def set_initial_gains(SS, K0_method, K0=None):
# Initial gains
if K0_method == 'random':
K0 = randn(*SS.Kare.shape)
SS.setK(K0)
while not SS.c < np.inf:
K0 = randn(*SS.Kare.shape) # May take forever
SS.setK(K0)
elif K0_method == 'random_olmss':
K0 = randn(*SS.Kare.shape)
SS.setK(K0)
while not SS.c < np.inf:
K0 = 0.9 * K0 # Only guaranteed to work if sys open loop mean-square stable
SS.setK(K0)
elif K0_method == 'are':
K0 = SS.Kare
print('Initializing at the ARE solution')
elif K0_method == 'are_perturbed':
perturb_scale = 10.0 / SS.n # scale with 1/n as rough heuristic
safety_scale = 10.0
Kp = randn(SS.Kare.shape[0], SS.Kare.shape[1])
K0 = SS.Kare + perturb_scale * Kp
SS.setK(K0)
while not SS.c < safety_scale * SS.ccare:
perturb_scale *= 0.2
K0 = SS.Kare + perturb_scale * Kp
SS.setK(K0)
while not SS.c > safety_scale * SS.ccare:
perturb_scale *= 1.01
K0 = SS.Kare + perturb_scale * Kp
SS.setK(K0)
perturb_scale /= 1.01
K0 = SS.Kare + perturb_scale * Kp
SS.setK(K0)
elif K0_method == 'zero':
K0 = np.zeros([SS.m, SS.n])
SS.setK(K0)
P = dlyap_obj(SS, algo='iterative', show_warn=False)
if P is not None:
print('Initializing with zero gain solution')
else:
print(
'System not open-loop mean-square stable, use a different initial gain setting'
)
elif K0_method == 'user':
K0 = K0
SS.setK(K0)
return K0
def set_initial_gains(SS, K0_method):
# Initial gains
if K0_method == 'random':
K0 = randn(SS.Kare.shape)
SS.setK(K0)
while not SS.c < np.inf:
K0 = 0.9 * K0 # Only guaranteed to work if sys open loop mean-square stable
SS.setK(K0)
if K0_method == 'are':
K0 = SS.Kare
print('Initializing at the ARE solution')
elif K0_method == 'perturbed_are_safe':
perturb_scale = float(10) / float(
SS.n) # scale with 1/n as rough heuristic
safety_scale = float(10)
Kp = randn(SS.Kare.shape[0], SS.Kare.shape[1])
K0 = SS.Kare + perturb_scale * Kp
SS.setK(K0)
while not SS.c < safety_scale * SS.ccare:
perturb_scale *= 0.5
K0 = SS.Kare + perturb_scale * Kp
elif K0_method == 'zero':
K0 = np.zeros([SS.m, SS.n])
SS.setK(K0)
P = dlyap_obj(SS, algo='iterative', show_warn=False)
if P is not None:
print('Initializing with zero gain solution')
else:
print(
'System not open-loop mean-square stable, use a different initial gain setting'
)
elif K0_method == 'user':
K0[K1_subs] = 7.2
K0[K2_subs] = -0.37
SS.setK(K0)
def sample_Brand(self):
self._Brand = np.copy(self.B)
for j in range(self.q):
self._Brand += (self.b[j]**0.5) * randn() * self.Bb[:, :, j]
return self._Brand
def sample_Arand(self):
self._Arand = np.copy(self.A)
for i in range(self.p):
self._Arand += (self.a[i]**0.5) * randn() * self.Aa[:, :, i]
return self._Arand