$title Mission Planning for Synthetic Aperture Radar Surveillance (SWATH,SEQ=325)
$onText
The Microwave Radar Division of the Defence Sciences and Technology
Organisation employs synthetic aperture radar to obtain
high-resolution images of selected ground targets. It uses this
device, mounted on an aircraft, to scan up to 20 rectangular regions
called swaths to obtain images with resolutions down to one
meter. Missions consisting of a designated sequence of swaths and
flight paths are planned using mapping software. Previously DSTO had
been determining the best tours for missions by visually tracking
possible swath sequences from a starting base to an ending base. This
method was time consuming and did not guarantee optimality interms of
distance traveled. We developed optimization software tools to plan
mission tours more efficiently. DSTO can now plan missions with up to
20 swaths in a few seconds, rather than the hour it took using the
visual approach. Proposed tour lengths show an average improvement of
15 percent over those manually produced. The software incorporates
methods for dealing with the operational problems of no-fly zones and
shadowing associated with images.
Panton, D M, and Elbers, A W, Mission Planning for Synthetic Aperture
Radar Surveillance. Interfaces 29, 2 (1999), 73-88.
Keywords: mixed integer linear programming, Miller-Tucker-Zemlin subtour elimination,
iterative subtour elimination, traveling salesman problem, military application,
network optimization, synthetic aperture radar surveillance
$offText
Set
s 'swaths' / s0*s20 /
n 'nodes' / n1*n4 /
sx(s,n) 'valid swath node combinations' / s0.n1, (s1*s20).set.n /
a(s,n,s,n) 'arcs';
Alias (s,i,j), (n,np);
Parameter l(s,n,s,n) 'arc length';
$gdxIn swathdat.gdx
$load l
a(sx,s,n) = l(sx,s,n);
* TSP Type model
Variable
x(s,n,s,n) 'TSP tour between nodes'
y(s,s) 'TSP tour between swath'
z 'objective';
Binary Variable x, y;
Equation
defobj 'objective'
defone(s) 'one entering arc per swath'
defbal(s,n) 'flow balance'
defy(s,s) 'TSP swath tour determined by TSP node tour';
defobj.. z =e= sum(a, l(a)*x(a));
defone(s).. sum(a(sx,s,n), x(a)) =e= 1;
defbal(sx).. sum(a(s,n,sx), x(a)) - sum(a(sx,s,n), x(a)) =e= 0;
defy(i,j)$(not sameas(i,j)).. y(i,j) =e= sum(a(i,n,j,np), x(a));
$if not set orgse $goto secuts
* Original subtour elimination constraint from the SWATH MIPLIB 2003 instance
Positive Variable u(s);
Equation se(s,s) 'subtour elimination';
se(i,j)$(ord(i) > 1 and ord(j) > 1 and not sameas(i,j))..
u(i) - u(j) + card(s)*y(i,j) =l= card(s)-1;
u.fx('s0') = 0;
$goto solve
$label secuts
Set
cc 'subtour elimination cuts' / c1*c150 /
c(cc) 'active cuts';
c(cc) = no;
Parameter
cutcoeff(cc,s,s) 'coeffients for the subtour elimination cuts'
rhs(cc) 'right hand side for the subtour elimination cuts';
cutcoeff(c,s,s) = 0;
rhs(c) = 0;
Equation cut(cc) 'dynamic subtour elimination cuts';
cut(c).. sum((i,j), cutcoeff(c,i,j)*y(i,j)) =l= rhs(c);
$label solve
Model swath / all /;
option optCr = 0, limRow = 0, limCol = 0, solPrint = off;
* Solve without subtour elimination constraints
solve swath min z using mip;
$if set orgse $exit
Set
t 'tours' / t1*t25 /
tour(i,j,t) 'subtours'
from(i) 'contains always one element: the from swath'
next(j) 'contains always one element: the to swath'
visited(i) 'flag whether a swath is already visited'
tt(t) 'contains always one element: the current subtour'
curc(cc) 'contains always one element: the current SE cut' / c1 /;
Scalar goon 'go on flag used to control loop' / 1 /;
$eolCom //
while(goon = 1,
// Start tour in first swath
from(i) = no;
tt(t) = no;
tour(i,j,t) = no;
visited(i) = no;
from('s0') = yes;
tt('t1') = yes;
y.l(i,j) = round(y.l(i,j));
while(card(from),
// find swath visited after swath 'from'
next(i) = no;
loop((from,i)$y.l(from,i), next(i) = yes;);
tour(from,next,tt) = yes; // store in table
visited(from) = yes; // mark swath 'from' as visited
from(j) = next(j);
if(sum(visited(next),1) > 0, // if we are back at the start of the tour
// find starting point of new subtour
tt(t) = tt(t-1);
from(i) = no;
goon = 1;
loop(i$(not visited(i) and goon), from(i) = yes; goon = 0;);
);
);
display tour;
if(tt('t2'),
goon = 0; // just one tour -> stop
else // else: introduce cut(s)
// in case of two tours, we get complement cuts so eliminate the long tour
if(tt('t3'),
if(sum(tour(i,j,'t1'),1) < sum(tour(i,j,'t2'),1),
tour(i,j,'t2') = no;
else
tour(i,j,'t1') = tour(i,j,'t2');
tour(i,j,'t2') = no;
);
tt('t2') = yes;
);
goon = 1;
loop(t$goon,
rhs(curc) = -1;
visited(i) = sum(tour(i,j,t),1);
loop(tour(i,j,t),
cutcoeff(curc,i,visited) = 1;
rhs(curc) = rhs(curc) + 1;
);
c(curc) = yes;
curc(cc) = curc(cc-1);
goon = card(curc) and not tt(t+1);
);
abort$(card(curc) = 0) 'set cc to small';
solve swath using mip minimizing z;
goon = 1;
);
);