Reference
Category : GAMS NOA library
Mainfile : glider.gms
$ontext
Maximize the final horizontal position of a hang glider while in the
presence of a thermal updraft.
This model is from the COPS benchmarking suite.
See http://www-unix.mcs.anl.gov/~more/cops/.
The number of discretization points can be specified using the GAMS user1
parameter.
COPS performance tests have been reported for nh = 50, 100, 200, 400
Dolan, E D, and More, J J, Benchmarking Optimization Software with COPS.
Tech. rep., Mathematics and Computer Science Division, 2000.
Bulirsch, R, Nerz, E, Pesch, H J, and von Stryk, O, Combining Direct and
Indirect Methods in Nonlinear Optimal Control: Range Maximization of a
Hang Glider. In Bulirsch, R, Miele, A, Stoer, J, and Well, K H, Eds,
Optimal Control. Birkhauser Verlag, 1993, pp. 273-288.
$offtext
$if set n $set nh %n%
$if not set nh $set nh 800
sets c coordinates / x distance
y altitude /
h intervals / h0 * h%nh%/ ;
alias(h,i);
scalars nh Number of intervals in mesh / %nh% /
cL_min bound on control variable / 0.0 /
cL_max bound on control variable / 1.4 /
u_c / 2.5 /
r_0 / 100 /
m / 100 /
g / 9.81 /
c0 / 0.034 /
c1 / 0.069662 /
S / 14 /
rho / 1.13 / ;
parameters c_0(c) initial position / x 0, y 1000/
v_0(c) initial velocity / x 13.23, y -1.288/
c_f(c) final position / y 900/
v_f(c) final velocity / x 13.23, y -1.288/ ;
variables t_f
pos(c,h) position x distance y altitude
vel(c,h) velocity x distance y altitude
cl(h) control variables
r(h) the r function
u(h) the u function
w(h) the w function
v(h) the v function
D(h) the D function
L(h) the L function
v_dot(c,h)
final_x
step step size
positive variables step;
equations tf_eqn
rdef(h)
udef(h)
wdef(h)
vdef(h)
Ddef(h)
Ldef(h)
vx_dot_def(h)
vy_dot_def(h)
obj
pos_eqn(c,h)
vel_eqn(c,h);
tf_eqn.. t_f =e= step*nh;
rdef(i).. r[i] =e= sqr(pos['x',i]/r_0 - 2.5);
udef(i).. u[i] =e= u_c*(1-r[i])*exp(-r[i]);
wdef(i).. w[i] =e= vel['y',i] - u[i];
vdef(i).. v[i] =e= sqrt(sqr(vel['x',i]) + sqr(w[i]));
Ddef(i).. D[i] =e= .5*(c0+c1*sqr(cL[i]))*rho*S*sqr(v[i]);
Ldef(i).. L[i] =e= .5* cL[i] *rho*S*sqr(v[i]);
vx_dot_def(i).. v_dot['x',i] =e= (-L[i]*w[i]/v[i] - D[i]*vel['x',i]/v[i])/m;
vy_dot_def(i).. v_dot['y',i] =e= ( L[i]*vel['x',i]/v[i] - D[i]*w[i]/v[i])/m - g;
obj.. final_x =e= pos('x','h%nh%');
pos_eqn(c,i-1).. pos[c,i] =e= pos[c,i-1] + .5*step*(vel[c,i] + vel[c,i-1]);
vel_eqn(c,i-1).. vel[c,i] =e= vel[c,i-1] + .5*step*(v_dot[c,i] + v_dot[c,i-1]);
* Boundary Conditions
cl.lo(h) = cL_min;
cl.up(h) = cL_max;
pos.lo('x',h) = 0;
vel.lo('x',h) = 0;
v.lo(h) = 0.01;
* Fixed values
pos.fx(c,'h0') = c_0(c);
pos.fx('y','h%nh%') = c_f('y');
vel.fx(c,'h0') = v_0(c);
vel.fx(c,'h%nh%') = v_f(c);
* Initial point
pos.l('x',h) = c_0('x') + v_0('x')*((ord(h)-1)/nh);
pos.l('y',h) = c_0('y') + ((ord(h)-1)/nh)*(c_f('y') - c_0('y'));
vel.l(c,h) = v_0(c);
cL.l(h) = cL_max/2;
step.l = 1.0/nh;
* Initial values for intermediate variables
t_f.l = step.l*nh;
r.l[i] = sqr(pos.l['x',i]/r_0 - 2.5);
u.l[i] = u_c*(1-r.l[i])*exp(-r.l[i]);
w.l[i] = vel.l['y',i] - u.l[i];
v.l[i] = sqrt(sqr(vel.l['x',i]) + sqr(w.l[i]));
D.l[i] = .5*(c0+c1*sqr(cL.l[i]))*rho*S*sqr(v.l[i]);
L.l[i] = .5* cL.l[i] *rho*S*sqr(v.l[i]);
v_dot.l['x',i] = (-L.l[i]*w.l[i]/v.l[i] - D.l[i]*vel.l['x',i]/v.l[i])/m;
v_dot.l['y',i] = ( L.l[i]*vel.l['x',i]/v.l[i] - D.l[i]*w.l[i]/v.l[i])/m - g;
model glider /all/;
$iftheni x%mode%==xbook
glider.workspace = 40;
$endif
solve glider maximizing final_x using nlp;
$iftheni x%mode%==xbook
file res /g8.dat/;
put res
loop(h, put vel.l('y',h):10:7, put/)
$endif
* --------------------------- Numerical Experiments -------------------
* January 25, 2011
*
* For nh=1200 I got:
* CONOPT: 1547 iterations, 217.242 seconds, vfo=1247.9859370
* KNITRO: 604 iterations, 631.468 seconds, vfo=1247.9858984
*------
* End Glider