def run_example(with_plots=True):
"""
This is the same example from the Sundials package (idasRoberts_FSA_dns.c)
This simple example problem for IDA, due to Robertson,
is from chemical kinetics, and consists of the following three
equations::
dy1/dt = -p1*y1 + p2*y2*y3
dy2/dt = p1*y1 - p2*y2*y3 - p3*y2**2
0 = y1 + y2 + y3 - 1
"""
def f(t, y, yd, p):
res1 = -p[0]*y[0]+p[1]*y[1]*y[2]-yd[0]
res2 = p[0]*y[0]-p[1]*y[1]*y[2]-p[2]*y[1]**2-yd[1]
res3 = y[0]+y[1]+y[2]-1
return N.array([res1,res2,res3])
#The initial conditons
y0 = [1.0, 0.0, 0.0] #Initial conditions for y
yd0 = [0.1, 0.0, 0.0] #Initial conditions for dy/dt
p0 = [0.040, 1.0e4, 3.0e7] #Initial conditions for parameters
#Create an Assimulo implicit problem
imp_mod = Implicit_Problem(f, y0, yd0,p0=p0)
#Create an Assimulo implicit solver (IDA)
imp_sim = IDA(imp_mod) #Create a IDA solver
#Sets the paramters
imp_sim.atol = N.array([1.0e-8, 1.0e-14, 1.0e-6])
imp_sim.algvar = [1.0,1.0,0.0]
imp_sim.suppress_alg = False #Suppres the algebraic variables on the error test
imp_sim.continuous_output = True #Store data continuous during the simulation
imp_sim.pbar = p0
imp_sim.suppress_sens = False #Dont suppress the sensitivity variables in the error test.
#Let Sundials find consistent initial conditions by use of 'IDA_YA_YDP_INIT'
imp_sim.make_consistent('IDA_YA_YDP_INIT')
#Simulate
t, y, yd = imp_sim.simulate(4,400) #Simulate 4 seconds with 400 communication points
print imp_sim.p_sol[0][-1] , imp_sim.p_sol[1][-1], imp_sim.p_sol[0][-1]
#Basic test
nose.tools.assert_almost_equal(y[-1][0], 9.05518032e-01, 4)
nose.tools.assert_almost_equal(y[-1][1], 2.24046805e-05, 4)
nose.tools.assert_almost_equal(y[-1][2], 9.44595637e-02, 4)
nose.tools.assert_almost_equal(imp_sim.p_sol[0][-1][0], -1.8761, 2) #Values taken from the example in Sundials
nose.tools.assert_almost_equal(imp_sim.p_sol[1][-1][0], 2.9614e-06, 8)
nose.tools.assert_almost_equal(imp_sim.p_sol[2][-1][0], -4.9334e-10, 12)
#Plot
if with_plots:
P.plot(t,y)
P.show()
def run_example(with_plots=True):
"""
This is the same example from the Sundials package (idasRoberts_FSA_dns.c)
This simple example problem for IDA, due to Robertson,
is from chemical kinetics, and consists of the following three
equations::
dy1/dt = -p1*y1 + p2*y2*y3
dy2/dt = p1*y1 - p2*y2*y3 - p3*y2**2
0 = y1 + y2 + y3 - 1
"""
def f(t, y, yd, p):
res1 = -p[0] * y[0] + p[1] * y[1] * y[2] - yd[0]
res2 = p[0] * y[0] - p[1] * y[1] * y[2] - p[2] * y[1]**2 - yd[1]
res3 = y[0] + y[1] + y[2] - 1
return N.array([res1, res2, res3])
#The initial conditons
y0 = [1.0, 0.0, 0.0] #Initial conditions for y
yd0 = [0.1, 0.0, 0.0] #Initial conditions for dy/dt
p0 = [0.040, 1.0e4, 3.0e7] #Initial conditions for parameters
#Create an Assimulo implicit problem
imp_mod = Implicit_Problem(f, y0, yd0, p0=p0)
#Create an Assimulo implicit solver (IDA)
imp_sim = IDA(imp_mod) #Create a IDA solver
#Sets the paramters
imp_sim.atol = N.array([1.0e-8, 1.0e-14, 1.0e-6])
imp_sim.algvar = [1.0, 1.0, 0.0]
imp_sim.suppress_alg = False #Suppres the algebraic variables on the error test
imp_sim.continuous_output = True #Store data continuous during the simulation
imp_sim.pbar = p0
imp_sim.suppress_sens = False #Dont suppress the sensitivity variables in the error test.
#Let Sundials find consistent initial conditions by use of 'IDA_YA_YDP_INIT'
imp_sim.make_consistent('IDA_YA_YDP_INIT')
#Simulate
t, y, yd = imp_sim.simulate(
4, 400) #Simulate 4 seconds with 400 communication points
print imp_sim.p_sol[0][-1], imp_sim.p_sol[1][-1], imp_sim.p_sol[0][-1]
#Basic test
nose.tools.assert_almost_equal(y[-1][0], 9.05518032e-01, 4)
nose.tools.assert_almost_equal(y[-1][1], 2.24046805e-05, 4)
nose.tools.assert_almost_equal(y[-1][2], 9.44595637e-02, 4)
nose.tools.assert_almost_equal(
imp_sim.p_sol[0][-1][0], -1.8761,
2) #Values taken from the example in Sundials
nose.tools.assert_almost_equal(imp_sim.p_sol[1][-1][0], 2.9614e-06, 8)
nose.tools.assert_almost_equal(imp_sim.p_sol[2][-1][0], -4.9334e-10, 12)
#Plot
if with_plots:
P.plot(t, y)
P.show()
def simulate(model, init_cond, start_time=0., final_time=1., input=(lambda t: []), ncp=500, blt=True,
causalization_options=sp.CausalizationOptions(), expand_to_sx=True, suppress_alg=False,
tol=1e-8, solver="IDA"):
"""
Simulate model from CasADi Interface using CasADi.
init_cond is a dictionary containing initial conditions for all variables.
"""
if blt:
t_0 = timing.time()
blt_model = sp.BLTModel(model, causalization_options)
blt_time = timing.time() - t_0
print("BLT analysis time: %.3f s" % blt_time)
blt_model._model = model
model = blt_model
if causalization_options['closed_form']:
solved_vars = model._solved_vars
solved_expr = model._solved_expr
#~ for (var, expr) in itertools.izip(solved_vars, solved_expr):
#~ print('%s := %s' % (var.getName(), expr))
return model
dh() # This is not a debug statement!
# Extract model variables
model_states = [var for var in model.getVariables(model.DIFFERENTIATED) if not var.isAlias()]
model_derivatives = [var for var in model.getVariables(model.DERIVATIVE) if not var.isAlias()]
model_algs = [var for var in model.getVariables(model.REAL_ALGEBRAIC) if not var.isAlias()]
model_inputs = [var for var in model.getVariables(model.REAL_INPUT) if not var.isAlias()]
states = [var.getVar() for var in model_states]
derivatives = [var.getMyDerivativeVariable().getVar() for var in model_states]
algebraics = [var.getVar() for var in model_algs]
inputs = [var.getVar() for var in model_inputs]
n_x = len(states)
n_y = len(states) + len(algebraics)
n_w = len(algebraics)
n_u = len(inputs)
# Create vectorized model variables
t = model.getTimeVariable()
y = MX.sym("y", n_y)
yd = MX.sym("yd", n_y)
u = MX.sym("u", n_u)
# Extract the residuals and substitute the (x,z) variables for the old variables
scalar_vars = states + algebraics + derivatives + inputs
vector_vars = [y[k] for k in range(n_y)] + [yd[k] for k in range(n_x)] + [u[k] for k in range(n_u)]
[dae] = substitute([model.getDaeResidual()], scalar_vars, vector_vars)
# Fix parameters
if not blt:
# Sort parameters
par_kinds = [model.BOOLEAN_CONSTANT,
model.BOOLEAN_PARAMETER_DEPENDENT,
model.BOOLEAN_PARAMETER_INDEPENDENT,
model.INTEGER_CONSTANT,
model.INTEGER_PARAMETER_DEPENDENT,
model.INTEGER_PARAMETER_INDEPENDENT,
model.REAL_CONSTANT,
model.REAL_PARAMETER_INDEPENDENT,
model.REAL_PARAMETER_DEPENDENT]
pars = reduce(list.__add__, [list(model.getVariables(par_kind)) for
par_kind in par_kinds])
# Get parameter values
model.calculateValuesForDependentParameters()
par_vars = [par.getVar() for par in pars]
par_vals = [model.get_attr(par, "_value") for par in pars]
# Eliminate parameters
[dae] = casadi.substitute([dae], par_vars, par_vals)
# Extract initial conditions
y0 = [init_cond[var.getName()] for var in model_states] + [init_cond[var.getName()] for var in model_algs]
yd0 = [init_cond[var.getName()] for var in model_derivatives] + n_w * [0.]
# Create residual CasADi functions
dae_res = MXFunction([t, y, yd, u], [dae])
dae_res.setOption("name", "complete_dae_residual")
dae_res.init()
###################
#~ import matplotlib.pyplot as plt
#~ h = MX.sym("h")
#~ iter_matrix_expr = dae_res.jac(2)/h + dae_res.jac(1)
#~ iter_matrix = MXFunction([t, y, yd, u, h], [iter_matrix_expr])
#~ iter_matrix.init()
#~ n = 100;
#~ hs = np.logspace(-8, 1, n);
#~ conds = [np.linalg.cond(iter_matrix.call([0, y0, yd0, input(0), hval])[0].toArray()) for hval in hs]
#~ plt.close(1)
#~ plt.figure(1)
#~ plt.loglog(hs, conds, 'b-')
#~ #plt.gca().invert_xaxis()
#~ plt.grid('on')
#~ didx = range(4, 12) + range(30, 33)
#~ aidx = [i for i in range(33) if i not in didx]
#~ didx = range(10)
#~ aidx = []
#~ F = MXFunction([t, y, yd, u], [dae[didx]])
#~ F.init()
#~ G = MXFunction([t, y, yd, u], [dae[aidx]])
#~ G.init()
#~ dFddx = F.jac(2)[:, :n_x]
#~ dFdx = F.jac(1)[:, :n_x]
#~ dFdy = F.jac(1)[:, n_x:]
#~ dGdx = G.jac(1)[:, :n_x]
#~ dGdy = G.jac(1)[:, n_x:]
#~ E_matrix = MXFunction([t, y, yd, u, h], [dFddx])
#~ E_matrix.init()
#~ E_cond = np.linalg.cond(E_matrix.call([0, y0, yd0, input(0), hval])[0].toArray())
#~ iter_matrix_expr = vertcat([horzcat([dFddx + h*dFdx, h*dFdy]), horzcat([dGdx, dGdy])])
#~ iter_matrix = MXFunction([t, y, yd, u, h], [iter_matrix_expr])
#~ iter_matrix.init()
#~ n = 100
#~ hs = np.logspace(-8, 1, n)
#~ conds = [np.linalg.cond(iter_matrix.call([0, y0, yd0, input(0), hval])[0].toArray()) for hval in hs]
#~ plt.loglog(hs, conds, 'b--')
#~ plt.gca().invert_xaxis()
#~ plt.grid('on')
#~ plt.xlabel('$h$')
#~ plt.ylabel('$\kappa$')
#~ plt.show()
#~ dh()
###################
# Expand to SX
if expand_to_sx:
dae_res = SXFunction(dae_res)
dae_res.init()
# Create DAE residual Assimulo function
def dae_residual(t, y, yd):
dae_res.setInput(t, 0)
dae_res.setInput(y, 1)
dae_res.setInput(yd, 2)
dae_res.setInput(input(t), 3)
dae_res.evaluate()
return dae_res.getOutput(0).toArray().reshape(-1)
# Set up simulator
problem = Implicit_Problem(dae_residual, y0, yd0, start_time)
if solver == "IDA":
simulator = IDA(problem)
elif solver == "Radau5DAE":
simulator = Radau5DAE(problem)
else:
raise ValueError("Unknown solver %s" % solver)
simulator.rtol = tol
simulator.atol = 1e-4 * np.array([model.get_attr(var, "nominal") for var in model_states + model_algs])
#~ simulator.atol = tol * np.ones([n_y, 1])
simulator.report_continuously = True
# Log method order
if solver == "IDA":
global order
order = []
def handle_result(solver, t, y, yd):
global order
order.append(solver.get_last_order())
solver.t_sol.extend([t])
solver.y_sol.extend([y])
solver.yd_sol.extend([yd])
problem.handle_result = handle_result
# Suppress algebraic variables
if suppress_alg:
if isinstance(suppress_alg, bool):
simulator.algvar = n_x * [True] + (n_y - n_x) * [False]
else:
simulator.algvar = n_x * [True] + suppress_alg
simulator.suppress_alg = True
# Simulate
t_0 = timing.time()
(t, y, yd) = simulator.simulate(final_time, ncp)
simul_time = timing.time() - t_0
stats = {'time': simul_time, 'steps': simulator.statistics['nsteps']}
if solver == "IDA":
stats['order'] = order
# Generate result for time and inputs
class SimulationResult(dict):
pass
res = SimulationResult()
res.stats = stats
res['time'] = t
if u.numel() > 0:
input_names = [var.getName() for var in model_inputs]
for name in input_names:
res[name] = []
for time in t:
input_val = input(time)
for (name, val) in itertools.izip(input_names, input_val):
res[name].append(val)
# Create results for everything else
if blt:
# Iteration variables
i = 0
for var in model_states:
res[var.getName()] = y[:, i]
res[var.getMyDerivativeVariable().getName()] = yd[:, i]
i += 1
for var in model_algs:
res[var.getName()] = y[:, i]
i += 1
# Create function for computing solved algebraics
for (_, sol_alg) in model._explicit_solved_algebraics:
res[sol_alg.name] = []
alg_sol_f = casadi.MXFunction(model._known_vars + model._explicit_unsolved_vars, model._solved_expr)
alg_sol_f.init()
if expand_to_sx:
alg_sol_f = casadi.SXFunction(alg_sol_f)
alg_sol_f.init()
# Compute solved algebraics
for k in xrange(len(t)):
for (i, var) in enumerate(model._known_vars + model._explicit_unsolved_vars):
alg_sol_f.setInput(res[var.getName()][k], i)
alg_sol_f.evaluate()
for (j, sol_alg) in model._explicit_solved_algebraics:
res[sol_alg.name].append(alg_sol_f.getOutput(j).toScalar())
else:
res_vars = model_states + model_algs
for (i, var) in enumerate(res_vars):
res[var.getName()] = y[:, i]
der_var = var.getMyDerivativeVariable()
if der_var is not None:
res[der_var.getName()] = yd[:, i]
# Add results for all alias variables (only treat time-continuous variables) and convert to array
if blt:
res_model = model._model
else:
res_model = model
for var in res_model.getAllVariables():
if var.getVariability() == var.CONTINUOUS:
res[var.getName()] = np.array(res[var.getModelVariable().getName()])
res["time"] = np.array(res["time"])
res._blt_model = blt_model
return res
