View source for Dynamic Simulation Tutorial with DWSIM and Python, Part 4: Tuning the PID Controller through Non-Linear Optimization

<div id="mw-content-text" lang="en" dir="ltr" class="mw-content-ltr"><table style="color: white; background-color: steelblue;" cellpadding="10" width="100%">
<tr>
<td colspan="1"> [[File:Dialog-warning.png]] This tutorial requires advanced or above average Python programming skills.
</td></tr>
</table>
<table style="color: white; background-color: steelblue; margin-top: 5px" cellpadding="10" width="100%">
<tr>
<td colspan="1"> [[File:Dialog-information.png]] You'll need at least <b>DWSIM v5.7 (Cross-Platform UI)</b> on Windows, Linux or macOS to follow/reproduce the tasks within this tutorial.
</td></tr>
</table><br/>
<h1><span class="mw-headline" id="Introduction">Introduction</span></h1>
<p>We can use built-in DWSIM non-linear optimizers to tune our PID controller by finding a parameter set (Kp, Ki, Kd) which minimizes the total error between the set-point and process variable during the whole simulation period.
</p>
<h1><span class="mw-headline" id="Setup">Setup</span></h1>
<p>Modify the 'RunDynamicProcess_ClosedLoop' to read the PID parameters stored in the PID Controller, by changing
</p>
<source lang=python>P = 0.5
I = 0.01
D = 0.1</source>
<p>to
</p>
<source lang=python>P = controller.ExtraProperties.Kp
I = controller.ExtraProperties.Ki
D = controller.ExtraProperties.Kd</source>

<p>When you save the simulation, the latest parameter values will be saved as extra properties of the PID Controller object.
</p><p>You can comment the chart generation lines so you won't get a chart window at each optimizer iteration:
</p>
<source lang=python>#outputresults = Flowsheet.Scripts.Values.Where(lambda x: x.Title == 'GenerateCharts').FirstOrDefault().ScriptText.replace('\r', '')
#exec(outputresults)</source>

<h1><span class="mw-headline" id="Tuning_the_PID_Controller_using_the_Simplex_method">Tuning the PID Controller using the Simplex method</span></h1>
<p>Create a new script and name it 'TunePID_Simplex', with the following content:
</p>
<source lang=python>import clr
import System
from System import *
 
clr.AddReference('DWSIM.MathOps.DotNumerics')
clr.AddReference('System.Core')
clr.ImportExtensions(System.Linq)
 
from System import Func
 
from DotNumerics import *
from DotNumerics.Optimization import *
 
controller = Flowsheet.GetFlowsheetSimulationObject("PID Controller")
 
controller.ExtraProperties.Kp = 0.5
controller.ExtraProperties.Ki = 0.01
controller.ExtraProperties.Kd = 0.1
 
P = controller.ExtraProperties.Kp
I = controller.ExtraProperties.Ki
D = controller.ExtraProperties.Kd
 
simplex = Simplex()
 
Pvar = OptBoundVariable(P, 0.0, 10.0)
Ivar = OptBoundVariable(I, 0.0, 10.0)
Dvar = OptBoundVariable(D, 0.0, 10.0)
 
vars = [Pvar, Ivar, Dvar]
 
tx0 = DateTime.Now
 
counter = 1
 
def RunIteration(vars):
    global counter
    Flowsheet.SupressMessages = False
    Flowsheet.WriteMessage("Run #" + str(counter))
    Flowsheet.WriteMessage("P = " + str(controller.ExtraProperties.Kp))
    Flowsheet.WriteMessage("I = " + str(controller.ExtraProperties.Ki))
    Flowsheet.WriteMessage("D = " + str(controller.ExtraProperties.Kd))
    controller.ExtraProperties.Kp = vars[0]
    controller.ExtraProperties.Ki = vars[1]
    controller.ExtraProperties.Kd = vars[2]
    Flowsheet.SupressMessages = True
    itloop = Flowsheet.Scripts.Values.Where(lambda x: x.Title == 'RunDynamicProcess_ClosedLoop').FirstOrDefault().ScriptText.replace('\r', '')
    exec(itloop)
    Flowsheet.SupressMessages = False
    Flowsheet.WriteMessage("Error = " + str(controller.ExtraProperties.TotalError))
    counter += 1
    return controller.ExtraProperties.TotalError
 
simplex.Tolerance = 0.000001
simplex.MaxFunEvaluations = 1000
simplex.ComputeMin(lambda x: RunIteration(x), Array[OptBoundVariable](vars))
 
tx1 = DateTime.Now
 
controller.ExtraProperties.TotalTuningTime = (tx1 - tx0).TotalSeconds</source>

<p>Run the above script. It can take several hours to complete, depending on your system's processing power.
</p>
<h1><span class="mw-headline" id="Tuning_the_PID_Controller_using_IPOPT">Tuning the PID Controller using IPOPT</span></h1>
<p>Create a new script and name it 'TunePID_IPOPT', with the following content:
</p>
<source lang=python>import clr
import System
from System import *
 
clr.AddReference('Cureos.Numerics')
clr.AddReference('System.Core')
clr.ImportExtensions(System.Linq)
 
from System import Func
 
from Cureos.Numerics import *
 
controller = Flowsheet.GetFlowsheetSimulationObject("PID Controller")
 
controller.ExtraProperties.Kp = 0.5
controller.ExtraProperties.Ki = 0.01
controller.ExtraProperties.Kd = 0.1
 
P = controller.ExtraProperties.Kp
I = controller.ExtraProperties.Ki
D = controller.ExtraProperties.Kd
 
lconstr = [0.0, 0.0, 0.0]
uconstr = [10.0, 10.0, 10.0]
vars = [P, I, D]
 
tx0 = DateTime.Now
 
counter = 1
 
Flowsheet.SupressMessages = False
     
def RunIteration(vars):
    global counter
    Flowsheet.SupressMessages = False
    Flowsheet.WriteMessage("Run #" + str(counter))
    Flowsheet.WriteMessage("P = " + str(controller.ExtraProperties.Kp) + ", I = " + str(controller.ExtraProperties.Ki) + ", D = " + str(controller.ExtraProperties.Kd))
    controller.ExtraProperties.Kp = vars[0]
    controller.ExtraProperties.Ki = vars[1]
    controller.ExtraProperties.Kd = vars[2]
    Flowsheet.SupressMessages = True
    itloop = Flowsheet.Scripts.Values.Where(lambda x: x.Title == 'RunDynamicProcess_ClosedLoop').FirstOrDefault().ScriptText.replace('\r', '')
    exec(itloop)
    Flowsheet.SupressMessages = False
    Flowsheet.WriteMessage("Error = " + str(controller.ExtraProperties.TotalError))
    counter += 1
    return controller.ExtraProperties.TotalError
     
def CalculateGradient(x):
    epsilon = 0.001
    x1 = [0.0, 0.0, 0.0]
    x2 = [0.0, 0.0, 0.0]
    g = [0.0, 0.0, 0.0]
    for j in range(0, x.Length):
        for k in range(0, x.Length):
            x1[k] = x[k]
            x2[k] = x[k]
        x1[j] = x[j] + epsilon
        x2[j] = x[j] - epsilon
        f1 = RunIteration(x1)
        f2 = RunIteration(x2)
        g[j] = (f2 - f1) / (x2[j] - x1[j])
    return Array[Double](g)
     
def func(n, x, new_x, obj_value):
    obj_value.Value = RunIteration(x)
    return True
     
def gradfunc(n, x, new_x, grad_f):
    grad_f.Value = CalculateGradient(x)
    return True
 
ipopt = Ipopt(3, Array[float](lconstr), Array[float](uconstr), 
                0, None, None, 0, 0, 
                lambda n, x, new_x, obj_value: func(n, x, new_x, obj_value), 
                lambda n, x, new_x, m, g: True, 
                lambda n, x, new_x, grad_f: gradfunc(n, x, new_x, grad_f), 
                lambda n, x, new_x, m, nele_jac, iRow, jCol, values: False, 
                lambda n, x, new_x, obj_factor, m, lambda0, new_lambda0, nele_hess, iRow, jCol, values: False)
 
obj = clr.Reference[float]()
 
ipopt.AddOption("tol", 0.000001)
ipopt.AddOption("max_iter", 1000)
ipopt.AddOption("mu_strategy", "adaptive")
ipopt.AddOption("hessian_approximation", "limited-memory")
status = ipopt.SolveProblem(Array[float](vars), obj, None, None, None, None)
 
print status
 
tx1 = DateTime.Now
 
controller.ExtraProperties.TotalTuningTime = (tx1 - tx0).TotalSeconds</source>

<p>Run the above script. It can take several hours to complete, depending on your system's processing power.
</p>
<h1><span class="mw-headline" id="Viewing_Tuning_Results">Viewing Tuning Results</span></h1>
<p>After successfully tuning the PID, you should get charts that look like the following ones:
</p><p>[[File:Dynamic_tuned.jpg]]
</p>
<h2><span class="mw-headline" id="Download_File">Download File</span></h2>
<p>Download the simulation file with what has been done so far: [http://dwsim.inforside.com.br/files/dynamic_part4.dwxmz dynamic_part4.dwxmz]
</p><p>Return to [[Dynamic Simulation Tutorial with DWSIM and Python, Part 3: Adding a PID Controller]]
</p>