def simulation():
# Plant Model
model_plant = SIMOSystem(nsim=30,
x0=np.array([2, 2 / 3]),
u0=np.array([2]),
xs=np.array([4, 4 / 3]),
us=np.array([4]),
step_size=0.2,
control=False,
q_cost=1,
r_cost=0.5,
random_seed=1)
# Build Controller Model
model_control = SIMOSystem(nsim=model_plant.Nsim,
x0=np.array([2., 2 / 3, 0, 0]),
u0=np.array([2.]),
xs=np.array([4., 4 / 3, 0, 0]),
us=np.array([4.]),
step_size=0.2,
control=True,
q_cost=1,
r_cost=0.5,
random_seed=1)
# MPC Object Initiation
control = ModelPredictiveControl(model_control.Nsim,
30,
model_control.Nx,
model_control.Nu,
1,
0.5,
0.0,
model_control.xs,
model_control.us,
eval_time=model_plant.step_size,
dist=True,
gamma=0.5)
# MPC Construction
mpc_control = control.get_mpc_controller(model_control.system_ode,
delta=control.eval_time,
x0=model_control.x0,
random_guess=False)
"""
Simulation portion
"""
for t in range(1, model_plant.Nsim + 1):
# Solve the MPC optimization problem, obtain current input and predicted state
model_control.u[t, :], model_control.x[t, :] = control.solve_mpc(
model_plant.x, model_plant.xsp, mpc_control, t, control.p)
# Calculate the next states for the plant
model_plant.x[t, :] = model_plant.system_sim.sim(
model_plant.x[t - 1, :], model_control.u[t, :])
# Disturbance
# if t % 20 == 0:
# model_plant.x[t, 1] -= 5
# Update the P parameters for offset-free control
control.p = model_plant.x[t, :] - model_control.x[t, 0:model_plant.Nx]
# Update the plant inputs to be same as controller inputs
if t == model_plant.Nsim:
model_plant.u = deepcopy(model_control.u)
return model_plant, model_control, control
def simulation():
# Plant Model
model_plant = LinearSystem(nsim=100,
model_type='SISO',
x0=np.array([0.5]),
u0=np.array([1]),
xs=np.array([5]),
us=np.array([10]),
step_size=0.2)
# Build Controller Model
model_control = LinearSystem(nsim=model_plant.Nsim,
model_type=model_plant.model_type,
x0=np.array([0.5, 0]),
u0=np.array([1]),
xs=np.array([5, 0]),
us=np.array([10]),
control=True,
step_size=0.2)
# MPC Object Initiation
control = ModelPredictiveControl(model_control.Nsim,
10,
model_control.Nx,
model_control.Nu,
1,
0.5,
0.0,
model_control.xs,
model_control.us,
eval_time=model_plant.step_size,
dist=True,
gamma=0.9)
# MPC Construction
mpc_control = control.get_mpc_controller(model_control.system_ode,
delta=control.eval_time,
x0=model_control.x0,
verbosity=0,
random_guess=False)
"""
Simulation portion
"""
for t in range(1, model_plant.Nsim + 1):
# Setpoint Change
if t % 51 == 0:
# MPC Object Initiation
control = ModelPredictiveControl(49,
10,
model_control.Nx,
model_control.Nu,
1,
0.5,
0.0,
np.array([2.5]),
np.array([5]),
eval_time=model_plant.step_size,
dist=True,
gamma=0.9)
# MPC Construction
mpc_control = control.get_mpc_controller(model_control.system_ode,
delta=control.eval_time,
x0=model_control.x0,
verbosity=0,
random_guess=False)
# Solve the MPC optimization problem, obtain current input and predicted state
model_control.u[t, :], model_control.x[t, :] = control.solve_mpc(
model_plant.x, model_plant.xsp, mpc_control, t, control.p)
# Calculate the next states for the plant
model_plant.x[t, :] = model_plant.system_sim.sim(
model_plant.x[t - 1, :], model_control.u[t, :])
# Update the P parameters for offset-free control
control.p = model_plant.x[t, :] - model_control.x[t, 0:model_plant.Nx]
# Update the plant inputs to be same as controller inputs
if t == model_plant.Nsim:
model_plant.u = deepcopy(model_control.u)
return model_plant, model_control, control
def simulation():
# Build PID Objects
PID1 = DiscretePIDControl(kp=1.31, ki=0.21, kd=0)
PID2 = DiscretePIDControl(kp=-0.28, ki=-0.06, kd=0)
# Initialize PIDs
PID1.u = [3.9, 3.9, 3.9, 3.9, 3.9, 3.9, 3.9, 3.9]
PID2.u = [0, 0, 0, 0, 0, 0, 0, 0]
input_1 = PID1.u[-1]
input_2 = PID2.u[-1]
# Initial set-points
set_point1 = 100
set_point2 = 0
# Plant model
model_plant = WoodberryDistillation(nsim=2000)
# Build Controller Model
model_control = WoodberryDistillation(
nsim=model_plant.Nsim,
x0=np.array([65.13, 42.55, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]),
xs=np.array([261.78, 171.18, 111.43, 76.62, 0.0, 0.0, 0.0, 0.0]),
control=True)
"""
Below are all the MPC and controller objects
"""
# Build MPC Object
control = ModelPredictiveControl(
model_control.Nsim,
10,
model_control.Nx,
model_control.Nu,
1,
0.0,
0.0,
model_control.xs,
model_control.us,
eval_time=15,
dist=True,
gamma=0.9,
upp_u_const=[99, 99, 99, 99],
low_u_const=[0.0, 0.0, 0.0, 0.0],
upp_x_const=[1000, 1000, 1000, 1000, 1000, 1000, 1000, 1000],
low_x_const=[-1000, -1000, -1000, -1000, -1000, -1000, -1000, -1000])
# MPC Construction
mpc_control = control.get_mpc_controller(model_control.system_ode,
delta=control.eval_time,
x0=model_control.x0,
verbosity=0,
random_guess=False)
# Build FTC-MPC Object
ftc_control = ModelPredictiveControl(
model_control.Nsim,
10,
model_control.Nx,
model_control.Nu,
1,
0.0,
0.0,
np.array([200.33, 130.86, 60.87, 41.74, 0, 0, 0, 0]),
model_control.us,
eval_time=15,
dist=True,
gamma=0.9,
upp_u_const=[12.1, 12.1, 99, 99],
low_u_const=[11.9, 11.9, 0, 0],
upp_x_const=[1000, 1000, 1000, 1000, 1000, 1000, 1000, 1000],
low_x_const=[-1000, -1000, -1000, -1000, -1000, -1000, -1000, -1000])
# MPC FTCConstruction
ftc_mpc_control = ftc_control.get_mpc_controller(model_control.system_ode,
delta=control.eval_time,
x0=model_control.x0,
verbosity=0,
random_guess=False)
for t in range(7, model_plant.Nsim + 1):
# Initial 355 minutes of simulation
if t % 15 == 0 and t <= 355:
# Solve the MPC optimization problem, obtain current input and predicted state
model_control.u[t, :], model_control.x[t, :] = control.solve_mpc(
model_plant.x, model_plant.xsp, mpc_control, t, control.p)
# Evaluate FT-MPC
elif t % 15 == 0 and t > 355:
# Solve the MPC optimization problem, obtain current input and predicted state
model_control.u[t, :], model_control.x[
t, :] = ftc_control.solve_mpc(model_plant.x, model_plant.xsp,
ftc_mpc_control, t, control.p)
print(t, model_control.u[t, :])
# When MPC does not evaluate
else:
model_control.u[t, :] = model_control.u[t - 1, :]
model_control.x[t, :] = model_control.x[t - 1, :]
# Fault in the system
if t >= 340:
model_control.u[t, 0:2] = 12
if t % 4 == 0:
input_1 = PID1(model_control.u[t, 0], env.y[t - 1, 0],
env.y[t - 2, 0], env.y[t - 3, 0])
input_2 = PID2(model_control.u[t, 2], env.y[t - 1, 1],
env.y[t - 2, 1], env.y[t - 3, 1])
# Generate input tuple
control_input = np.array([input_1, input_2])
# Calculate the next states for the plant
model_plant.x[t, :] = model_plant.system_sim.sim(
model_plant.x[t - 1, :], control_input)
# Populate the output values for model and control
model_control.y[t, 0] = model_control.C[0, 0] * model_control.x[
t, 0] + model_control.C[0, 2] * model_control.x[t, 2]
model_control.y[t, 1] = model_control.C[1, 1] * model_control.x[
t, 1] + model_control.C[1, 3] * model_control.x[t, 3]
model_plant.y[t, 0] = model_plant.C[0, 0] * model_plant.x[
t, 0] + model_plant.C[0, 2] * model_plant.x[t, 2]
model_plant.y[t, 1] = model_plant.C[1, 1] * model_plant.x[
t, 1] + model_plant.C[1, 3] * model_plant.x[t, 3]
# Disturbance
# if t % 20 == 0:
# model_plant.x[t, 1] -= 5
# Update the P parameters for offset-free control
control.p = model_plant.x[t, :] - model_control.x[t, 0:model_plant.Nx]
ftc_control.p = model_plant.x[t, :] - model_control.x[t,
0:model_plant.Nx]
# Update the plant inputs to be same as controller inputs
if t == model_plant.Nsim:
model_plant.u = deepcopy(model_control.u)
return model_plant, model_control, control
def simulation():
# Build PID Objects
PID1 = DiscretePIDControl(kp=1.31, ki=0.21, kd=0)
PID2 = DiscretePIDControl(kp=-0.28, ki=-0.06, kd=0)
# Initialize PIDs
PID1.u = [3.9, 3.9, 3.9, 3.9, 3.9, 3.9, 3.9, 3.9]
PID2.u = [0, 0, 0, 0, 0, 0, 0, 0]
input_1 = PID1.u[-1]
input_2 = PID2.u[-1]
# Initial set-points
set_point1 = 100
set_point2 = 0
# Overcome index error
set_point1_list = [100, 100]
set_point2_list = [0, 0]
# Plant model
model_plant = WoodberryDistillation(nsim=2000)
# Build Controller Model
model_control = WoodberryDistillation(
nsim=model_plant.Nsim,
x0=np.array([65.13, 42.55, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]),
xs=np.array([261.78, 171.18, 111.43, 76.62, 0.0, 0.0, 0.0, 0.0]),
control=True)
"""
Below are all the MPC and controller objects
"""
# Build MPC Object
control = ModelPredictiveControl(
model_control.Nsim,
10,
model_control.Nx,
model_control.Nu,
1,
0.0,
0.0,
model_control.xs,
model_control.us,
eval_time=15,
dist=True,
gamma=0.9,
upp_u_const=[99, 99, 99, 99],
low_u_const=[0.0, 0.0, 0.0, 0.0],
upp_x_const=[1000, 1000, 1000, 1000, 1000, 1000, 1000, 1000],
low_x_const=[-1000, -1000, -1000, -1000, -1000, -1000, -1000, -1000])
# MPC Construction
mpc_control = control.get_mpc_controller(model_control.system_ode,
delta=control.eval_time,
x0=model_control.x0,
verbosity=0,
random_guess=False)
# Build FTC-MPC Object
ftc_control = ModelPredictiveControl(
model_control.Nsim,
10,
model_control.Nx,
model_control.Nu,
1,
0.0,
0.0,
np.array([200.33, 130.86, 60.87, 41.74, 0, 0, 0, 0]),
model_control.us,
eval_time=15,
dist=True,
gamma=0.9,
upp_u_const=[12.1, 12.1, 99, 99],
low_u_const=[11.9, 11.9, 0, 0],
upp_x_const=[1000, 1000, 1000, 1000, 1000, 1000, 1000, 1000],
low_x_const=[-1000, -1000, -1000, -1000, -1000, -1000, -1000, -1000])
# MPC FTCConstruction
ftc_mpc_control = ftc_control.get_mpc_controller(model_control.system_ode,
delta=control.eval_time,
x0=model_control.x0,
verbosity=0,
random_guess=False)
for t in range(10, model_plant.Nsim + 1):
# How often to evaluate MPC
eval_time = 15
# Initial 355 minutes of simulation
if t % eval_time == 0 and t <= 340:
# Solve the MPC optimization problem, obtain current input and predicted state
model_control.u[t, :], model_control.x[t, :] = control.solve_mpc(
model_plant.x, model_plant.xsp, mpc_control, t, control.p)
# Evaluate FT-MPC
elif t % eval_time == 0 and t > 359:
# Introduce noise in states
# model_plant.x += np.random.uniform(-5, 5)
# Solve the MPC optimization problem, obtain current input and predicted state
model_control.u[t, :], model_control.x[
t, :] = ftc_control.solve_mpc(model_plant.x, model_plant.xsp,
ftc_mpc_control, t, control.p)
# When MPC does not evaluate
else:
model_control.u[t, :] = model_control.u[t - 1, :]
model_control.x[t, :] = model_control.x[t - 1, :]
# Convert model output to a setpoint
if t % eval_time == 0:
set_point1 = 0.7665 * model_control.x[
t, 0] - 0.9 * model_control.x[t, 2]
set_point2 = 0.6055 * model_control.x[
t, 1] - 1.3472 * model_control.x[t, 3]
# Input constraints
if (set_point1 - set_point1_list[-2]) > 10:
set_point1 = set_point1_list[-2] + 10
elif (set_point1 - set_point1_list[-2]) < -10:
set_point1 = set_point1_list[-2] - 10
if (set_point2 - set_point2_list[-2]) > 10:
set_point2 = set_point2_list[-2] + 10
elif (set_point2 - set_point2_list[-2]) < -10:
set_point2 = set_point2_list[-2] - 10
set_point1_list.append(set_point1)
set_point2_list.append(set_point2)
if t < 360:
set_point1 = 100
set_point2 = 0
if t % 4 == 0:
input_1 = PID1(set_point1, model_plant.y[t - 1, 0],
model_plant.y[t - 2, 0], model_plant.y[t - 3, 0])
input_2 = PID2(set_point2, model_plant.y[t - 1, 1],
model_plant.y[t - 2, 1], model_plant.y[t - 3, 1])
# Fault occurs at 340
if 347 <= t < 363:
control_input = np.array([12, 12, 5.337, 5.337])
elif t >= 363:
control_input = np.array([12, 12, input_2, input_2])
else:
control_input = np.array([15.7, 15.7, 5.337, 5.337])
# Calculate the next states for the plant
model_plant.x[t, :] = model_plant.system_sim.sim(
model_plant.x[t - 1, :], control_input)
# Populate the output values for model and control
model_control.y[t, 0] = model_control.C[0, 0] * model_control.x[
t, 0] + model_control.C[0, 2] * model_control.x[t, 2]
model_control.y[t, 1] = model_control.C[1, 1] * model_control.x[
t, 1] + model_control.C[1, 3] * model_control.x[t, 3]
model_plant.y[t, 0] = model_plant.C[0, 0] * model_plant.x[
t, 0] + model_plant.C[0, 2] * model_plant.x[t, 2]
model_plant.y[t, 1] = model_plant.C[1, 1] * model_plant.x[
t, 1] + model_plant.C[1, 3] * model_plant.x[t, 3]
if 99 < model_plant.y[t, 0] < 101 and t > 390:
print(
'Time to mediate: {} | RMSE: {} | t: {} | Value: {:2f}'.format(
t - 360, np.sum(np.abs(model_plant.y[360:t, 0] - 100)), t,
model_plant.y[t, 0]))
# Update the P parameters for offset-free control
control.p = model_plant.x[t, :] - model_control.x[t, 0:model_plant.Nx]
ftc_control.p = model_plant.x[t, :] - model_control.x[t,
0:model_plant.Nx]
# Update the plant inputs to be same as controller inputs
if t == model_plant.Nsim:
model_plant.u = deepcopy(model_control.u)
# Plots the output
plt.plot(model_plant.y[330:1000, 0])
plt.plot(model_plant.y[330:1000, 1])
plt.show()
return model_plant, model_control, control, set_point2_list