7462- 2 =≤+≤ ktu (42)
1-N,1,...,0k,2,1,66.0)1()(33.0 ==≤−+−+≤− iTktuktuT siis (43)
In order to avoid the steady state offset of the MPC controller, two more states are added to the model
(39), which take into account the integral error:
)()()1( 133 txTtxtx s+=+ , )()()1( 244 txTtxtx s+=+ (44)
Thus, the linear discrete-time model of the gas-liquid separation unit becomes:
=
100010000.99990.0006-000.0001-0.9719
s
s
TT
A ,
=
0000
0.0023-00.0041-0.0832-
B (45)
4. Real-time performance of explicit MPC controller for the gas-liquid
separation plant.
4.1. Design of approximate explicit MPC controller for the gas-liquid separation
plant.
The approximate mp-QP approach described in section 2 is applied to design an explicit MPC controller
for the gas-liquid separation plant. The controller minimizes the cost function (4) subject to the system
equation (45) and the input constraints (42). In (4), P is chosen as the solution of the discrete-time
algebraic Riccati equation and the cost matrices are:
}0001.0,005.0,100,05.0{diag=Q , }1,1{diag=R (46)
The horizon is 500=N and the time instants at which the input variables can change are:
11 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
12
]310308306304302300
11010810610410210050454035302520151051[1
=uN(47)
]30010050454035302520151051[2
=uN (48)
which makes totally 36 optimization variables. With this MPC problem, numerous tests showed that the
rate constraints (43) almost never become active and there are no practical benefits of including them in
the problem formulation. Also, in order to limit the complexity of the explicit MPC by keeping the
dimension of the state space low, the rate constraints were only ensured via rate limiters rather than
directly included in the optimization problem.
The state space to be partitioned is 4-dimensional and is defined by
]60,10[]3,3[]2.0,2.0[]5.0,5.0[ −×−×−×−=X . The size of the regions on each of the state variables is
restricted to be larger than 01.0∆ 1 =x , 004.0∆ 2 =x , 06.0∆ 3 =x and 7.0∆ 4 =x . The approximation
tolerance 0>ε is chosen according to the approach in (Johansen & Grancharova, 2003) and it depends
on 0X such that 2
5.0 T00Σxx+
=ε , where Σxxxx0
T
0
minargX∈
= . The MPC controller has 2693 regions in
its state space partition and 24 levels of search. With one scalar comparison required at each level of the
k-d tree, 24 arithmetic operations are required in the worst case to determine which region the state
belongs to. Totally, 40 arithmetic operations are needed in real-time to compute the two control inputs
(24 comparisons, 8 multiplications and 8 additions). For comparison, the average computation time with
the conventional MPC approach is 0.02 sec for 2.8 GHz Pentium 4 computer, which indicates that
millions of arithmetic operations are required in real-time to solve the QP problem (8).
4.2. Environment for the real-time experiments.
The real-time experiments were pursued in the environment schematically shown in Fig. 4. This
environment encompasses supervisory control on two levels: upper level with Factory Link SCADA
system and lower procedural and basic control levels implemented in two PLCs. This is one of the
possible configurations of control, which can be found in industry. User-friendly experimentation with
the process plant is enabled through interface with Matlab/Simulink environment. This interface enables
PLC access with Matlab/Simulink using DDE protocol via Serial Communication Link RS232 or
TCP/IPv4 over Ethernet IEEE802.3.
12 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
13
Control algorithms for experimentation can be prepared in Matlab code or as Simulink blocks
and extended with functions/blocks, which access PLC. This interface also enables user-friendly data
acquisition for Matlab users. In our case the control schemes were put together as Simulink blocks and
tested at the plant operating points as described in the previous section.
4.3. Real-time performance of the explicit MPC controller.
In Fig. 5 to 8, the real-time performance of the approximate explicit MPC controller, in closed-loop with
the plant is shown. The initial condition is ]bar[3.0)0(1 =p , ]m[5.1)0(1 =h and the set point is
]bar[5.01*1 == spp , ]m[4.11
*1 == shh . The experimental results (the solid line) are compared with the
exact MPC trajectory (the dotted line) computed by solving the optimization problem at each time
instant, based on the process model and with the simulated approximate trajectory (the dashed line). The
latter two curves (the dotted curve and the dashed curve) are difficult to distinguish since they are almost
matching. It can be seen from the figures that the explicit MPC controller brings the plant to the desired
set-point despite of the error in the steady state process model and the transient performance is close to
that of the optimal trajectory. The set point changes are handled by using the new set-point values *1p
and *1h when determining the values of the state variables *
111 ppx −= and *112 hhx −= , where 1p and
1h are the measured variables.
It can be seen from Fig. 6 that there is a slight chattering of the signal for the second valve. This
can be explained as follows. The signal we depict has come out of analog digital converter which has 10
bit A/D converter resolution that means approximately 0.1 % of quantisation noise on the range 0 to 1
which was used in our case. Afterwards this signal has gone through the controller with gain of about 10
which amplified the quantisation noise to about 1% (as it can be seen in the figure). This is not a problem
for the valve and actually even helps beating the small dead zone contained in valve and makes it react
faster when the change of control signal occurs.
In Fig. 9 and Fig. 10, the computed approximate PWL feedback laws for valve positions 1v and
2v are shown, together with the exact control laws computed by solving the QP problem (8) on a grid.
The corresponding sub-optimal and optimal cost functions are given in Fig. 11. It can be noticed from
Fig.11 that for ]m[5.11 =h the cost function makes a jump. This artifact is due to the implementation of
13 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
14
the anti-windup strategy. The cost function is computed by taking into account also the integral error (cf.
equation (44)), which is reset by applying the anti-windup strategy in (Åström & Wittenmark, 1997).
Table 1 shows statistics about the performance decrease Jε as well as the errors 1uε and
2uε in
the two control inputs with the approximate explicit MPC controller, based on simulations for 3600
combinations of initial and target states. The errors Jε ,1uε and
2uε are computed as the relative
deviations (measured in %) of the sub-optimal cost function and control input values (respectively )(ˆ xJ
and 2,1,)(ˆ =ixui ) from the optimal ones (respectively )(* xJ and 2,1,)(* =ixui ):
%100)(
)()(ˆ)( *
*
⋅−=
xJxJxJxJε (49)
2,1,%100)()(ˆ
min,max,
*
=⋅−
−= i
uuxuxu
ii
iiui
ε (50)
In (50), iumin, and iumax, , 2,1=i are the lower and upper bounds in the control input constraints (42). It
can be seen from Table 1 and from Fig. 5 to 11 that the performance decrease with the approximate
explicit MPC is not prohibitive, and the experimental results for initial condition ]bar[3.0)0(1 =p ,
]m[5.1)0(1 =h and set point ]bar[5.01*1 == spp , ]m[4.11
*1 == shh correspond to a typical amount of
sub-optimality.
Another approximate explicit MPC controller for the gas-liquid separation plant has been
designed by applying the method based on the control input approximation error (Grancharova &
Johansen, 2002). The experiments showed that this controller has performance very similar to that of the
approximate explicit MPC controller which minimizes the cost function approximation error (described
above).
5. Conclusions.
An approximate explicit MPC approach has been experimentally tested on a two-input two-output
laboratory gas-liquid separation plant. The approach achieves performance close to that of conventional
MPC, but requires only a fraction of the real-time computational machinery, leading to fast and reliable
computations. The main limitation of the approach is that the complexity of state space partition tends to
14 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
15
increase rapidly with the state space dimension, which makes the approach unsuitable for larger
problems. A convex nonlinear extension of the approximate explicit MPC approach has recently been
reported by Johansen (2004).
Appendix
List of the constants and variables related to the gas-liquid separation unit (Vrančić et al., 1995):
Constant Value Unit Description
1S 0.312 m2 Cross-section area of R1
2S 0.32 m2 Cross-section area of R2
0p 1.033 bar Normal atmospheric pressure
1Rh 2.25 m Height of the separator R1
2Rh 2 m Height of the reservoir R2
01K 75.1 )bar/(sl Flow coefficient of valve 1V
1VR 46.1 - The open-close flow ratio of valve 1V
02K 0.742 )bar/(sl Flow coefficient of valve 2V
2VR 75.66 - The open-close flow ratio of valve 2V
wK 0.0981 bar/m The proportional factor between the water level in [m] and
pressure in [bar]
FK 1e-3 m3/l Proportional factor between the flow in m3/s and l/s
wΦ 0.1644 l/s Water flow to the R1
0airΦ 6.46 l/s Air flow to R1 (at the atmospheric pressure)
1airΦ -1.615 l/(s⋅bar) The proportional factor between air flow and pressure
inside R1
max2r 0.8625 - The maximum value of the variable 2v (the position of 2V )
maxv� 0.66 s-1 Maximum speed of valves opening
minv� -0.33 s-1 Maximum speed of valves closing
15 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
16
Variable Unit Description
airΦ l/s Air flow to the separator R1
1Φ l/s Air flow from the separator R1
2Φ l/s Water flow from the separator R1
1p bar Excess of gas pressure above the normal atmospheric pressure inside R1
p bar Absolute air pressure inside R1 : )( 10 ppp +=
sp1 bar Steady-state value of pressure 1p used for linearization
sVp 2 bar Steady-state pressure on the valve 2V
1h m Water level inside R1
sh1 m Steady-state value of water level 1h used for linearization
2h m Water level inside R2
V m3 Gas volume inside R1
sV m3 Steady-state value of gas volume V
m kg Mass of the gas inside R1
1K )bar/(sl Flow coefficient of valve 1V
sK1 )bar/(sl Steady-state value of the flow coefficient 1K
2K )bar/(sl Flow coefficient of valve 2V
sK2 )bar/(sl Steady-state value of the flow coefficient 2K
1v - Actual position of the valve 1V
1v∆ - Deviations from the steady-state svvv 111 −=∆
1r - Valve command signal (desired position of 1V )
2v - Actual position of valve 2V
2v∆ - Deviations from the steady-state svvv 222 −=∆
2r - Valve command signal (desired position of 2V )
16 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
17
Acknowledgements
We would like to thank to Boštjan Čečko for his assistance when performing the experiments on the gas-
liquid separation plant and to Dr. Damir Vrančić for his useful advices.
This work was sponsored by the European Commission through the Research Training Network MAC
(“Multi-Agent Control: Probabilistic reasoning, optimal coordination, stability analysis and controller
design for intelligent hybrid systems”, HPRN-CT-1999-00107).
References
Åström, K., & Wittenmark, B. (1997). Computer-controlled systems. Theory and design. Prentice-Hall,
Inc., Upper Saddle River, New Jersey.
Bemporad, A., Morari, M., Dua, V., & Pistikopoulos, E. N. (2000). The explicit solution of model
predictive control via multiparametric quadratic programming. Proc. of American Control
Conference, Chicago, 872–876.
Bemporad, A., Morari, M., Dua, V., & Pistikopoulos, E. N. (2002). The explicit linear quadratic
regulator for constrained systems. Automatica, 38, 3-20.
Bemporad, A., & Filippi, C. (2001). Suboptimal explicit MPC via approximate quadratic programming.
Proc. of IEEE Conf. Decision and Control, Orlando, 4851-4856.
Bentley, J. L. (1975). Multidimensional binary search trees used for associative searching.
Communications of the ACM, 18, 509–517.
Biegler, L. T., & Rawlings, J. B. (1991). Optimization approaches to nonlinear model predictive control.
Proc. of Chemical Process Control IV, 543–571.
Chmielewski, D., & Manousiouthakis, V. (1996). On constrained infinite-time linear quadratic optimal
control. Systems and Control Letters, 29, 121–129.
Garcia, C. E., Prett, D. M., & Morari, M. (1987). Model predictive control: Theory and practice, A
survey. Automatica, 25, 335-348.
Grancharova, A., & Johansen, T. A. (2002). Approximate explicit model predictive control incorporating
heuristics. Proc. of IEEE International Symposium on Computer Aided Control System Design,
Glasgow, Scotland, U.K., 92-97.
17 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
18
Grancharova, A., Johansen, T. A., & Kocijan, J. (2003). Explicit model predictive control of gas-liquid
separation plant. Proc. of European Control Conference, Cambridge, U.K.
Johansen, T. A., Petersen, I., & Slupphaug, O. (2002). Explicit sub-optimal linear quadratic regulation
with state and input constraints. Automatica, 38, 1099-1111.
Johansen, T. A., & Grancharova, A. (2002). Approximate explicit model predictive control implemented
via orthogonal search tree partitioning. Proc. of 15-th IFAC World Congress, Barcelona, Spain,
session T-We-M17.
Johansen, T. A., & Grancharova, A. (2003). Approximate explicit constrained linear model predictive
control via orthogonal search tree. IEEE Trans. Automatic Control, 48, 810-815.
Johansen, T. A. (2004). Approximate explicit receding horizon control of constrained nonlinear systems.
Automatica, 40, 293-300.
Mayne, D. Q., Rawlings, J. B., Rao, C. V., & Scokaert, P. O. M. (2000). Constrained model predictive
control: Stability and optimality. Automatica, 36, 789-814.
Ogata, K. (1995). Discrete-time control systems. Prentice-Hall, Inc., Englewood Cliffs, New Jersey.
Pistikopoulos, E. N., Dua, V., Bozinis, N. A., Bemporad, A. & Morari, M. (2000). On-line optimization
via off-line parametric optimization tools. Computers & Chemical Engineering, 24, 183-188.
Seron, M., De Dona, J. A., & Goodwin, G. C. (2000). Global analytical model predictive control with
input constraints. Proc. of IEEE Conf. Decision and Control, Sydney, TuA05–2.
Tøndel, P., Johansen, T. A., & Bemporad, A. (2003). An algorithm for multi-parametric quadratic
programming and explicit MPC solutions. Automatica, 39, 489-497.
Tøndel, P., & Johansen, T. A. (2002). Complexity reduction in explicit linear model predictive control.
Proc. of 15-th IFAC World Congress, Barcelona, Spain, session T-We-M17.
Vrančić, D., Juričić, D., & Petrovčič, J. (1995). Measurements and mathematical modelling of a semi-
industrial liquid-gas separator for the purpose of fault diagnosis. Report DP-7260, University of
Ljubljana, J. Stefan Institute, Ljubljana.
18 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
19
List of Tables
Table 1. The average and the maximal approximation errors in the cost function and in the two
control inputs for the approximate explicit MPC controller.
19 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
20
List of Figures
Fig. 1. Quad-tree partition.
Fig. 2. k – d tree partition.
Fig. 3. Process scheme of the separation unit.
Fig. 4. Scheme showing environment for control and experimentation.
Fig. 5. Trajectory of 1v (position of valve 1).
Fig. 6. Trajectory of 2v (position of valve 2).
Fig. 7. Trajectory of 1p (pressure in the separator).
Fig. 8. Trajectory of 1h (liquid level in the separator).
Fig. 9. Piecewise linear approximate feedback control law (top) and the exact feedback control
law (bottom) for valve position 1v .
Fig. 10. Piecewise linear approximate feedback control law (top) and the exact feedback control
law (bottom) for valve position 2v .
Fig. 11. Sub-optimal cost function (top) and optimal cost function (bottom).
20 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
21
Table 1.
Error in cost function
Jε [%]
Error in 1u
1uε [%]
Error in 2u
2uε [%]
Average 0.2491 0.5660 0.3652
Maximal 4.3625 7.3928 7.1224
21 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
22
��#$��� ��� ��� �� ����������� �� � �����#���� �$������ '�� )�������� � �������( ��� 2��$�� ��! 4555��( ���# �#�� �#� �����# ������ �" ��$�� �#� � � �# ������� � �#� ������ � ��$ ��%*#� ��#$��� �����# ���� � � # �����# ��� ������������� �#��� � #"������� ��� �� # �����# ����"���� � ��� �� ������� #"�������� ���� �$ �#����� ������ � � �� �������( ��% : $��� 4% �#�������# �$ �#� ����( ��" � ������ ����� ��� �����)� ��� �� ���# �����%
: $% 4% 6������� ���� � � � � ������$���� ��$ � � � 4�� ���� ��� �����������%
,� � ���" �#� ��$� �#� � �� ��� ��� �#� �#����$ �� B � % *#� �� � ��� � �#� ����� �������6� ��$� �#� � � ������ �#� ���� � � �#������� �<���=� �� �#� 4� ���� ��� � �#� #"��������( �" ��� �$ �� � 4� 6��% 2���� � �#������� �� �+D�( �� ���� � �#� ������ ���� �
� ��� B& �A ' �4D�
�� ������ � ���� ����� ��� � J���� � �#���� �� ��� ���� � �����( ����� ���� � �#�#"������� �% �� ����� ��� ����( �#� � �������� � �4D� � ����� � ��� ��"#����� �� � �����$ � ���� � �$ �#� ������ % � � ���� ������ ��� � � �#� � ���� ����� ��� �
J���� B&�A ' �4E�
�� �$ �����$�� $� � ���� ���
& B+
4�
��
��
& � ' B
+
4�
��
��
'
2���� � �# � ������ � �� ��� ���� � �������#� ���� ����
%B�����%�� %� �4<�
��� �4+� ��� �44�% ,� ��� � � �� ���� � ���� ��� � ��� � � �#� ������ ��� �� �������� �"�#� ����� I��� ��!��� �"
I��� B(���&�A '��� � ��� �4=�
�#��� ( � � � �$��� ���� �$ ���� � � �# ��� � �� ��������% *#� ��� ��� � ��� � � ���#������ �� � �# � � ��"#���� � � ��������" ��� �$ �#� ��� � ���
� B �������
I���� �4G�
�� ) B +� 4� ���� �% -�( � �#� ��� ����� � ����% � ������� �#�� ��� ������ ��� �������� % ����#� ������ �� � ��� �� � ��� ������� �#�� ���
������ ��� �������� � ( � ����#�� ��!������ ��#� ��$ � � � ������% /�#��� ��( �� ���� � �� �� 4
� �)����� '�� #"��������( ��� ������ �#��������� ����� ��� ���� �� ���# � �#���% *# ���$� �#� ��� �� ������ '�� �� �����%
��%��&'� ! (�""���& �'� ")*�+
+% ,� � �� '� �#� ���� � � � �#� �#�� #"�������( %�% � B ��% ��� �#� #"������� � �� ����������%
4% ����� ��" ��������� #"������� � � � % ,� ����# #"������� �� ���( �#� ��$� �#� ���� ������ �# �#� ���� � � � %
D% ������ �#� ���� � � �#� 6� �<���=� �� �!��� � ���# � �#� 4� ���� ��� � �#� #"�������� ��� �� ������ � �� ���� � �#��� 6����" #��� ���� ����� � ���� �� ������% :�� �#��� ��� ��� �� ��� �� ���# ���� �( ������ �#����� � ���� �� ��� ���� � �+D�%
E% :�� �#� ���� � ���� �� ��� ���� �� �� �#����� ��� � �#� #"�������( ������ � ��� �������� � ���� ����� ������� �4E� �� �� ����� ���� � � �� ���� � �#� #"������� �%
<% 0����� �� � �#� #"������� ����� � �� ��� � ����� � ������ �#� ��� ����� � ����� ��� ����� ���� % � �#� ������ �� � ��� �� ����Æ% ,� �( $ � ���� =% /�#��� ��( ��� � ���������� $ � ���� 4%
=% �� � �#� #"������� � �� #"�������� ��(�( %%%( �� % ��� �#�� ��� ���������( ������ ��� � ( ��� ��( �( %%%( �� � � ��� $ ����� 4%
�
*# � ��$� �#� � �� ���� ���� � �# � � ���� ����� ���� ��� ��� ����� � �#�� � �� ������ ��� � � �#� ��� ���� ��� ����� ���� �%
������ ,� ��� ��� �#� ���6� ������ �<���=� ��!��� � � #"������� � % ����� ( � 5
��� %� Æ � 5 ��� $ ��� ��� �� ��)� �� � ��� < ���$� �#� + �#�� % � % ��� Æ � Æ ��������� ���%*#�� ��$� �#� + ���� ����� ����� � !� �� ������� ����� � �# �� ����� ���� ���� � J���� ���
���� ���� ��� J� ��� �#�� ��� �!��
������
J� ��� � � ���� � % �4>�
��� ������ ��� ��� ��" �$
������
��J���� �� � ����
�
*���� �%%%
*���� �
� � �43�
Fig. 1.
22 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
23
Fig. 2.
23 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
24
Fig. 3.
24 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
25
Personal computer 1
Personal computer 2
FactoryLink
Matlab-Simulink
SCADAfor proc.control
SCADAfor basiccontrol
DDE
MelDDE server
Procedural control
Basic control
Batchprocesscontrol
Continuous process control
Equipment + Supervisionand control
PLC Mitsubishi Q2AS-S1
PLC Mitsubishi A2SH-S1
R 232Et
hern
et
Mits
ubis
hi N
etB
ProcessOS IS
S
Fig. 4.
25 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
26
0 500 1000 15000
0.1
0.2
0.3
0.4
0.5
0.6
0.7
v1
time [s]
experimentssimulated exact MPC
simulated approximate explicit MPC
Fig. 5.
26 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
27
0 500 1000 15000.72
0.74
0.76
0.78
0.8
0.82
0.84
0.86
0.88
v2
time [s]
experiments
simulated exact MPC
simulated approximate explicit MPC
Fig. 6.
27 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
28
0 500 1000 15000.2
0.3
0.4
0.5
0.6
0.7
0.8p1 [bar]
time [s]
experimentssimulated exact MPC
simulated approximate explicit MPC
Fig. 7.
28 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
29
0 500 1000 15001.38
1.4
1.42
1.44
1.46
1.48
1.5
h1 [m]
time [s]
experimentssimulated exact MPC
simulated approximate explicit MPC
Fig. 8.
29 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
30
00.2
0.40.6
0.81
1.2
1.3
1.4
1.5
1.6
0
0.5
1
h1
p1
v1(p1,h1)
00.2
0.40.6
0.81
1.2
1.3
1.4
1.5
1.6
0
0.5
1
h1
p1
v1* (p1,h1)
Fig. 9.
30 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
31
0
0.2
0.4
0.6
0.8
1 1.2
1.3
1.4
1.5
1.6
0
0.2
0.4
0.6
0.8
h1
p1
v2(p1,h1)
0
0.2
0.4
0.6
0.8
1 1.2
1.3
1.4
1.5
1.6
0
0.2
0.4
0.6
0.8
h1
p1
v2* (p1,h1)
Fig. 10.
31 of 32
Monday , May 03, 2004
Elsevier
Revie
w C
opy
32
0
0.2
0.4
0.6
0.8
1 1.2
1.3
1.4
1.5
1.6
0
500
1000
1500
h1
p1
J(p1,h1)
0
0.2
0.4
0.6
0.8
1 1.2
1.3
1.4
1.5
1.6
0
500
1000
1500
h1
p1
J*(p1,h1)
Fig. 11.
32 of 32
Monday , May 03, 2004
Elsevier