else:
TS = m.addConstrs(inflow_now[i] + inflow_past[i] == 0
for i in range(4))
m.add_continuous_uncertainty(
uncertainty=sampler(t - 1),
locations=([(TS[i], inflow_past[i])
for i in range(4)] + [TS[i] for i in range(4)]),
)
if t < T / 2:
z = m.addVars(4, vtype='B')
m.addConstrs(hydro[i] >= z[i] * hydro_['UB'][:4][i] / 2
for i in range(4))
m.addConstrs(thermal_sum[i] >= (1 - z[i]) * thermal_mid_level[i]
for i in range(4))
HydroThermal.discretize(n_samples=100, random_state=seed)
if solver == 'Extensive':
HT_extensive = Extensive(HydroThermal)
HT_extensive.solve(outputFlag=0, TimeLimit=time_limit)
print('lower bound', HT_extensive.objBound)
print('upper bound', HT_extensive.objVal)
print('total time', HT_extensive.total_time)
print('gap', HT_extensive.MIPGap)
if solver == 'SDDiP':
HydroThermal.binarize(bin_stage=T)
HT_sddp = SDDiP(HydroThermal)
HT_sddp.solve(
logFile=0,
logToConsole=0,
cuts=[cut],
n_processes=1,
for i in I for j in J)
# capacity constraint
m.addConstrs(
gurobipy.quicksum(B_now[(i, j)] - C_now[(i, j)] for i in I
for j in J
if l in i and j.startswith(k)) <= R[k]
for k in K for l in L)
m.addConstrs(
(b[(i, j)] <= 0 for i in I for j in J),
uncertainty_dependent=[
6 * i + j for i in range(len(I)) for j in range(len(J))
])
start_mca = time.time()
markovian = airline.discretize(
method='SA',
n_Markov_states=n_Markov_states,
n_sample_paths=n_sample_paths,
int_flag=1,
)
time_mca = time.time() - start_mca
mca['model'].append(idx)
mca['iter_MCA'].append(n_sample_paths)
mca['time'].append(time_mca)
for t in range(1, 15):
mca[t].append(markovian.n_Markov_states[t - 1])
# comparison visually of the true process and its Markov chain approximation
if idx == 2:
simulation = markovian.simulate(1000)
true = demand_sampler(numpy.random.RandomState(0), 1000)
fig_mca, ax_mca = plt.subplots(6,
2,
figsize=(10, 10),