def test_modify_problem(self):
from dymos.examples.vanderpol.vanderpol_dymos import vanderpol
from dymos.examples.vanderpol.vanderpol_dymos_plots import vanderpol_dymos_plots
from dymos.run_problem import modify_problem, run_problem
from scipy.interpolate import interp1d
from numpy.testing import assert_almost_equal
# Create the Dymos problem instance
p = vanderpol(transcription='gauss-lobatto', num_segments=75)
# Run the problem (simulate only)
p.run_model()
# simulate and record
p.model.traj.simulate(record_file='vanderpol_simulation.sql')
# create a new problem for restart to simulate a different command line execution
q = vanderpol(transcription='gauss-lobatto', num_segments=75)
# Call modify_problem with simulation restart database
# modify_problem(q, restart='vanderpol_simulation.sql')
# # Run the model
run_problem(q, restart='vanderpol_simulation.sql')
# The solution should look like the explicit time history for the states and controls.
DO_PLOTS = False
if DO_PLOTS:
vanderpol_dymos_plots(q) # only for visual inspection and debug
else: # automate comparison
s = q.model.traj.simulate()
# get_val returns data for duplicate time points; remove them before interpolating
tq = q.get_val('traj.phase0.timeseries.time')[:, 0]
nodup = np.insert(tq[1:] != tq[:-1], 0, True)
tq = tq[nodup]
x1q = q.get_val('traj.phase0.timeseries.states:x1')[:, 0][nodup]
x0q = q.get_val('traj.phase0.timeseries.states:x0')[:, 0][nodup]
uq = q.get_val('traj.phase0.timeseries.controls:u')[:, 0][nodup]
ts = s.get_val('traj.phase0.timeseries.time')[:, 0]
nodup = np.insert(ts[1:] != ts[:-1], 0, True)
ts = ts[nodup]
x1s = s.get_val('traj.phase0.timeseries.states:x1')[:, 0][nodup]
x0s = s.get_val('traj.phase0.timeseries.states:x0')[:, 0][nodup]
us = s.get_val('traj.phase0.timeseries.controls:u')[:, 0][nodup]
# create interpolation functions so that values can be looked up at matching time points
fx1s = interp1d(ts, x1s, kind='cubic')
fx0s = interp1d(ts, x0s, kind='cubic')
fus = interp1d(ts, us, kind='cubic')
assert_almost_equal(x1q, fx1s(tq), decimal=2)
assert_almost_equal(x0q, fx0s(tq), decimal=2)
assert_almost_equal(uq, fus(tq), decimal=5)
def test_vanderpol_for_docs_optimize_refine(self):
import dymos as dm
from dymos.examples.plotting import plot_results
from dymos.examples.vanderpol.vanderpol_dymos import vanderpol
# Create the Dymos problem instance
p = vanderpol(transcription='gauss-lobatto',
num_segments=15,
transcription_order=3,
compressed=True,
optimizer='SLSQP')
# Enable grid refinement and find optimal control solution to stop oscillation
p.model.traj.phases.phase0.set_refine_options(refine=True)
dm.run_problem(p, refine=True, refine_iteration_limit=10)
# check validity by using scipy.integrate.solve_ivp to integrate the solution
exp_out = p.model.traj.simulate()
# Display the results
plot_results([
('traj.phase0.timeseries.time', 'traj.phase0.timeseries.states:x1',
'time (s)', 'x1 (V)'),
('traj.phase0.timeseries.time', 'traj.phase0.timeseries.states:x0',
'time (s)', 'x0 (V/s)'),
('traj.phase0.timeseries.states:x0',
'traj.phase0.timeseries.states:x1', 'x0 vs x1', 'x0 vs x1'),
('traj.phase0.timeseries.time',
'traj.phase0.timeseries.controls:u', 'time (s)', 'control u'),
],
title='Van Der Pol Optimization with Grid Refinement',
p_sol=p,
p_sim=exp_out)
plt.show()
def test_vanderpol_for_docs_simulation(self):
from dymos.examples.plotting import plot_results
from dymos.examples.vanderpol.vanderpol_dymos import vanderpol
# Create the Dymos problem instance
p = vanderpol(transcription='gauss-lobatto', num_segments=75)
# Run the problem (simulate only)
p.run_model()
# check validity by using scipy.integrate.solve_ivp to integrate the solution
exp_out = p.model.traj.simulate()
# Display the results
plot_results([
('traj.phase0.timeseries.time', 'traj.phase0.timeseries.states:x1',
'time (s)', 'x1 (V)'),
('traj.phase0.timeseries.time', 'traj.phase0.timeseries.states:x0',
'time (s)', 'x0 (V/s)'),
('traj.phase0.timeseries.states:x0',
'traj.phase0.timeseries.states:x1', 'x0 vs x1', 'x0 vs x1'),
('traj.phase0.timeseries.time',
'traj.phase0.timeseries.controls:u', 'time (s)', 'control u'),
],
title='Van Der Pol Simulation',
p_sol=p,
p_sim=exp_out)
plt.show()
def test_vanderpol_optimal_mpi(self):
"""to test with MPI:
OPENMDAO_REQUIRE_MPI=1 mpirun -n 4 python
dymos/examples/vanderpol/test/test_vanderpol.py TestVanderpolExampleMPI.test_vanderpol_optimal_mpi
(using varying values for n should give the same answer)
"""
p = vanderpol(transcription='gauss-lobatto',
num_segments=75,
delay=True,
use_pyoptsparse=True,
optimizer='IPOPT')
p.run_driver() # find optimal control solution to stop oscillation
if SHOW_PLOTS:
vanderpol_dymos_plots(p)
print('Objective function minimized to',
p.get_val('traj.phase0.states:J')[-1, ...])
# check that ODE states (excluding J) and control are driven to near zero
assert_almost_equal(
p.get_val('traj.phase0.states:x0')[-1, ...], np.zeros(1))
assert_almost_equal(
p.get_val('traj.phase0.states:x1')[-1, ...], np.zeros(1))
assert_almost_equal(p.get_val('traj.phase0.controls:u')[-1, ...],
np.zeros(1),
decimal=3)
def test_vanderpol_simulate(self):
# simulate only: with no control, the system oscillates
p = vanderpol(transcription='gauss-lobatto', num_segments=75)
p.run_model()
if SHOW_PLOTS:
vanderpol_dymos_plots(p)
def test_vanderpol_optimal_slow(self):
"""to test with MPI:
OPENMDAO_REQUIRE_MPI=1 mpirun -n 4 python
dymos/examples/vanderpol/test/test_vanderpol.py TestVanderpolExample.test_vanderpol_optimal_slow
(using varying values for n should give the same answer)
"""
p = vanderpol(transcription='gauss-lobatto',
num_segments=75,
delay=True)
dm.run_problem(p) # find optimal control solution to stop oscillation
if SHOW_PLOTS:
vanderpol_dymos_plots(p)
print('Objective function minimized to',
p['traj.phases.phase0.final_conditions.states:J++'])
# check that ODE states (excluding J) and control are driven to near zero
assert_almost_equal(
p['traj.phases.phase0.final_conditions.states:x0++'], np.zeros(1))
assert_almost_equal(
p['traj.phases.phase0.final_conditions.states:x1++'], np.zeros(1))
assert_almost_equal(
p['traj.phases.phase0.final_conditions.controls:u++'],
np.zeros(1),
decimal=3)
def test_vanderpol_for_docs_optimize_refine(self):
import dymos as dm
from dymos.examples.vanderpol.vanderpol_dymos import vanderpol
from openmdao.utils.assert_utils import assert_near_equal
# Create the Dymos problem instance
p = vanderpol(transcription='gauss-lobatto',
num_segments=15,
transcription_order=3,
compressed=True,
optimizer='SLSQP')
# Enable grid refinement and find optimal control solution to stop oscillation
p.model.traj.phases.phase0.set_refine_options(refine=True)
dm.run_problem(p,
refine_iteration_limit=10,
simulate=True,
make_plots=True)
assert_near_equal(p.get_val('traj.phase0.states:x0')[-1, ...], 0.0)
assert_near_equal(p.get_val('traj.phase0.states:x1')[-1, ...], 0.0)
assert_near_equal(p.get_val('traj.phase0.states:J')[-1, ...],
5.2808,
tolerance=1.0E-3)
assert_near_equal(p.get_val('traj.phase0.controls:u')[-1, ...],
0.0,
tolerance=1.0E-3)
def test_modify_problem(self):
from dymos.examples.vanderpol.vanderpol_dymos import vanderpol
from dymos.run_problem import run_problem
from scipy.interpolate import interp1d
from numpy.testing import assert_almost_equal
# Create the Dymos problem instance
p = vanderpol(transcription='gauss-lobatto', num_segments=75)
# Run the problem (simulate only)
p.run_model()
# simulate and record
p.model.traj.simulate(record_file='vanderpol_simulation.sql')
# create a new problem for restart to simulate a different command line execution
q = vanderpol(transcription='gauss-lobatto', num_segments=75)
# # Run the model
run_problem(q, restart='vanderpol_simulation.sql')
s = q.model.traj.simulate(rtol=1.0E-9, atol=1.0E-9)
# get_val returns data for duplicate time points; remove them before interpolating
tq = q.get_val('traj.phase0.timeseries.time')[:, 0]
nodup = np.insert(tq[1:] != tq[:-1], 0, True)
tq = tq[nodup]
x1q = q.get_val('traj.phase0.timeseries.states:x1')[:, 0][nodup]
x0q = q.get_val('traj.phase0.timeseries.states:x0')[:, 0][nodup]
uq = q.get_val('traj.phase0.timeseries.controls:u')[:, 0][nodup]
ts = s.get_val('traj.phase0.timeseries.time')[:, 0]
nodup = np.insert(ts[1:] != ts[:-1], 0, True)
ts = ts[nodup]
x1s = s.get_val('traj.phase0.timeseries.states:x1')[:, 0][nodup]
x0s = s.get_val('traj.phase0.timeseries.states:x0')[:, 0][nodup]
us = s.get_val('traj.phase0.timeseries.controls:u')[:, 0][nodup]
# create interpolation functions so that values can be looked up at matching time points
fx1s = interp1d(ts, x1s, kind='cubic')
fx0s = interp1d(ts, x0s, kind='cubic')
fus = interp1d(ts, us, kind='cubic')
assert_almost_equal(x1q, fx1s(tq), decimal=2)
assert_almost_equal(x0q, fx0s(tq), decimal=2)
assert_almost_equal(uq, fus(tq), decimal=5)
def test_vanderpol_for_docs_simulation(self):
import dymos as dm
from dymos.examples.vanderpol.vanderpol_dymos import vanderpol
# Create the Dymos problem instance
p = vanderpol(transcription='gauss-lobatto', num_segments=75)
dm.run_problem(p, run_driver=False, simulate=True, make_plots=True)
def test_vanderpol_optimal(self):
p = vanderpol(transcription='gauss-lobatto', num_segments=75)
dm.run_problem(p) # find optimal control solution to stop oscillation
print('Objective function minimized to', p.get_val('traj.phase0.states:J')[-1, ...])
# check that ODE states (excluding J) and control are driven to near zero
assert_almost_equal(p.get_val('traj.phase0.states:x0')[-1, ...], np.zeros(1))
assert_almost_equal(p.get_val('traj.phase0.states:x1')[-1, ...], np.zeros(1))
assert_almost_equal(p.get_val('traj.phase0.controls:u')[-1, ...], np.zeros(1), decimal=3)
def test_vanderpol_optimal_grid_refinement(self):
# enabling grid refinement gives a faster and better solution with fewer segments
p = vanderpol(transcription='gauss-lobatto', num_segments=15)
p.model.traj.phases.phase0.set_refine_options(refine=True)
dm.run_problem(p, refine_iteration_limit=10) # enable grid refinement and find optimal solution
print('Objective function minimized to', p.get_val('traj.phase0.timeseries.states:J')[-1, ...])
# check that ODE states (excluding J) and control are driven to near zero
assert_almost_equal(p.get_val('traj.phase0.timeseries.states:x0')[-1, ...], np.zeros(1))
assert_almost_equal(p.get_val('traj.phase0.timeseries.states:x1')[-1, ...], np.zeros(1))
assert_almost_equal(p.get_val('traj.phase0.timeseries.controls:u')[-1, ...], np.zeros(1), decimal=4)
def test_vanderpol_simulate_true(self):
# simulate true
p = vanderpol(transcription='radau-ps',
num_segments=30,
transcription_order=3,
compressed=True,
optimizer='SLSQP',
delay=0.005,
distrib=True,
use_pyoptsparse=True)
dm.run_problem(p, run_driver=True, simulate=True)
def test_vanderpol_for_docs_optimize(self):
import dymos as dm
from dymos.examples.vanderpol.vanderpol_dymos import vanderpol
# Create the Dymos problem instance
p = vanderpol(transcription='gauss-lobatto',
num_segments=75,
transcription_order=3,
compressed=True,
optimizer='SLSQP')
dm.run_problem(p, simulate=True, make_plots=True)
def test_vanderpol_optimal(self):
p = vanderpol(transcription='gauss-lobatto', num_segments=75)
dm.run_problem(p) # find optimal control solution to stop oscillation
if SHOW_PLOTS:
vanderpol_dymos_plots(p)
print('Objective function minimized to',
p['traj.phases.phase0.final_conditions.states:J++'])
# check that ODE states (excluding J) and control are driven to near zero
assert_almost_equal(
p['traj.phases.phase0.final_conditions.states:x0++'], np.zeros(1))
assert_almost_equal(
p['traj.phases.phase0.final_conditions.states:x1++'], np.zeros(1))
assert_almost_equal(
p['traj.phases.phase0.final_conditions.controls:u++'],
np.zeros(1),
decimal=3)
def test_vanderpol_optimal_grid_refinement(self):
# enabling grid refinement gives a faster and better solution with fewer segments
p = vanderpol(transcription='gauss-lobatto', num_segments=15)
p.model.traj.phases.phase0.set_refine_options(refine=True)
dm.run_problem(p, refine=True, refine_iteration_limit=10
) # enable grid refinement and find optimal solution
if SHOW_PLOTS:
vanderpol_dymos_plots(p)
print('Objective function minimized to',
p['traj.phases.phase0.final_conditions.states:J++'])
# check that ODE states (excluding J) and control are driven to near zero
assert_almost_equal(
p['traj.phases.phase0.final_conditions.states:x0++'], np.zeros(1))
assert_almost_equal(
p['traj.phases.phase0.final_conditions.states:x1++'], np.zeros(1))
assert_almost_equal(
p['traj.phases.phase0.final_conditions.controls:u++'],
np.zeros(1),
decimal=4)
