def test_robust_regression(self):
#
# Compare cvxpy solutions to cvxopt ground truth
#
from cvxpy import (matrix, variable, program, minimize,
huber, sum, leq, geq, square, norm2,
ones, zeros, quad_form)
tol_exp = 5 # Check solution to within 5 decimal places
m, n = self.m, self.n
A = matrix(self.A)
b = matrix(self.b)
# Method 1: using huber directly
x = variable(n)
p = program(minimize(sum(huber(A*x - b, 1.0))))
p.solve(True)
np.testing.assert_array_almost_equal(
x.value, self.xh, tol_exp)
# Method 2: solving the dual QP
x = variable(n)
u = variable(m)
v = variable(m)
p = program(minimize(0.5*quad_form(u, 1.0) + sum(v)),
[geq(u, 0.0), leq(u, 1.0), geq(v, 0.0),
leq(A*x - b, u + v), geq(A*x - b, -u - v)])
p.solve(True)
np.testing.assert_array_almost_equal(
x.value, self.xh, tol_exp)
def create(m, n):
np.random.seed(0)
x0 = np.random.randn(n)
A = np.random.randn(m,n)
A = A*sp.diags([1 / np.sqrt(np.sum(A**2, 0))], [0])
b = A.dot(x0) + np.sqrt(0.01)*np.random.randn(m)
b = b + 10*np.asarray(sp.rand(m, 1, 0.05).todense()).ravel()
x = cp.Variable(n)
return cp.Problem(cp.Minimize(cp.sum_entries(cp.huber(A*x - b))))
### Part (b) ###
objective2 = cp.Minimize(cp.sum_squares(Y - X * beta - gamma * np.ones(n)))
constrains2 = [beta >= 0]
prob2 = cp.Problem(objective2, constrains2)
print('Part (b):')
print('Optimal value:', prob2.solve())
print('Optimal beta:', beta.value)
print('Optimal gamma:', gamma.value)
print()
### Part (c) ###
objective3 = cp.Minimize(
cp.sum_squares(Y - X * beta - gamma * np.ones(n)) + lam * cp.norm(beta, 1))
prob3 = cp.Problem(objective3)
print('Part (c):')
print('Optimal value:', prob3.solve())
print('Optimal beta:', beta.value)
print('Optimal gamma:', gamma.value)
print()
### Part (d) ###
objective4 = cp.Minimize(
cp.sum(cp.huber(Y - X * beta - gamma * np.ones(n), 0.5)))
prob4 = cp.Problem(objective4)
print('Part (d):')
print('Optimal value:', prob4.solve())
print('Optimal beta:', beta.value)
print('Optimal gamma:', gamma.value)
print()
def test_huber(self):
# Valid.
cp.huber(self.x, 1)
with self.assertRaises(Exception) as cm:
cp.huber(self.x, -1)
self.assertEqual(str(cm.exception),
"M must be a non-negative scalar constant.")
with self.assertRaises(Exception) as cm:
cp.huber(self.x, [1, 1])
self.assertEqual(str(cm.exception),
"M must be a non-negative scalar constant.")
# M parameter.
M = Parameter(nonneg=True)
# Valid.
cp.huber(self.x, M)
M.value = 1
self.assertAlmostEqual(cp.huber(2, M).value, 3)
# Invalid.
M = Parameter(nonpos=True)
with self.assertRaises(Exception) as cm:
cp.huber(self.x, M)
self.assertEqual(str(cm.exception),
"M must be a non-negative scalar constant.")
# Test copy with args=None
atom = cp.huber(self.x, 2)
copy = atom.copy()
self.assertTrue(type(copy) is type(atom))
# A new object is constructed, so copy.args == atom.args but copy.args
# is not atom.args.
self.assertEqual(copy.args, atom.args)
self.assertFalse(copy.args is atom.args)
# As get_data() returns a Constant, we have to check the value
self.assertEqual(copy.get_data()[0].value, atom.get_data()[0].value)
# Test copy with new args
copy = atom.copy(args=[self.y])
self.assertTrue(type(copy) is type(atom))
self.assertTrue(copy.args[0] is self.y)
self.assertEqual(copy.get_data()[0].value, atom.get_data()[0].value)
def loss(self, A, U): return cp.sum_entries(cp.huber(cp.Constant(A) - U, self.a)) def __str__(self): return "huber loss"
def loss(self, A, U):
return cp.sum_entries(cp.huber(cp.Constant(A) - U, self.a))
Constant([[5, 2], [3, 1]])), (lambda x: cp.cumsum(x, axis=1), (2, 2), [[[-5, 2], [-3, 1]]], Constant([[-5, 2], [-8, 3]])), (lambda x: cp.cumsum(x, axis=0), (2, 2), [[[-5, 2], [-3, 1]]], Constant([[-5, -3], [-3, -2]])), (lambda x: cp.cummax(x, axis=1), (2, 2), [[[-5, 2], [-3, 1]]], Constant([[-5, 2], [-3, 2]])), (lambda x: cp.cummax(x, axis=0), (2, 2), [[[-5, 2], [-3, 1]]], Constant([[-5, 2], [-3, 1]])), (cp.diag, (2,), [[[-5, 2], [-3, 1]]], Constant([-5, 1])), (cp.diag, (2, 2), [[-5, 1]], Constant([[-5, 0], [0, 1]])), (cp.exp, (2, 2), [[[1, 0], [2, -1]]], Constant([[math.e, 1], [math.e**2, 1.0 / math.e]])), (cp.huber, (2, 2), [[[0.5, -1.5], [4, 0]]], Constant([[0.25, 2], [7, 0]])), (lambda x: cp.huber(x, 2.5), (2, 2), [[[0.5, -1.5], [4, 0]]], Constant([[0.25, 2.25], [13.75, 0]])), (cp.inv_pos, (2, 2), [[[1, 2], [3, 4]]], Constant([[1, 1.0 / 2], [1.0 / 3, 1.0 / 4]])), (lambda x: (x + Constant(0))**-1, (2, 2), [[[1, 2], [3, 4]]], Constant([[1, 1.0 / 2], [1.0 / 3, 1.0 / 4]])), (cp.kl_div, tuple(), [math.e, 1], Constant([1])), (cp.kl_div, tuple(), [math.e, math.e], Constant([0])), (cp.kl_div, (2,), [[math.e, 1], 1], Constant([1, 0])), (cp.rel_entr, tuple(), [math.e, 1], Constant([math.e])), (cp.rel_entr, tuple(), [math.e, math.e], Constant([0])), (cp.rel_entr, (2,), [[math.e, 1], 1], Constant([math.e, 0])), (lambda x: cp.kron(np.array([[1, 2], [3, 4]]), x), (4, 4), [np.array([[5, 6], [7, 8]])], Constant(np.kron(np.array([[1, 2], [3, 4]]), np.array([[5, 6], [7, 8]])))), (cp.lambda_max, tuple(), [[[2, 0], [0, 1]]], Constant([2])), (cp.lambda_max, tuple(), [[[2, 0, 0], [0, 3, 0], [0, 0, 1]]], Constant([3])),
def loss(self, A, U): return cp.sum(cp.huber(cp.Constant(A) - U, self.a)) def __str__(self): return "huber loss"
def Classo_R2(pb, lam, compute=True):
pb_type = pb.type # can be 'Path-Alg', 'P-PDS' , 'PF-PDS' or 'DR'
(m, d, k), (A, C, y) = pb.dim, pb.matrix
lamb, rho = lam * pb.lambdamax, pb.rho
#ODE
# here we compute the path algo until our lambda, and just take the last beta
if (pb_type == 'Path-Alg'):
beta = solve_path((A, C, y), lam, False, rho, 'huber')[0]
return (beta[-1])
# 2 prox :
regpath = pb.regpath
r = lamb / (2 * rho)
if (pb_type == 'DR'):
Ahuber = np.concatenate((A, r * np.eye(len(A))), axis=1)
Chuber = np.concatenate((C, np.zeros((len(C), len(y)))), axis=1)
matrices_huber = (Ahuber, Chuber, y)
prob = problem_R1(matrices_huber, 'DR')
prob.regpath = regpath
if (len(pb.init) == 3): prob.init = pb.init
if not (regpath): return (Classo_R1(prob, lamb / prob.lambdamax)[:d])
x, warm_start = Classo_R1(prob, lamb / prob.lambdamax)
return (x[:d], warm_start)
tol = pb.tol * LA.norm(y) / LA.norm(A, 'fro') # tolerance rescaled
#cvx
# call to the cvx function of minimization
if (pb_type == 'cvx'):
import cvxpy as cp
x = cp.Variable(d)
objective, constraints = cp.Minimize(
cp.sum(cp.huber(A * x - y, rho)) +
lamb * cp.norm(x, 1)), [C * x == 0]
prob = cp.Problem(objective, constraints)
result = prob.solve(warm_start=regpath, eps_abs=tol)
if (regpath): return (x.value, True)
return (x.value)
if (compute): pb.compute_param()
tau, Proj, AtA, Aty = pb.tauN, proj_c(C, d), pb.AtA, pb.Aty
gamma = pb.gam / (2 * (pb.AtAnorm + r**2))
t = lamb * gamma
w, tm, zerom, zerod = t * pb.weights, t * np.ones(m), np.zeros(
m), np.zeros(d)
o, xbar, x, v = pb.init
# vectors usefull to compute the prox of f(b)= sum(wi |bi|)
#FORWARD BACKWARD
if (pb_type == 'P-PDS'):
for i in range(pb.N):
grad = (AtA.dot(x) - Aty)
v = v + tau * C.dot(xbar)
S = x - 2 * gamma * grad - 2 * gamma * r * (A.T).dot(o) - (
C.T).dot(v)
o = prox(
o * (1 - 2 * gamma * r**2) + 2 * gamma * r * (y - A.dot(x)),
tm, zerom)
p = prox(S, w, zerod)
nw_x = Proj.dot(p)
eps = nw_x - x
xbar = p + eps
if (i % 10 == 2 and LA.norm(eps) < tol): # 0.6
if (regpath): return (x, (o, xbar, x, v))
else: return (x)
x = nw_x
if (LA.norm(x) > 1e10):
print('DIVERGENCE')
return (x)
print('NO CONVERGENCE')
return (x)
# NO PROJ
if (pb_type == 'PF-PDS'): # y1 --> S ; p1 --> p . ; p2 --> y2
for i in range(pb.N):
grad = (AtA.dot(x) - Aty)
S1 = x - 2 * gamma * grad - 2 * gamma * r * (A.T).dot(o) - (
C.T).dot(v)
S2 = o * (1 - 2 * gamma * r**2) + 2 * gamma * r * (y - A.dot(x))
p1 = prox(S1, w, zerod)
p2 = prox(S2, tm, zerom)
v = v + tau * C.dot(p1)
v2 = v + tau * C.dot(x)
eps1 = p1 + 2 * gamma * (Aty - AtA.dot(p1) -
r * A.T.dot(o)) - C.T.dot(v2) - S1
eps2 = p2 + 2 * r * gamma * (y - r * p2 - A.dot(x)) - S2
x = x + eps1
o = o + eps2
if (i > 0 and LA.norm(eps1) + LA.norm(eps2) < tol):
if (regpath): return (x, (o, xbar, x, v))
else: return (x)
if (LA.norm(x) + LA.norm(o) + LA.norm(v) > 1e6):
print('DIVERGENCE')
return (x)
print('NO CONVERGENCE')
return (x)
linf(np_x[2 * k + 2, :], np_u[k + 1, :], x[2 * k + 2, :],
u[k + 1, :]))
]
const += [
x[2 * k + 2, :] - x[2 * k, :] ==
(linf(np_x[2 * k, :], np_u[k, :], x[2 * k, :], u[k, :]) +
4 * linf(np_x[2 * k + 1, :],
(np_u[k, :] + np_u[k + 1, :]) / 2, x[2 * k + 1],
(u[k, :] + u[k + 1, :]) / 2) +
linf(np_x[2 * k + 2, :], np_u[k + 1, :], x[2 * k + 2], u[k, :])) *
dt / 6
]
#const += [x[k+1,:] - x[k,:] == 0.5*dt*(linf(np_x[k+1,:],np_u[k+1,:],x[k+1,:],u[k+1,:])+linf(np_x[k,:],np_u[k,:],x[k,:],u[k,:]))]
cost = cost + 0.5 * cvx.huber(
x[k, 0] - np.pi, M=0.5) + 0.01 * cvx.huber(
u[k, :]) + 0.5 * cvx.huber(x[k, 1] - 0, M=0.25)
cost = cost + 100 * cvx.square(x[K, 0] -
np.pi) + 100 * cvx.square(x[K, 1] - 0)
objective = cvx.Minimize(cost)
print("Iteration", j + 1)
prob = cvx.Problem(objective, const)
sol = prob.solve(verbose=True)
np_x = x.value
np_u = u.value
#print((np_x))
# print(np.shape(np_u))
plt.plot(np_x[:, 0])
plt.plot(np_x[:, 1])
new_pos = new_pos.astype(int)
# true permutation matrix
P=np.identity(m)
P[perm_idxs]=P[new_pos,:]
true_perm=[]
for i in range(k):
if perm_idxs[i] != new_pos[i]:
true_perm = np.append(true_perm, perm_idxs[i])
y=P.dot(y_true)
new_pos = None
# naive estimator (P=I)
x_naive = np.linalg.lstsq(A,y)[0]
# robust estimator
x_hub = cp.Variable(n)
obj = cp.sum(cp.huber(A@x_hub-y)) #TODO dimention bugs
cp.Problem(cp.Minimize(obj)).solve()
plt.figure(1)
plt.plot(np.arange(m), np.abs(A.dot(x_hub.value)-y), '.')
plt.ylabel('residual')
plt.xlabel('idx')
plt.savefig('prob_152.png')
# remove k largest residuals
cand_idxs = np.zeros(m)
cand_idxs[:] = np.flip(np.argsort(np.abs(A.dot(x_hub.value)-y)))
cand_idxs = np.sort(cand_idxs[:k])
cand_idxs = cand_idxs.astype(int)
keep_idxs = np.zeros(m)
keep_idxs[:] = np.argsort(np.abs(A.dot(x_hub.value)-y).T)
keep_idxs = np.sort(keep_idxs[:(m-k)])
def main_optimize(elec_storage, battery_depth, thermal_storage, CHP_Runtime, CHP_Downtime, capacities_conv,
capacities_stor, tech_details, storage_details, list_techs, list_storage, time_now):
# Original boundary condition
hub_el = 'elec_' + str(hub)
hub_ht = 'heat_' + str(hub)
P_Demand = demand_data.loc[time_now: time_now + dt.timedelta(hours=23), hub_el].values.tolist()
Q_Demand = demand_data.loc[time_now: time_now + dt.timedelta(hours=23), hub_ht].values.tolist()
Ta = demand_data.loc[time_now: time_now + dt.timedelta(hours=23), 'temp'].values.tolist()
I_Solar = demand_data.loc[time_now: time_now + dt.timedelta(hours=23), 'solar_roof'].values.tolist()
Rini_CHP = CHP_Runtime
Dini_CHP = CHP_Downtime
power_list = []
heat_list = []
cost = 0
constr = []
Qmax_GSHP = 0
comp_list = []
it = 0
for item in list_techs:
if item == 'solar_PV':
Pmax_PV = math.ceil(capacities_conv.loc[((capacities_conv.hub == hub) & (
capacities_conv.tech == item)), 'value'].squeeze())
Eff_PV = math.ceil(tech_details.loc[(tech_details.tech == item), 'eff'].squeeze()) * 0.01
Pmin_PV = 0
comp_list.append(PV(num_opt_var))
power_list.append(comp_list[it].P_PV)
for t in range(num_opt_var):
comp_list[it].I_PV[t] = I_Solar[t] * (Pmax_PV / d_PV)
comp_list[it].TempEff_PV[t] = 1 + ((-beta) * ((Ta[t] - Tstc) +
(Tnoct - Ta[t]) * (I_Solar[t] / 0.8)))
for t in range(num_opt_var):
constr += [comp_list[it].P_PV[t] >= 0,
comp_list[it].P_PV[t] <= (comp_list[it].b_PV[t] * Eff_PV *
comp_list[it].TempEff_PV[t] * comp_list[it].I_PV[t])]
it = it + 1
elif item == 'solar_PVT':
Pmax_PVT = math.ceil(capacities_conv.loc[((capacities_conv.hub == hub) & (
capacities_conv.tech == item)), 'value'].squeeze())
Eff_PVT = tech_details.loc[(tech_details.tech == item), 'outshare'].squeeze()
list1 = Eff_PVT.split(",")
li1 = []
for t in list1:
li1.append(float(t))
PEff_PVT = li1[0]
QEff_PVT = li1[1]
Eff_PVT = tech_details.loc[(tech_details.tech == item), 'eff'].squeeze() * 0.01
comp_list.append(PVT(num_opt_var))
power_list.append(comp_list[it].P_PVT)
heat_list.append(comp_list[it].Q_PVT)
for t in range(num_opt_var):
comp_list[it].I_PVT[t] = I_Solar[t] * (Pmax_PVT / d_PVT)
comp_list[it].TempEff_PVT[t] = 1 + ((-beta) * ((Ta[t] - Tstc) +
(Tnoct - Ta[t]) * (I_Solar[t] / 0.8)))
for t in range(num_opt_var):
constr += [comp_list[it].P_PVT[t] >= 0, comp_list[it].Q_PVT[t] >= 0,
comp_list[it].Out_PVT[t] <= (comp_list[it].b_PVT[t] * Eff_PVT *
comp_list[it].TempEff_PVT[t] * comp_list[it].I_PVT[t]),
comp_list[it].P_PVT[t] == PEff_PVT * comp_list[it].Out_PVT[t],
comp_list[it].Q_PVT[t] == QEff_PVT * comp_list[it].Out_PVT[t]]
it = it + 1
elif item == 'Gas_CHP_unit_1':
Pmax_mCHP = math.ceil(capacities_conv.loc[((capacities_conv.hub == hub) & (
capacities_conv.tech == item)), 'value'].squeeze())
Pmin_mCHP = 0
Eff_mCHP = tech_details.loc[(tech_details.tech == item), 'outshare'].squeeze()
list2 = Eff_mCHP.split(",")
li = []
for k in list2:
li.append(float(k))
PEff_mCHP = li[0]
QEff_mCHP = li[1]
Eff_mCHP = tech_details.loc[(tech_details.tech == item), 'eff'].squeeze() * 0.01
comp_list.append(mCHP(num_opt_var))
power_list.append(comp_list[it].P_mCHP)
heat_list.append(comp_list[it].Q_mCHP)
for t in range(num_opt_var):
cost += comp_list[it].C_mCHP[t]
constr += [comp_list[it].Out_mCHP[t] >= comp_list[it].b_mCHP[t] * Pmin_mCHP,
comp_list[it].Out_mCHP[t] <= comp_list[it].b_mCHP[t] * Pmax_mCHP,
comp_list[it].P_mCHP[t] == PEff_mCHP * comp_list[it].Out_mCHP[t],
comp_list[it].Q_mCHP[t] == QEff_mCHP * comp_list[it].Out_mCHP[t],
comp_list[it].C_mCHP[t] == C_Fuel * comp_list[it].F_mCHP[t],
comp_list[it].F_mCHP[t] == comp_list[it].P_mCHP[t] / Eff_mCHP]
it = it + 1
elif item == 'Gas_CHP_unit_2':
Pmax_CHP = math.ceil(capacities_conv.loc[((capacities_conv.hub == hub) & (
capacities_conv.tech == item)), 'value'].squeeze())
Pmin_CHP = 0
Eff_CHP = tech_details.loc[(tech_details.tech == item), 'outshare'].squeeze()
list2 = Eff_CHP.split(",")
li = []
for m in list2:
li.append(float(m))
PEff_CHP = li[0]
QEff_CHP = li[1]
Eff_CHP = tech_details.loc[(tech_details.tech == item), 'eff'].squeeze() * 0.01
CHP_points = CHP_operation(Pmax_CHP, Eff_CHP, PEff_CHP, QEff_CHP, n_CHP, min_cap_CHP)
CHP_fuelcap = Pmax_CHP / Eff_CHP
P11_CHP = CHP_points[1][0]
P12_CHP = CHP_points[1][0]
P13_CHP = CHP_points[1][3]
P14_CHP = CHP_points[1][4]
P21_CHP = CHP_points[1][0]
P22_CHP = CHP_points[1][1]
P23_CHP = CHP_points[1][2]
P24_CHP = CHP_points[1][3]
Q11_CHP = CHP_points[0][0]
Q12_CHP = CHP_points[0][0]
Q13_CHP = CHP_points[0][3]
Q14_CHP = CHP_points[0][4]
Q21_CHP = CHP_points[0][0]
Q22_CHP = CHP_points[0][1]
Q23_CHP = CHP_points[0][2]
Q24_CHP = CHP_points[0][3]
comp_list.append(CHP(num_opt_var))
power_list.append(comp_list[it].P_CHP)
heat_list.append(comp_list[it].Q_CHP)
constr += [comp_list[it].ysum_CHP[0] == comp_list[it].yon_CHP[0],
comp_list[it].zsum_CHP[0] == comp_list[it].zoff_CHP[0],
comp_list[it].R_CHP[0] == (Rini_CHP + 1) * comp_list[it].b_CHP[0],
comp_list[it].D_CHP[0] == (Dini_CHP + 1) * (1 - comp_list[it].b_CHP[0])]
if Rini_CHP == 0:
constr += [comp_list[it].zoff_CHP[0] == 0, comp_list[it].yon_CHP[0] == comp_list[it].b_CHP[0]]
elif Dini_CHP == 0:
constr += [comp_list[it].zoff_CHP[0] == 1 - comp_list[it].b_CHP[0], comp_list[it].yon_CHP[0] == 0]
for t in range(num_opt_var):
cost += comp_list[it].C_CHP[t]
constr += [comp_list[it].P_CHP[t] <= Pmax_CHP * 10 * comp_list[it].b_CHP[t],
comp_list[it].Q_CHP[t] <= Pmax_CHP * 10 * comp_list[it].b_CHP[t],
comp_list[it].P_CHP[t] >= Pmin_CHP * comp_list[it].b_CHP[t],
comp_list[it].Q_CHP[t] >= Pmin_CHP * comp_list[it].b_CHP[t],
comp_list[it].P_CHP[t] == (
comp_list[it].w11_CHP[t] * P11_CHP + comp_list[it].w12_CHP[t] * P12_CHP +
comp_list[it].w13_CHP[t] * P13_CHP + comp_list[it].w14_CHP[t] * P14_CHP +
comp_list[it].w21_CHP[t] * P21_CHP + comp_list[it].w22_CHP[t] * P22_CHP +
comp_list[it].w23_CHP[t] * P23_CHP + comp_list[it].w24_CHP[t] * P24_CHP),
comp_list[it].Q_CHP[t] == (
comp_list[it].w11_CHP[t] * Q11_CHP + comp_list[it].w12_CHP[t] * Q12_CHP +
comp_list[it].w13_CHP[t] * Q13_CHP + comp_list[it].w14_CHP[t] * Q14_CHP +
comp_list[it].w21_CHP[t] * Q21_CHP + comp_list[it].w22_CHP[t] * Q22_CHP +
comp_list[it].w23_CHP[t] * Q23_CHP + comp_list[it].w24_CHP[t] * Q24_CHP),
comp_list[it].b1_CHP[t] + comp_list[it].b2_CHP[t] == comp_list[it].b_CHP[t],
comp_list[it].w11_CHP[t] + comp_list[it].w12_CHP[t] + comp_list[it].w13_CHP[t] +
comp_list[it].w14_CHP[t] == comp_list[it].b1_CHP[t],
comp_list[it].w21_CHP[t] + comp_list[it].w22_CHP[t] + comp_list[it].w23_CHP[t] +
comp_list[it].w24_CHP[t] == comp_list[it].b2_CHP[t],
comp_list[it].w11_CHP[t] >= 0, comp_list[it].w12_CHP[t] >= 0,
comp_list[it].w13_CHP[t] >= 0, comp_list[it].w14_CHP[t] >= 0,
comp_list[it].w21_CHP[t] >= 0, comp_list[it].w22_CHP[t] >= 0,
comp_list[it].w23_CHP[t] >= 0, comp_list[it].w24_CHP[t] >= 0,
comp_list[it].w11_CHP[t] <= 1, comp_list[it].w12_CHP[t] <= 1,
comp_list[it].w13_CHP[t] <= 1, comp_list[it].w14_CHP[t] <= 1,
comp_list[it].w21_CHP[t] <= 1, comp_list[it].w22_CHP[t] <= 1,
comp_list[it].w23_CHP[t] <= 1, comp_list[it].w24_CHP[t] <= 1,
comp_list[it].yon_CHP[t] + comp_list[it].zoff_CHP[t] <= 1,
comp_list[it].C_CHP[t] == C_Fuel * comp_list[it].F_CHP[t],
comp_list[it].F_CHP[t] == comp_list[it].P_CHP[t] / Eff_CHP]
if t >= 1:
constr += [comp_list[it].P_CHP[t] <= (comp_list[it].P_CHP[t - 1] + 0.5 * Pmax_CHP *
(comp_list[it].b_CHP[t - 1] + comp_list[it].yon_CHP[t])),
comp_list[it].P_CHP[t] >= (comp_list[it].P_CHP[t - 1] - 0.5 * Pmax_CHP *
(comp_list[it].b_CHP[t] + comp_list[it].zoff_CHP[t])),
comp_list[it].Q_CHP[t] <= (comp_list[it].Q_CHP[t - 1] + 0.5 * Pmax_CHP *
(comp_list[it].b_CHP[t - 1] + comp_list[it].yon_CHP[t])),
comp_list[it].Q_CHP[t] >= (comp_list[it].Q_CHP[t - 1] - 0.5 * Pmax_CHP *
(comp_list[it].b_CHP[t] + comp_list[it].zoff_CHP[t])),
comp_list[it].yon_CHP[t] - comp_list[it].zoff_CHP[t] == (comp_list[it].b_CHP[t] -
comp_list[it].b_CHP[t - 1])]
# min up time constraints
if UPmin_CHP > 0:
if t >= UPmin_CHP:
constr += [comp_list[it].b_CHP[t] >= comp_list[it].ysum_CHP[t],
comp_list[it].ysum_CHP[t] == (comp_list[it].ysum_CHP[t - 1] -
comp_list[it].yon_CHP[t - UPmin_CHP] +
comp_list[it].yon_CHP[t])]
elif 1 <= t < UPmin_CHP:
constr += [comp_list[it].b_CHP[t] >= comp_list[it].ysum_CHP[t],
comp_list[it].ysum_CHP[t] == (comp_list[it].ysum_CHP[t - 1] +
comp_list[it].yon_CHP[t])]
if 0 < Rini_CHP < UPmin_CHP:
if t < (UPmin_CHP - Rini_CHP):
constr += [comp_list[it].b_CHP[t] == 1,
comp_list[it].yon_CHP[t] == 0, comp_list[it].zoff_CHP[t] == 0]
elif Rini_CHP >= UPmin_CHP:
if t == 0:
constr += [comp_list[it].zoff_CHP[t] == 1 - comp_list[it].b_CHP[t],
comp_list[it].yon_CHP[t] == 0]
if DNmin_CHP > 0:
if t >= DNmin_CHP:
constr += [(1 - comp_list[it].b_CHP[t]) >= comp_list[it].zsum_CHP[t],
comp_list[it].zsum_CHP[t] == (comp_list[it].zsum_CHP[t - 1] -
comp_list[it].zoff_CHP[t - DNmin_CHP] +
comp_list[it].zoff_CHP[t])]
if 1 <= t < DNmin_CHP:
constr += [(1 - comp_list[it].b_CHP[t]) >= comp_list[it].zsum_CHP[t],
comp_list[it].zsum_CHP[t] == (comp_list[it].zsum_CHP[t - 1] +
comp_list[it].zoff_CHP[t])]
if 0 < Dini_CHP < DNmin_CHP:
if t < (DNmin_CHP - Dini_CHP):
constr += [comp_list[it].b_CHP[t] == 0, comp_list[it].yon_CHP[t] == 0,
comp_list[it].zoff_CHP[t] == 0]
elif Dini_CHP >= DNmin_CHP:
if t == 0:
constr += [comp_list[it].zoff_CHP[t] == 0,
comp_list[it].yon_CHP[t] == comp_list[it].b_CHP[t]]
it = it + 1
elif item == 'GSHP_1' or item == 'GSHP_2':
Qmax_GSHP = math.ceil(capacities_conv.loc[((capacities_conv.hub == hub) &
(capacities_conv.tech == item)), 'value'].squeeze())
Qmin_GSHP = 0
comp_list.append(GSHP(num_opt_var))
power_list.append(-comp_list[it].P_GSHP)
heat_list.append(comp_list[it].Q_GSHP)
for t in range(num_opt_var):
constr += [comp_list[it].Q_GSHP[t] <= comp_list[it].b_GSHP[t] * Qmax_GSHP,
comp_list[it].Q_GSHP[t] >= comp_list[it].b_GSHP[t] * Qmin_GSHP,
comp_list[it].Q_GSHP[t] == comp_list[it].P_GSHP[t] * COP]
it = it + 1
elif item == 'gas_boiler_1' or item == 'gas_boiler_2':
Qmax_GB = math.ceil(capacities_conv.loc[((capacities_conv.hub == hub) &
(capacities_conv.tech == item)), 'value'].squeeze())
Qmin_GB = 0
x0_GB = 0
x1_GB = 0.25 * Qmax_GB
x2_GB = 0.5 * Qmax_GB
x3_GB = 0.75 * Qmax_GB
x4_GB = Qmax_GB
Eff1_GB = 0.01 * (21.49 + (182.18 * (1 / 4)) + ((-120.67) * (1 / 4) ** 2))
Eff2_GB = 0.01 * (21.49 + (182.18 * (2 / 4)) + ((-120.67) * (2 / 4) ** 2))
Eff3_GB = 0.01 * (21.49 + (182.18 * (3 / 4)) + ((-120.67) * (3 / 4) ** 2))
Eff4_GB = 0.01 * (21.49 + (182.18 * (4 / 4)) + ((-120.67) * (4 / 4) ** 2))
comp_list.append(GB(num_opt_var))
heat_list.append(comp_list[it].Q_GB)
for t in range(num_opt_var):
cost += comp_list[it].C_GB[t]
constr += [comp_list[it].Q_GB[t] <= comp_list[it].b_GB[t] * Qmax_GB,
comp_list[it].Q_GB[t] >= comp_list[it].b_GB[t] * Qmin_GB,
comp_list[it].Q_GB[t] == (comp_list[it].w1_GB[t] * x1_GB +
comp_list[it].w2_GB[t] * x2_GB +
comp_list[it].w3_GB[t] * x3_GB +
comp_list[it].w4_GB[t] * x4_GB),
(comp_list[it].b1_GB[t] + comp_list[it].b2_GB[t] +
comp_list[it].b3_GB[t] + comp_list[it].b4_GB[t]) == comp_list[it].b_GB[t],
(comp_list[it].w0_GB[t] + comp_list[it].w1_GB[t] + comp_list[it].w2_GB[t] +
comp_list[it].w3_GB[t] + comp_list[it].w4_GB[t]) == comp_list[it].b_GB[t],
comp_list[it].w0_GB[t] <= comp_list[it].b1_GB[t],
comp_list[it].w1_GB[t] <= comp_list[it].b1_GB[t] + comp_list[it].b2_GB[t],
comp_list[it].w2_GB[t] <= comp_list[it].b3_GB[t] + comp_list[it].b2_GB[t],
comp_list[it].w3_GB[t] <= comp_list[it].b3_GB[t] + comp_list[it].b4_GB[t],
comp_list[it].w4_GB[t] <= comp_list[it].b4_GB[t],
comp_list[it].w0_GB[t] >= 0, comp_list[it].w1_GB[t] >= 0, comp_list[it].w2_GB[t] >= 0,
comp_list[it].w3_GB[t] >= 0, comp_list[it].w4_GB[t] >= 0,
comp_list[it].w0_GB[t] <= 1, comp_list[it].w1_GB[t] <= 1, comp_list[it].w2_GB[t] <= 1,
comp_list[it].w3_GB[t] <= 1, comp_list[it].w4_GB[t] <= 1,
comp_list[it].C_GB[t] == C_Fuel * comp_list[it].F_GB[t],
comp_list[it].F_GB[t] == (comp_list[it].w1_GB[t] * (x1_GB / Eff1_GB) +
comp_list[it].w2_GB[t] * (x2_GB / Eff2_GB) +
comp_list[it].w3_GB[t] * (x3_GB / Eff3_GB) +
comp_list[it].w4_GB[t] * (x4_GB / Eff4_GB))]
it = it + 1
for item in list_storage:
if item == 'heat_storage':
Qmax_Storage = capacities_stor.loc[((capacities_stor.hub == hub) &
(capacities_stor.tech == item)), 'value'].squeeze()
Eff_Storage = storage_details.loc[(storage_details.tech == item), 'stateff'].squeeze()
Eff_StorageCh = storage_details.loc[(storage_details.tech == item), 'cyceff'].squeeze()
Eff_StorageDc = Eff_StorageCh
comp_list.append(Heat_Storage(num_opt_var))
constr += [comp_list[it].Q_StorageTot[0] == (Eff_Storage * thermal_storage +
Eff_StorageCh * comp_list[it].Q_StorageCh[0] -
Eff_StorageDc * comp_list[it].Q_StorageDc[0])]
heat_list.append(comp_list[it].Q_StorageDc)
heat_list.append(-comp_list[it].Q_StorageCh)
for t in range(num_opt_var):
constr += [comp_list[it].Q_StorageCh[t] >= 0,
comp_list[it].Q_StorageCh[t] <= BigM * comp_list[it].b_StorageCh[t],
comp_list[it].Q_StorageDc[t] >= 0,
comp_list[it].Q_StorageDc[t] <= BigM * comp_list[it].b_StorageDc[t],
comp_list[it].Q_StorageTot[t] >= 0.2 * Qmax_Storage,
comp_list[it].Q_StorageTot[t] <= Qmax_Storage,
comp_list[it].b_StorageCh[t] + comp_list[it].b_StorageDc[t] == 1,
comp_list[it].b_StorageCh[t] >= 0, comp_list[it].b_StorageDc[t] >= 0,
comp_list[it].b_StorageCh[t] <= 1, comp_list[it].b_StorageDc[t] <= 1]
if t >= 1:
constr += [comp_list[it].Q_StorageTot[t] == (Eff_Storage * comp_list[it].Q_StorageTot[t - 1] +
Eff_StorageCh * comp_list[it].Q_StorageCh[t] -
Eff_StorageDc * comp_list[it].Q_StorageDc[t])]
it = it + 1
elif item == 'Battery':
Pmax_Battery = capacities_stor.loc[((capacities_stor.hub == hub) &
(capacities_stor.tech == item)), 'value'].squeeze()
Pmin_Battery = 0
Eff_Battery = storage_details.loc[(storage_details.tech == item), 'stateff'].squeeze()
Eff_BatteryCh = storage_details.loc[(storage_details.tech == item), 'cyceff'].squeeze()
Eff_BatteryDc = Eff_BatteryCh
c1_Battery = 0.11
c2_Battery = 0.32
c3_Battery = 0.55
c4_Battery = 0.76
comp_list.append(Elec_Storage(num_opt_var))
power_list.append(comp_list[it].P_BatteryDc)
power_list.append(-comp_list[it].P_BatteryCh)
constr += [comp_list[it].P_BatteryTot[0] == (Eff_Battery * elec_storage +
Eff_BatteryCh * comp_list[it].P_BatteryCh[0] -
(1 / Eff_BatteryDc) * comp_list[it].P_BatteryDc[0]),
comp_list[it].w1_Battery[0] == (battery_depth[0] + comp_list[it].p1_BatteryCh[0] -
comp_list[it].p1_BatteryDc[0]),
comp_list[it].w2_Battery[0] == (battery_depth[1] + comp_list[it].p2_BatteryCh[0] -
comp_list[it].p2_BatteryDc[0]),
comp_list[it].w3_Battery[0] == (battery_depth[2] + comp_list[it].p3_BatteryCh[0] -
comp_list[it].p3_BatteryDc[0]),
comp_list[it].w4_Battery[0] == (battery_depth[3] + comp_list[it].p4_BatteryCh[0] -
comp_list[it].p4_BatteryDc[0])]
for t in range(num_opt_var):
cost += comp_list[it].C_Battery[t]
constr += [comp_list[it].P_BatteryTot[t] >= 0.2 * Pmax_Battery,
comp_list[it].P_BatteryTot[t] <= 0.8 * Pmax_Battery,
comp_list[it].P_BatteryCh[t] >= 0,
comp_list[it].P_BatteryCh[t] <= BigM * comp_list[it].b_BatteryCh[t],
comp_list[it].P_BatteryDc[t] >= 0,
comp_list[it].P_BatteryDc[t] <= BigM * comp_list[it].b_BatteryDc[t],
comp_list[it].b_BatteryCh[t] + comp_list[it].b_BatteryDc[t] == 1,
comp_list[it].w1_Battery[t] <= 0.2, comp_list[it].w2_Battery[t] <= 0.2,
comp_list[it].w3_Battery[t] <= 0.2, comp_list[it].w4_Battery[t] <= 0.2,
comp_list[it].w1_Battery[t] >= 0, comp_list[it].w2_Battery[t] >= 0,
comp_list[it].w3_Battery[t] >= 0, comp_list[it].w4_Battery[t] >= 0,
comp_list[it].p1_BatteryCh[t] >= 0, comp_list[it].p2_BatteryCh[t] >= 0,
comp_list[it].p3_BatteryCh[t] >= 0, comp_list[it].p4_BatteryCh[t] >= 0,
comp_list[it].p1_BatteryDc[t] >= 0, comp_list[it].p2_BatteryDc[t] >= 0,
comp_list[it].p3_BatteryDc[t] >= 0, comp_list[it].p4_BatteryDc[t] >= 0,
comp_list[it].p1_BatteryCh[t] <= 0.2 - comp_list[it].w1_Battery[t],
comp_list[it].p2_BatteryCh[t] <= 0.2 - comp_list[it].w2_Battery[t],
comp_list[it].p3_BatteryCh[t] <= 0.2 - comp_list[it].w3_Battery[t],
comp_list[it].p4_BatteryCh[t] <= 0.2 - comp_list[it].w4_Battery[t],
comp_list[it].p1_BatteryDc[t] <= comp_list[it].w1_Battery[t],
comp_list[it].p2_BatteryDc[t] <= comp_list[it].w2_Battery[t],
comp_list[it].p3_BatteryDc[t] <= comp_list[it].w3_Battery[t],
comp_list[it].p4_BatteryDc[t] <= comp_list[it].w4_Battery[t],
(1 / Eff_BatteryDc) * comp_list[it].P_BatteryDc[t] == ((comp_list[it].p1_BatteryDc[t] +
comp_list[it].p2_BatteryDc[t] +
comp_list[it].p3_BatteryDc[t] +
comp_list[it].p4_BatteryDc[t])
* Pmax_Battery),
Eff_BatteryCh * comp_list[it].P_BatteryCh[t] == ((comp_list[it].p1_BatteryCh[t] +
comp_list[it].p2_BatteryCh[t] +
comp_list[it].p3_BatteryCh[t] +
comp_list[it].p4_BatteryCh[t])
* Pmax_Battery),
comp_list[it].C_Battery[t] == ((c1_Battery * (comp_list[it].p1_BatteryCh[t] +
comp_list[it].p1_BatteryDc[t]) +
c2_Battery * (comp_list[it].p2_BatteryCh[t] +
comp_list[it].p2_BatteryDc[t]) +
c3_Battery * (comp_list[it].p3_BatteryCh[t] +
comp_list[it].p3_BatteryDc[t]) +
c4_Battery * (comp_list[it].p4_BatteryCh[t] +
comp_list[it].p4_BatteryDc[t]))
* Pmax_Battery)]
if t >= 1:
constr += [
comp_list[it].P_BatteryTot[t] == (Eff_Battery * comp_list[it].P_BatteryTot[t - 1] +
Eff_BatteryCh * comp_list[it].P_BatteryCh[t] -
(1 / Eff_BatteryDc) * comp_list[it].P_BatteryDc[t]),
comp_list[it].w1_Battery[t] == (comp_list[it].w1_Battery[t - 1] +
comp_list[it].p1_BatteryCh[t] -
comp_list[it].p1_BatteryDc[t]),
comp_list[it].w2_Battery[t] == (comp_list[it].w2_Battery[t - 1] +
comp_list[it].p2_BatteryCh[t] -
comp_list[it].p2_BatteryDc[t]),
comp_list[it].w3_Battery[t] == (comp_list[it].w3_Battery[t - 1] +
comp_list[it].p3_BatteryCh[t] -
comp_list[it].p3_BatteryDc[t]),
comp_list[it].w4_Battery[t] == (comp_list[it].w4_Battery[t - 1] +
comp_list[it].p4_BatteryCh[t] -
comp_list[it].p4_BatteryDc[t])]
it = it + 1
R_GridOut = demand_data.loc[time_now: time_now + dt.timedelta(hours=23), 'el_tariff'].values.tolist()
R_GridIn = demand_data.loc[time_now: time_now + dt.timedelta(hours=23), 'feed_in_tariff'].values.tolist()
comp_list.append(Elec_Grid(num_opt_var))
if len(power_list) < 24:
t = len(power_list)
while t <= 23:
power_list.append([0] * num_opt_var)
t = t + 1
if len(heat_list) < 24:
t = len(heat_list)
while t <= 23:
heat_list.append([0] * num_opt_var)
t = t + 1
for t in range(num_opt_var):
# Demand
cost += comp_list[it].C_Grid[t] + (cp.huber(cp.norm(comp_list[it].P_Slack[t]), P_delta) +
cp.huber(cp.norm(comp_list[it].Q_Slack[t]), Q_delta))
constr += [P_Demand[t] == (power_list[0][t] + power_list[1][t] + power_list[2][t] + power_list[3][t] +
power_list[4][t] + power_list[5][t] + power_list[6][t] + power_list[7][t] +
power_list[8][t] + power_list[9][t] + power_list[10][t] + power_list[11][t] +
power_list[12][t] + power_list[13][t] + power_list[14][t] + power_list[15][t] +
power_list[16][t] + power_list[17][t] + power_list[18][t] + power_list[19][t] +
power_list[20][t] + power_list[21][t] + power_list[22][t] + power_list[23][t] +
comp_list[it].P_GridOut[t] - comp_list[it].P_GridIn[t]),
Q_Demand[t] == (heat_list[0][t] + heat_list[1][t] + heat_list[2][t] + heat_list[3][t] +
heat_list[4][t] + heat_list[5][t] + heat_list[6][t] + heat_list[7][t] +
heat_list[8][t] + heat_list[9][t] + heat_list[10][t] + heat_list[11][t] +
heat_list[12][t] + heat_list[13][t] + heat_list[14][t] + heat_list[15][t] +
heat_list[16][t] + heat_list[17][t] + heat_list[18][t] + heat_list[19][t] +
heat_list[20][t] + heat_list[21][t] + heat_list[22][t] + heat_list[23][t]),
comp_list[it].b_GridIn[t] + comp_list[it].b_GridOut[t] <= 1,
comp_list[it].P_GridIn[t] >= 0, comp_list[it].P_GridIn[t] <= BigM * comp_list[it].b_GridIn[t],
comp_list[it].P_GridOut[t] >= 0, comp_list[it].P_GridOut[t] <= BigM * comp_list[it].b_GridOut[t],
comp_list[it].C_Grid[t] == (R_GridOut[t] * comp_list[it].P_GridOut[t] -
R_GridIn[t] * comp_list[it].P_GridIn[t]),
comp_list[it].P_Slack[t] >= 0, comp_list[it].Q_Slack[t] >= 0]
# Solve with mosek or Gurobi
problem = cp.Problem(cp.Minimize(cost), constr)
problem.solve(solver=cp.MOSEK, verbose=True, save_file='opt_diagnosis.opf',
mosek_params={mosek.iparam.intpnt_solve_form: mosek.solveform.dual,
mosek.dparam.optimizer_max_time: 100.0})
opt_stat = problem.status
opt_val = problem.value
opt_time = problem.solver_stats
print(f"Status:{opt_stat}, with Value:{opt_val:.2f}")
power = []
heat = []
grid_in = []
grid_out = []
storage_heatin = []
storage_heatout = []
storage_elecin = []
storage_elecout = []
tech_cost = 0
storage_cost = 0
grid_cost = 0
final_cost = 0
storage_elec = 0
storage_heat = 0
batt_depth = [0, 0, 0, 0]
batt_in = []
batt_out = []
CHP_ontime = 0
CHP_offtime = 0
it = 0
for item in list_techs:
if item == 'solar_PV':
power.append(comp_list[it].P_PV.value[0])
heat.append(0)
it = it + 1
if item == 'solar_PVT':
power.append(comp_list[it].P_PVT.value[0])
heat.append(comp_list[it].Q_PVT.value[0])
it = it + 1
if item == 'Gas_CHP_unit_1':
power.append(comp_list[it].P_mCHP.value[0])
heat.append(comp_list[it].Q_mCHP.value[0])
tech_cost += comp_list[it].C_mCHP.value[0]
final_cost += comp_list[it].C_mCHP.value[0]
it = it + 1
if item == 'Gas_CHP_unit_2':
power.append(comp_list[it].P_CHP.value[0])
heat.append(comp_list[it].Q_CHP.value[0])
tech_cost += comp_list[it].C_CHP.value[0]
CHP_ontime = comp_list[it].R_CHP.value[0]
CHP_offtime = comp_list[it].D_CHP.value[0]
final_cost += comp_list[it].C_CHP.value[0]
it = it + 1
if item == 'GSHP_1' or item == 'GSHP_2':
power.append(-comp_list[it].P_GSHP.value[0])
heat.append(comp_list[it].Q_GSHP.value[0])
it = it + 1
if item == 'gas_boiler_1' or item == 'gas_boiler_2':
power.append(0)
heat.append(comp_list[it].Q_GB.value[0])
tech_cost += comp_list[it].C_GB.value[0]
final_cost += comp_list[it].C_GB.value[0]
it = it + 1
for item in list_storage:
if item == 'heat_storage':
storage_heatin.append(-comp_list[it].Q_StorageCh.value[0])
storage_heatout.append(comp_list[it].Q_StorageDc.value[0])
storage_heat = comp_list[it].Q_StorageTot.value[0]
it = it + 1
if item == 'Battery':
storage_elecin.append(-comp_list[it].P_BatteryCh.value[0])
storage_elecout.append(comp_list[it].P_BatteryDc.value[0])
storage_elec = comp_list[it].P_BatteryTot.value[0]
storage_cost += comp_list[it].C_Battery.value[0]
batt_depth = [comp_list[it].w1_Battery.value[0], comp_list[it].w2_Battery.value[0],
comp_list[it].w3_Battery.value[0], comp_list[it].w4_Battery.value[0]]
batt_in = [comp_list[it].p1_BatteryCh.value[0], comp_list[it].p2_BatteryCh.value[0],
comp_list[it].p3_BatteryCh.value[0], comp_list[it].p4_BatteryCh.value[0]]
batt_out = [comp_list[it].p1_BatteryDc.value[0], comp_list[it].p2_BatteryDc.value[0],
comp_list[it].p3_BatteryDc.value[0], comp_list[it].p4_BatteryDc.value[0]]
final_cost += comp_list[it].C_Battery.value[0]
it = it + 1
grid_cost = [comp_list[it].C_Grid.value[0]]
final_cost += comp_list[it].C_Grid.value[0]
grid_in.append(-comp_list[it].P_GridIn.value[0])
grid_out.append(comp_list[it].P_GridOut.value[0])
demand_power = [P_Demand[0]]
demand_heat = [Q_Demand[0]]
final_cost = [final_cost]
tech_cost = [tech_cost]
storage_cost = [storage_cost]
print('run complete')
return demand_power, demand_heat, power, heat, grid_in, grid_out, storage_heatin, storage_heatout, storage_elecin, storage_elecout, tech_cost, storage_cost, grid_cost, final_cost, storage_heat, storage_elec, batt_depth, CHP_ontime, CHP_offtime
def huber_loss_fn(residuals, params):
M = params
return np.sum([i for i in cp.huber(residuals, M)])
def huber_loss_fn(Kmat, Ymat, alpha_coef):
return cp.sum(cp.huber(Kmat.T * alpha_coef - Ymat, 1))
(u + next_u) / 2) +
linf(np_x[2 * i + 2, :], np_u[i + 1, :], next_x, next_u)) *
dt / 6)
#constraints.append(deldthetan == deldtheta + lin(at, deltheta) * dt)
#conditions on allowable control
constraints.append(next_u <= 8.0)
constraints.append(next_u >= -8.0)
#trust regions
#constraints.append(delthetan <= 0.5)
#constraints.append(delthetan >= -0.5)
#constraints.append(deldthetan <= 0.5)
#constraints.append(deldthetan >= -0.5)
#Cost calculation
cost = cost + cvx.huber(x[0] - np.pi, M=0.5) + 0.01 * cvx.huber(
u
) #+ (np.cos(np_x[2*i,:]) + 1) * (x[0] - np_x[2*i,:]) #+ cvx.square( x[0] - np.pi ) #+ cvx.square(u) #+ 0.1 * cvx.square(ut)
# + cvx.square(np.cos(np_x[2*i,:])*(x - np_x[2*i,:]))
x = next_x
u = next_u
cost = cost + 100 * cvx.square(
x[0] - np.pi) # cvx.huber( x[0] - np.pi, M=0.4)
objective = cvx.Minimize(cost)
#print(objective)
#print(constraints)
prob = cvx.Problem(objective, constraints)
sol = prob.solve(verbose=True)
#print(sol)
#update by the del
PTl[np.asmatrix(pid).T, pid] = PTs
#[pid,pid], [[pid],pid] don't work!!!!!!
#http://stackoverflow.com/questions/4257394/slicing-of-a-numpy-2d-array-or-how-do-i-extract-an-mxm-submatrix-from-an-nxn-ar
PTp.value = PTl * PTp.value
det = np.abs(np.linalg.det(PTp.value))
#print('det', det)
if det != 1:
input('...det not 1')
#
PTp = cv.Parameter(ct.m, ct.m)
PTp.value = np.asmatrix(np.eye(ct.m))
prbr = cv.Problem(cv.Minimize(\
cv.sum_entries(cv.huber(ct.A * xvr - PTp * ct.y))))
prb = cv.Problem(cv.Minimize(cv.norm(ct.A * xvr - PTp * ct.y)))
bestres = np.float("inf")
bestxer = np.float("inf")
bestidx = np.float("inf")
bestP = np.float("inf")
#
pl.figure(1)
for idx in np.arange(8):
#rough fit
prbr.solve()
#rearrange
construct_PT(np, ct, xvr, PTp)
#tight fit
prb.solve()
if prb.status != 'optimal':
def method1():
'''
First, try and fine the TF for vent at a few good points, then used that to find the scaling
at other points. Works alright, not the best though.
'''
st = 500
en = 1000
T = en - st
asm = a[st:en]
valsm = val[st:en]
Vtf = cvx.Variable(Vtf_end - Vtf_st)
Val = cvx.Variable(T)
g = 0
for i in range(-Vtf_st, T - Vtf_end):
if i + Vtf_st == 0:
g += valsm[i] * cvx.vstack(Vtf, np.zeros(T - Vtf_end + Vtf_st))
else:
g += valsm[i] * cvx.vstack(np.zeros(
(i + Vtf_st, 1)), Vtf, np.zeros((T - Vtf_end - i, 1)))
const = [Val == g]
obj = cvx.Minimize(cvx.sum_entries(cvx.huber(asm + Val, 0.2)))
prob = cvx.Problem(obj, const)
prob.solve(solver='ECOS',
verbose=True,
reltol=1e-10,
abstol=1e-10,
max_iters=200)
#plt.plot(Vtf.value)
#plt.show()
Vtf = Vtf.value
T = a.size
print(T, bal.size)
Bal = cvx.Variable(T)
Val = cvx.Variable(T)
Btf = cvx.Variable(Btf_end - Btf_st)
Vs = cvx.Variable(T)
const = []
f = 0
for i in range(-Btf_st, T - Btf_end):
if i + Btf_st == 0:
f += bal[i] * cvx.vstack(Btf, np.zeros(T - Btf_end + Btf_st))
else:
f += bal[i] * cvx.vstack(np.zeros(
(i + Btf_st, 1)), Btf, np.zeros((T - Btf_end - i, 1)))
const.append(Bal == f)
const.append(Btf >= 0)
const.append(cvx.sum_entries(Btf) == 400)
g = 0
for i in range(-Vtf_st, T - Vtf_end):
if i + Vtf_st == 0:
g += Vs[i] * val[i] * cvx.vstack(Vtf,
np.zeros(T - Vtf_end + Vtf_st))
else:
g += Vs[i] * val[i] * cvx.vstack(np.zeros(
(i + Vtf_st, 1)), Vtf, np.zeros((T - Vtf_end - i, 1)))
const.append(Val == g)
const.append(Vs >= 0)
#const.append(cvx.sum_entries(Vtf) == 400)
obj = cvx.Minimize(cvx.sum_entries(cvx.huber(a - Bal + Val, 0.2)))
prob = cvx.Problem(obj, const)
print("OHP")
r = prob.solve(solver='ECOS',
verbose=True,
reltol=1e-10,
abstol=1e-10,
max_iters=200)
print(r)
plt.subplot(411)
plt.plot(bal)
plt.plot(val)
plt.subplot(412)
plt.plot(a)
plt.plot(a + np.array((-Bal.value + Val.value))[:, 0])
plt.subplot(413)
plt.plot(Bal.value, 'r--')
plt.subplot(414)
plt.plot(Btf.value)
#plt.plot(V.value)
plt.show()
1
import scipy.sparse as sps
np.random.seed(0)
m = 5000
n = 200
x0 = np.random.randn(n)
A = np.random.randn(m, n)
A = A * sps.diags([1 / np.sqrt(np.sum(A**2, 0))], [0])
b = A.dot(x0) + np.sqrt(0.01) * np.random.randn(m)
b = b + 10 * np.asarray(sps.rand(m, 1, 0.05).todense()).ravel()
# Problem construction
x = cp.Variable(n)
prob = cp.Problem(cp.Minimize(cp.sum_entries(cp.huber(A * x - b))))
# Problem collection
# Single problem collection
problemDict = {"problemID": problemID, "problem": prob, "opt_val": opt_val}
problems = [problemDict]
# For debugging individual problems:
if __name__ == "__main__":
def printResults(problemID="", problem=None, opt_val=None):
print(problemID)
problem.solve()
print("\tstatus: {}".format(problem.status))
print("\toptimal value: {}".format(problem.value))