Main

Model Initialization Strategies

Main.ModelInitialization History

Hide minor edits - Show changes to output

Added line 23:
fid.write(' m ='+str(m)+' \n')
Changed line 26 from:
fid.write(' p[1:'+str(n)+'][1::'+str(m)+'] \n')
to:
fid.write(' p[1:n][1::m] \n')
Changed line 9 from:
Attach:download.png [[Attach:apmonitor_matrix.zip|Initialize Parameter Matrix]]
to:
Attach:download.png [[Attach:apmonitor_matrix.zip|Initialize Parameter Matrix (MATLAB and Python)]]
August 16, 2017, at 05:43 PM by 10.5.113.229 -
Changed line 54 from:
The following example is a demonstration of inserting different initial conditions or parameter values at points throughout the time horizon. A simulation solution is used to provide guess values for a subsequent simulation. The parameter [[http://apmonitor.com/wiki/index.php/Main/OptionApmCsvRead|CSV_READ]] can be set to ''2'' to provide the initial values for a calculated state. The default (''CSV_READ=1'') only updates the fixed values and skips the values that are calculated by the solver.
to:
The following example is a demonstration of inserting different initial conditions or parameter values at points throughout the time horizon. A simulation solution is used to provide guess values for a subsequent simulation. The parameter [[http://apmonitor.com/wiki/index.php/Main/OptionApmCsvRead|CSV_READ]] can be set to ''2'' to provide the initial values for a calculated state. The default (''CSV_READ=1'') only updates the fixed values and skips the values that are calculated by the solver. Setting [[http://apmonitor.com/wiki/index.php/Main/OptionApmColdstart|COLDSTART]] >= 1 also has the effect of using calculated values in the CSV file as initial guesses for the solver.
August 16, 2017, at 05:41 PM by 10.5.113.229 -
Changed lines 5-6 from:
Simulation is a first step after the model development to verify convergence, validate the model response to input changes, and manually adjust parameters to fit an expected response. This section demonstrates how to set up and initialize a dynamic model simulation. Options such as [[Main/OptionApmCsvRead|CSV_READ]] control how much information is read from a data (CSV) file.
to:
Simulation is a first step after the model development to verify convergence, validate the model response to input changes, and manually adjust parameters to fit an expected response. This section demonstrates how to set up and initialize a dynamic model simulation. Options such as [[http://apmonitor.com/wiki/index.php/Main/OptionApmCsvRead|CSV_READ]] control how much information is read from a data (CSV) file.
Changed line 54 from:
The following example is a demonstration of inserting different initial conditions or parameter values at points throughout the time horizon. A simulation solution is used to provide guess values for a subsequent simulation. The parameter [[Main/OptionApmCsvRead|CSV_READ]] can be set to ''2'' to provide the initial values for a calculated state. The default (''CSV_READ=1'') only updates the fixed values and skips the values that are calculated by the solver.
to:
The following example is a demonstration of inserting different initial conditions or parameter values at points throughout the time horizon. A simulation solution is used to provide guess values for a subsequent simulation. The parameter [[http://apmonitor.com/wiki/index.php/Main/OptionApmCsvRead|CSV_READ]] can be set to ''2'' to provide the initial values for a calculated state. The default (''CSV_READ=1'') only updates the fixed values and skips the values that are calculated by the solver.
August 16, 2017, at 05:40 PM by 10.5.113.229 -
Changed lines 5-6 from:
Simulation is a first step after the model development to verify convergence, validate the model response to input changes, and manually adjust parameters to fit an expected response. This section demonstrates how to set up and initialize a dynamic model simulation. A first example shows how to use a scripting language such as MATLAB or Python to provide input values for parameters.
to:
Simulation is a first step after the model development to verify convergence, validate the model response to input changes, and manually adjust parameters to fit an expected response. This section demonstrates how to set up and initialize a dynamic model simulation. Options such as [[Main/OptionApmCsvRead|CSV_READ]] control how much information is read from a data (CSV) file.

A first example shows how to use a scripting language such as MATLAB or Python to provide input values for a matrix of
parameters.
Changed line 54 from:
The following example is a demonstration of inserting different initial conditions or parameter values at points throughout the time horizon. A simulation solution is used to provide guess values for a subsequent simulation.
to:
The following example is a demonstration of inserting different initial conditions or parameter values at points throughout the time horizon. A simulation solution is used to provide guess values for a subsequent simulation. The parameter [[Main/OptionApmCsvRead|CSV_READ]] can be set to ''2'' to provide the initial values for a calculated state. The default (''CSV_READ=1'') only updates the fixed values and skips the values that are calculated by the solver.
July 26, 2017, at 11:23 PM by 189.223.138.84 -
Added lines 8-48:

(:source lang=python:)
from apm import *
import numpy as np

A = np.random.random((3,4))
n = np.size(A,0) # rows
m = np.size(A,1) # columns

# write model.apm
fid = open('model.apm','w')
fid.write('Constants \n')
fid.write(' n ='+str(n)+' \n')
fid.write(' \n')
fid.write('Parameters \n')
fid.write(' p[1:'+str(n)+'][1::'+str(m)+'] \n')
fid.write(' \n')
fid.write('Variables \n')
fid.write(' x \n')
fid.write('Equations \n')
fid.write(' x=p[1][1] \n')
fid.close()

# write data.csv
fid = open('data.csv','w')
for i in range(n):
    for j in range(m):
        fid.write(' p['+str(i+1)+']['+str(j+1)+'], '+str(A[i,j])+' \n')
fid.close()

# load model, data file, and solve
s = 'http://byu.apmonitor.com'
a = 'matrix_write'
apm(s,a,'clear all')
apm_load(s,a,'model.apm')
csv_load(s,a,'data.csv')
apm(s,a,'solve')

# retrieve solution
apm_web_var(s,a)
(:sourceend:)
July 26, 2017, at 11:07 PM by 189.223.138.84 -
Added lines 16-128:

(:source lang=python:)
import numpy as np
from apm import *  # load APMonitor library

##################################################################
## Step #1 - Solve model with p = 1
##################################################################

## Step #1a - write data.csv
n = 31
time = np.linspace(0,3,n)
p = np.ones(31)
x = 2 * np.ones(31)

fid = open('data.csv','w')

## write time row
fid.write('time, ')
for i in range(n-1):
    fid.write(str(time[i]) +  ', ')
fid.write(str(time[n-1]) + '\n')

## write 'p' row (input parameter)
fid.write('p, ')
for i in range(n-1):
    fid.write(str(p[i]) +  ', ')
fid.write(str(p[n-1]) + '\n')

## write 'x' row (state variable initialization)
fid.write('x, ')
# imode: http://apmonitor.com/wiki/index.php/Main/Modes
# for imode=4-6, include all initialization values
# for imode=7-9, include only the initial condition for variables
imode = 7
if ((imode>=4) and (imode<=6)):
    for i in range(n-1):
        fid.write(str(x[i]) + ', ')
    fid.write(str(x[n-1]) + '\n')
else:
    fid.write(str(x[0]) + ', ')
    for i in range(1,n-1):
        fid.write('-, ')
    fid.write('-\n')

# close file
fid.close()

## Step 1b - Load and solve model
s = 'http://byu.apmonitor.com'
a = 'model_init'

apm(s,a,'clear all')
apm_load(s,a,'model.apm')
csv_load(s,a,'data.csv')

apm_option(s,a,'apm.time_shift',1)
apm_option(s,a,'apm.imode',imode)
output1 = apm(s,a,'solve')

## Step 1c - Retrieve results with solution.csv
solution1 = apm_sol(s,a)

##################################################################
## Step 2 - Solve again with prior solution for initialization and
##          p as a step from 0 to 2
##################################################################

## Change to imode = 4 and change p trajectory
p[0:5] = 0.0
p[5:n] = 2.0

## Step 2a - Write new row at the end of solution.csv
fname = 'solution_' + a + '.csv'
fid = open(fname,'a')  # append to file
fid.write('p, ')
for i in range(n-1):
    fid.write(str(p[i]) +  ', ')
fid.write(str(p[n-1]) + '\n')
# close file
fid.close()

## Step 2b - Reload csv file for initialization
apm(s,a,'clear csv')
csv_load(s,a,fname)

## Step 2c - Solve again but with new inputs
imode = 4
apm_option(s,a,'apm.time_shift',0)
apm_option(s,a,'apm.imode',imode)
output2 = apm(s,a,'solve')
print(output2)

## Step 2d - Retrieve results with solution.csv
solution2 = apm_sol(s,a)

##################################################################
## Step 3 - Create plots
##################################################################
import matplotlib.pyplot as plt

plt.figure(1)
plt.subplot(2,1,1)
plt.plot(solution1['time'],solution1['p'],'k-',linewidth=2)
plt.plot(solution2['time'],solution2['p'],'b--',linewidth=2)
plt.legend([r'$p_1$',r'$p_2$'])
plt.subplot(2,1,2)
plt.plot(solution1['time'],solution1['x'],'r--',linewidth=2)
plt.plot(solution2['time'],solution2['x'],'g:',linewidth=2)
plt.legend([r'$x_1$',r'$x_2$'])
plt.xlabel('time')
plt.show()
(:sourceend:)
July 26, 2017, at 11:04 PM by 189.223.138.84 -
Changed lines 5-15 from:
Simulation is a first step in after the model development to verify convergence, validate the model response to input changes, and manually adjust parameters to fit an expected response. This section demonstrates how to set up and solve a dynamic model.

(:html:)
<iframe width="560" height="315" src="https://www.youtube.com/embed/-IDTagajoyA" frameborder="0" allowfullscreen></iframe>
(:htmlend:)

(:html:)
<iframe width="560" height="315" src="https://www.youtube.com/embed/-3FaZEfu7vE" frameborder="0" allowfullscreen></iframe>
(:htmlend:)

Although a problem may be written correctly, sometimes the solver fails to find a solution or requires excessive time to find a solution. Initialization strategies are critical in these situations to find a nearby solution that seeds the optimization problem with a starting point that allows convergence'^1^'.
to:
Simulation is a first step after the model development to verify convergence, validate the model response to input changes, and manually adjust parameters to fit an expected response. This section demonstrates how to set up and initialize a dynamic model simulation. A first example shows how to use a scripting language such as MATLAB or Python to provide input values for parameters.

Attach:download.png [[Attach:apmonitor_matrix.zip|Initialize Parameter Matrix]]

Although a problem may be written correctly, sometimes the solver fails to find a solution or requires excessive time to find a solution. Initialization strategies are critical in these situations to find a nearby solution that seeds the optimization problem with a starting point that allows convergence'^1^'.

The following example is a demonstration of inserting different initial conditions or parameter values at points throughout the time horizon. A simulation solution is used to provide guess values for a subsequent simulation.

Attach:download.png [[Attach:apmonitor_initialize.zip|Initialize for Dynamic Simulation]]

%width=550px%Attach:apmonitor_initialize.png
June 09, 2017, at 01:05 AM by 10.5.113.159 -
Changed line 63 from:
# 5.Safdarnejad, S.M., Hedengren, J.D., Lewis, N.R., Haseltine, E., Initialization Strategies for Optimization of Dynamic Systems, Computers and Chemical Engineering, 2015, Vol. 78, pp. 39-50, DOI: 10.1016/j.compchemeng.2015.04.016. [[http://www.sciencedirect.com/science/article/pii/S0098135415001179 | Article]]
to:
# Safdarnejad, S.M., Hedengren, J.D., Lewis, N.R., Haseltine, E., Initialization Strategies for Optimization of Dynamic Systems, Computers and Chemical Engineering, 2015, Vol. 78, pp. 39-50, DOI: 10.1016/j.compchemeng.2015.04.016. [[http://www.sciencedirect.com/science/article/pii/S0098135415001179 | Article]]
Changed lines 63-64 from:
# Safdarnejad, S.M., Hedengren, J.D., Lewis, N.R., Haseltine, E., Initialization Strategies for Optimization of Dynamic Systems, Computers and Chemical Engineering, DOI: 10.1016/j.compchemeng.2015.04.016. [[http://www.sciencedirect.com/science/article/pii/S0098135415001179 | Article]]
to:
# 5.Safdarnejad, S.M., Hedengren, J.D., Lewis, N.R., Haseltine, E., Initialization Strategies for Optimization of Dynamic Systems, Computers and Chemical Engineering, 2015, Vol. 78, pp. 39-50, DOI: 10.1016/j.compchemeng.2015.04.016. [[http://www.sciencedirect.com/science/article/pii/S0098135415001179 | Article]]
Added lines 66-68:

# Lewis, N.R., Hedengren, J.D., Haseltine, E.L., Hybrid Dynamic Optimization Methods for Systems Biology with Efficient Sensitivities, Special Issue on Algorithms and Applications in Dynamic Optimization, Processes, 2015, 3(3), 701-729; doi:10.3390/pr3030701. [[http://www.mdpi.com/2227-9717/3/3/701/html | Article]]

May 11, 2015, at 06:33 PM by 45.56.3.184 -
Changed line 51 from:
With guess values for parameters (kr'_1..6_'), approximately match the laboratory data for this patient. A subsequent section introduces methods for parameter estimation by minimizing an objective function.
to:
With guess values for parameters (kr'_1..6_'), approximately match the laboratory data for this patient. [[Main/EstimatorObjective|A subsequent section]] introduces methods for parameter estimation by minimizing an objective function.
May 01, 2015, at 01:58 PM by 45.56.3.184 -
Changed lines 30-34 from:
 dH/dt = kr^_1_^ - kr^_2_^ H - kr^_3_^ H V
 dI/dt = kr^_3_^ H V - kr^_4_^ I
 dV/dt = -kr^_3_^ H V - kr^_5_^ V + kr^_6_^ I
 LV = log^_10_^(V)
to:
 dH/dt = kr'_1_' - kr'_2_' H - kr'_3_' H V
 dI/dt = kr'_3_' H V - kr'_4_' I
 dV/dt = -kr'_3_' H V - kr'_5_' V + kr'_6_' I
 LV = log'_10_'(V)
Changed lines 38-43 from:
 kr^_1_^ = new healthy cells
 kr^_2_^ = death rate of healthy cells
 kr^_3_^ = healthy cells converting to infected cells
 kr^_4_^ = death rate of infected cells
 kr^_5_^ = death rate of virus
 kr^_6_^ = production of virus by infected cells
to:
 kr'_1_' = new healthy cells
 kr'_2_' = death rate of healthy cells
 kr'_3_' = healthy cells converting to infected cells
 kr'_4_' = death rate of infected cells
 kr'_5_' = death rate of virus
 kr'_6_' = production of virus by infected cells
May 01, 2015, at 01:56 PM by 45.56.3.184 -
Changed lines 21-45 from:
The spread of HIV in a patient is approximated with balance equations on (H)ealthy, (I)nfected, and (V)irus population counts'^2^'. There are six parameters (kr'_1..6_') in the model that provide the rates of cell death, infection spread, virus replication, and other processes that determine the spread of HIV in the body. The following data is provided from a virus count over the course of 15 years. Note that the virus count information is reported in log scale.
to:
The spread of HIV in a patient is approximated with balance equations on (H)ealthy, (I)nfected, and (V)irus population counts'^2^'.

 Initial Conditions
 H = healthy cells = 1,000,000
 I = infected cells = 0
 V = virus = 100
 LV = log virus = 2

 Equations
 dH/dt = kr^_1_^ - kr^_2_^ H - kr^_3_^ H V
 dI/dt = kr^_3_^ H V - kr^_4_^ I
 dV/dt = -kr^_3_^ H V - kr^_5_^ V + kr^_6_^ I
 LV = log^_10_^(V)

There are six parameters (kr'_1..6_') in the model that provide the rates of cell death, infection spread, virus replication, and other processes that determine the spread of HIV in the body.

 Parameters
 kr^_1_^ = new healthy cells
 kr^_2_^ = death rate of healthy cells
 kr^_3_^ = healthy cells converting to infected cells
 kr^_4_^ = death rate of infected cells
 kr^_5_^ = death rate of virus
 kr^_6_^ = production of virus by infected cells

The following data is provided from a virus count over the course of 15 years. Note that the virus count information is reported in log scale.
April 30, 2015, at 08:45 PM by 45.56.3.184 -
Added lines 32-35:

(:html:)
<iframe width="560" height="315" src="https://www.youtube.com/embed/0Et07u336Bo?rel=0" frameborder="0" allowfullscreen></iframe>
(:htmlend:)
April 27, 2015, at 07:03 PM by 10.5.113.179 -
Added lines 22-23:

Attach:download.png [[Attach:data_hiv.zip|HIV Data and Model Files]]
April 27, 2015, at 07:01 PM by 10.5.113.179 -
Changed lines 15-18 from:
Although a problem may be written correctly, sometimes the solver fails to find a solution or requires excessive time to find a solution. Initialization strategies are critical in these situations to find a nearby solution that seeds the optimization problem with a starting point that allows convergence.

* Safdarnejad, S.M., Hedengren, J.D., Lewis, N.R., Haseltine, E., Initialization Strategies for Optimization of Dynamic Systems, Computers and Chemical Engineering, DOI: 10.1016/j.compchemeng.2015.04.016. [[http://www.sciencedirect.com/science/article/pii/S0098135415001179 | Article]]

to:
Although a problem may be written correctly, sometimes the solver fails to find a solution or requires excessive time to find a solution. Initialization strategies are critical in these situations to find a nearby solution that seeds the optimization problem with a starting point that allows convergence'^1^'.
Changed lines 19-20 from:

to:
'''Objective:''' Simulate a highly nonlinear system, using initialization strategies to find a suitable approximation for a future parameter estimation exercise. Create a MATLAB or Python script to simulate and display the results. ''Estimated Time: 2 hours''

The spread of HIV in a patient is approximated with balance equations on (H)ealthy, (I)nfected, and (V)irus population counts'^2^'. There are six parameters (kr'_1..6_') in the model that provide the rates of cell death, infection spread, virus replication, and other processes that determine the spread of HIV in the body. The following data is provided from a virus count over the course of 15 years. Note that the virus count information is reported in log scale.

Attach:hiv_virus_count.png

With guess values for parameters (kr'_1..6_'), approximately match the laboratory data for this patient. A subsequent section introduces methods for parameter estimation by minimizing an objective function.

Added lines 29-35:
Attach:download.png [[Attach:simulate_hiv.zip|HIV Simulation in MATLAB and Python]]

!!!! References

# Safdarnejad, S.M., Hedengren, J.D., Lewis, N.R., Haseltine, E., Initialization Strategies for Optimization of Dynamic Systems, Computers and Chemical Engineering, DOI: 10.1016/j.compchemeng.2015.04.016. [[http://www.sciencedirect.com/science/article/pii/S0098135415001179 | Article]]

# Nowak, M. and May, R. M. ''Virus dynamics: mathematical principles of immunology and virology: mathematical principles of immunology and virology''. Oxford university press, 2000.
April 27, 2015, at 06:18 PM by 10.5.113.179 -
Changed lines 5-6 from:
Simulation is a first step in after the model development to verify convergence, validate the model response to input changes, and manually adjust parameters to fit an expected response. This section demonstrates how to set up and solve a simple dynamic model.
to:
Simulation is a first step in after the model development to verify convergence, validate the model response to input changes, and manually adjust parameters to fit an expected response. This section demonstrates how to set up and solve a dynamic model.
Added lines 15-18:
Although a problem may be written correctly, sometimes the solver fails to find a solution or requires excessive time to find a solution. Initialization strategies are critical in these situations to find a nearby solution that seeds the optimization problem with a starting point that allows convergence.

* Safdarnejad, S.M., Hedengren, J.D., Lewis, N.R., Haseltine, E., Initialization Strategies for Optimization of Dynamic Systems, Computers and Chemical Engineering, DOI: 10.1016/j.compchemeng.2015.04.016. [[http://www.sciencedirect.com/science/article/pii/S0098135415001179 | Article]]

Changed lines 21-24 from:
!!!! Solution
to:


!!!! Solution
Changed line 13 from:
(:html:)
to:
(:htmlend:)
Changed line 1 from:
(:title Initialization Strategies:)
to:
(:title Model Initialization Strategies:)
Added lines 4-17:

Simulation is a first step in after the model development to verify convergence, validate the model response to input changes, and manually adjust parameters to fit an expected response. This section demonstrates how to set up and solve a simple dynamic model.

(:html:)
<iframe width="560" height="315" src="https://www.youtube.com/embed/-IDTagajoyA" frameborder="0" allowfullscreen></iframe>
(:htmlend:)

(:html:)
<iframe width="560" height="315" src="https://www.youtube.com/embed/-3FaZEfu7vE" frameborder="0" allowfullscreen></iframe>
(:html:)

!!!! Exercise

!!!! Solution
Changed line 1 from:
(:title Initialization Strategies for Dynamic Systems:)
to:
(:title Initialization Strategies:)
Added lines 1-3:
(:title Initialization Strategies for Dynamic Systems:)
(:keywords initialization, strategy, modeling language, differential, algebraic, tutorial:)
(:description Model initialization strategies for Differential Algebraic Equations (DAEs) with use in dynamic simulation, estimation, and control:)