## Dynamic Estimation Objectives

## Main.EstimatorObjective History

Hide minor edits - Show changes to markup

The spread of HIV in a patient is approximated with balance equations on (H)ealthy, (I)nfected, and (V)irus population counts^{2}.

The spread of HIV in a patient is approximated with balance equations on (H)ealthy, (I)nfected, and (V)irus population counts^{2}. Additional information on the HIV model is at the Process Dynamic and Control Course.

$$x,u,p=\mathrm{states,\,inputs,\,parameters}$$

$$x,y,p=\mathrm{states,\,outputs,\,parameters}$$

$$\Delta p_U \ge p - \hat p$$

$$\Delta p_L \ge \hat p - p$$

$$\Delta p_U \ge p_i - p_{i-1}$$

$$\Delta p_L \ge p_{i-1} - p_i$$

$$\min_{x,y,p} \Phi = \left(y_x-y\right)^T W_x \left(y_x-y\right) + \left(y-\hat y\right)^T W_m \left(y-\hat y\right) + \Delta p^T c_{\Delta p} \Delta p$$

$$\min_{x,y,p} \Phi = \left(y_x-y\right)^T W_x \left(y_x-y\right) + \left(y-\hat y\right)^T W_m \left(y-\hat y\right) + \Delta p^T W_{\Delta p} \Delta p$$

$$c_{\Delta p}=\mathrm{parameter\,movement\,penalty\,(DCOST)}$$

$$w_{\Delta p},W_{\Delta p}=\mathrm{parameter\,movement\,penalty\,(DCOST)}$$

$$y=\mathrm{\mathrm{model\,predictions}$$

$$y=\mathrm{model\,predictions}$$

$$\Phi = \mathrm{Objective\,Function}$$

$$y_x$$ = measurements

$$y$$ = model predictions

$$\hat y$$ = prior model values

$$w_x, W_x$$ = measurement deviation penalty (WMEAS)

$$w_m, W_m$$ = prior prediction deviation penalty (WMODEL)

$$c_{\Delta p}$$ = parameter movement penalty (DCOST)

$$db$$ = dead-band for noise rejection

$$x,u,p$$ = states, inputs, parameters

$$\Delta p$$ = parameter change

$$f,g$$ = equality and inequality constraints

$$e_U,e_L$$ = upper and lower error outside dead-band

$$c_U,c_L$$ = upper and lower deviation from prior model prediction

$$\Delta p_U,\Delta p_L$$ = upper and lower parameter change

$$\Phi=\mathrm{Objective\,Function}$$

$$y_x=\mathrm{measurements}$$

$$y=\mathrm{\mathrm{model\,predictions}$$

$$\hat y=\mathrm{prior\,model\,values}$$

$$w_x, W_x=\mathrm{measurement\,deviation\,penalty\,(WMEAS)}$$

$$w_m, W_m=\mathrm{prior\,prediction\,deviation\,penalty\,(WMODEL)}$$

$$c_{\Delta p}=\mathrm{parameter\,movement\,penalty\,(DCOST)}$$

$$db=\mathrm{dead-band\,for\,noise\,rejection}$$

$$x,u,p=\mathrm{states,\,inputs,\,parameters}$$

$$\Delta p=\mathrm{parameter\,change}$$

$$f,g=\mathrm{equality\,and\,inequality\,constraints}$$

$$e_U,e_L=\mathrm{upper\,and\,lower\,error\,outside\,dead-band}$$

$$c_U,c_L=\mathrm{upper\,and\,lower\,deviation\,from\,prior\,model\,prediction}$$

$$\Delta p_U,\Delta p_L=\mathrm{upper\,and\,lower\,parameter\,change}$$

$$\Phi = \mathrm{Objective Function}$$

$$\Phi = \mathrm{Objective\,Function}$$

$$\Phi$$ = Objective Function

$$\Phi = \mathrm{Objective Function}$$

$$\Phi$$ = Objective Function

$$y_x$$ = measurements

$$y$$ = model predictions

$$\hat y$$ = prior model values

$$w_x, W_x$$ = measurement deviation penalty (WMEAS)

$$w_m, W_m$$ = prior prediction deviation penalty (WMODEL)

$$c_{\Delta p}$$ = parameter movement penalty (DCOST)

$$db$$ = dead-band for noise rejection

$$x,u,p$$ = states, inputs, parameters

$$\Delta p$$ = parameter change

$$f,g$$ = equality and inequality constraints

$$e_U,e_L$$ = upper and lower error outside dead-band

$$c_U,c_L$$ = upper and lower deviation from prior model prediction

$$\Delta p_U,\Delta p_L$$ = upper and lower parameter change

$$\min_{x,y,p} \Phi = \left(y_x-y\right)^T W_m \left(y_x-y\right) + \Delta p^T c_{\Delta p} \Delta p + \left(y-\hat y\right)^T W_p \left(y-\hat y\right)$$

$$\min_{x,y,p} \Phi = \left(y_x-y\right)^T W_x \left(y_x-y\right) + \left(y-\hat y\right)^T W_m \left(y-\hat y\right) + \Delta p^T c_{\Delta p} \Delta p$$

$$\min_{x,y,p} \Phi = w_m^T \left(e_U+e_L\right) + w_p^T \left(c_U+c_L\right) + w_{\Delta p}^T \left(\Delta p_U+\Delta p_L\right)$$

$$\min_{x,y,p} \Phi = w_x^T \left(e_U+e_L\right) + w_m^T \left(c_U+c_L\right) + w_{\Delta p}^T \left(\Delta p_U+\Delta p_L\right)$$

$$e_U, e_L, c_U, c_L \ge 0$$

$$\Delta p_U \ge p - \hat p$$

$$\Delta p_L \ge \hat p - p$$

$$e_U, e_L, c_U, c_L, \Delta p_U, \Delta p_L \ge 0$$

$$\min_{x,y,p} \Phi = w_m^T \left(e_U+e_L\right) + w_p^T \left(c_U+c_L\right) + \left| \Delta p \right|^T c_{\Delta p}$$

$$\min_{x,y,p} \Phi = w_m^T \left(e_U+e_L\right) + w_p^T \left(c_U+c_L\right) + w_{\Delta p}^T \left(\Delta p_U+\Delta p_L\right)$$

$$\min_{x,y,p} \Phi = w_m^T \left(e_U+e_L\right) + w_p^T \left(c_U+c_L\right) + \Delta p^T c_{\Delta p}$$

$$\min_{x,y,p} \Phi = w_m^T \left(e_U+e_L\right) + w_p^T \left(c_U+c_L\right) + \left| \Delta p \right|^T c_{\Delta p}$$

$$\min_{x,y,p} \Phi = \left(y_x-y\right)^T W_m \left(y_x-y\right) + \Delta p^T c_{\Delta p} + \left(y-\hat y\right)^T W_p \left(y-\hat y\right)$$

$$\min_{x,y,p} \Phi = \left(y_x-y\right)^T W_m \left(y_x-y\right) + \Delta p^T c_{\Delta p} \Delta p + \left(y-\hat y\right)^T W_p \left(y-\hat y\right)$$

$$\min_{x,y,p} \Phi = w_m^T \left(e_U+e_L\right) + w_p^T \left(c_U+c_L\right) + \Delta p^T c_{\Delta p} \Delta p$$

$$\min_{x,y,p} \Phi = w_m^T \left(e_U+e_L\right) + w_p^T \left(c_U+c_L\right) + \Delta p^T c_{\Delta p}$$

$$\min_{x,y,p} \Phi = w_m^T \left(e_U+e_L\right) + w_p^T \left(c_U+c_L\right) + \Delta p^T c_{\Delta p}$$

$$\min_{x,y,p} \Phi = w_m^T \left(e_U+e_L\right) + w_p^T \left(c_U+c_L\right) + \Delta p^T c_{\Delta p} \Delta p$$

$$c_U \ge \hat y - y$$

$$c_L \ge \hat y - y$$

$$e_U \ge y_{pred} - y_{meas} - \frac{db}{2}$$

$$e_L \ge y_{meas} - y_{pred} - \frac{db}{2}$$

$$e_U \ge y - y_x - \frac{db}{2}$$

$$e_L \ge y_x - y - \frac{db}{2}$$

$$\min_{x,y,p,d} \Phi$$

$$\Phi = \left(y_x-y\right)^T W_m \left(y_x-y\right) + \Delta p^T c_{\Delta p} + \left(y-\hat y\right)^T W_p \left(y-\hat y\right)$$

$$\min_{x,y,p} \Phi = \left(y_x-y\right)^T W_m \left(y_x-y\right) + \Delta p^T c_{\Delta p} + \left(y-\hat y\right)^T W_p \left(y-\hat y\right)$$

$$\mathrm{subject\;\;to}$$

$$0 = f\left(\frac{dx}{dt},x,y,p\right)$$

$$0 \le g\left(\frac{dx}{dt},x,y,p\right)$$

$$\mathrm{subject\;\;to}$$

$$0 = f\left(\frac{dx}{dt},x,y,p\right)$$

$$0 \le g\left(\frac{dx}{dt},x,y,p\right)$$

$$e_U \ge y_{pred} - y_{meas} - \frac{db}{2}$$

$$e_L \ge y_{meas} - y_{pred} - \frac{db}{2}$$

$$c_U \ge y - \hat y$$

$$c_U \ge \hat y - y$$

$$e_U, e_L, c_U, c_L \ge 0$$

$$\min_{x,y,p,d} \Phi$$

$$\Phi = \left(y_x-y\right)^T W_m \left(y_x-y\right) + \Delta p^T c_{\Delta p} + \left(y-\hat y\right)^T W_p \left(y-\hat y\right)$$

log_v = data[:,][:,1] # 2nd column of data

log_v = data[:,1] # 2nd column of data

(:toggle hide gekko button show="Show GEKKO (Python) Code":) (:div id=gekko:) (:source lang=python:) from __future__ import division from gekko import GEKKO import numpy as np

- Manually enter guesses for parameters

lkr = [3,np.log10(0.1),np.log10(2e-7), np.log10(0.5),np.log10(5),np.log10(100)]

- Model

m = GEKKO()

- Time

m.time = np.linspace(0,15,61)

- Parameters to estimate

lg10_kr = [m.FV(value=lkr[i]) for i in range(6)]

- Variables

kr = [m.Var() for i in range(6)] H = m.Var(value=1e6) I = m.Var(value=0) V = m.Var(value=1e2)

- Variable to match with data

LV = m.CV(value=2)

- Equations

m.Equations([10**lg10_kr[i]==kr[i] for i in range(6)]) m.Equations([H.dt() == kr[0] - kr[1]*H - kr[2]*H*V,

I.dt() == kr[2]*H*V - kr[3]*I, V.dt() == -kr[2]*H*V - kr[4]*V + kr[5]*I, LV == m.log10(V)])

- Estimation
- Global options

m.options.IMODE = 5 #switch to estimation m.options.TIME_SHIFT = 0 #don't timeshift on new solve m.options.EV_TYPE = 2 #l2 norm m.options.COLDSTART = 2 m.options.SOLVER = 1 m.options.MAX_ITER = 1000

m.solve()

for i in range(6):

lg10_kr[i].STATUS = 1 #Allow optimizer to fit these values lg10_kr[i].DMAX = 2 lg10_kr[i].LOWER = -10 lg10_kr[i].UPPER = 10

- patient virus count data

data = np.array()

- Convert log-scaled data for plotting

log_v = data[:,][:,1] # 2nd column of data v = np.power(10,log_v)

LV.FSTATUS = 1 #receive measurements to fit LV.STATUS = 1 #build objective function to match data and prediction LV.value = log_v #v data

m.solve()

- Plot results

import matplotlib.pyplot as plt plt.figure(1) plt.semilogy(m.time,H,'b-') plt.semilogy(m.time,I,'g:') plt.semilogy(m.time,V,'r--') plt.semilogy(data[:,][:,0],v,'ro') plt.xlabel('Time (yr)') plt.ylabel('States (log scale)') plt.legend(['H','I','V']) plt.show() (:sourceend:) (:divend:)

The l_{1}-norm objective is like an absolute value objective but also includes a dead-band to reject measurement error and stabilize the parameter estimates.

- # 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. Article

- 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. Article

- # 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. Article

(:html:) <iframe width="560" height="315" src="https://www.youtube.com/embed/5qY7WyngRbo" frameborder="0" allowfullscreen></iframe> (:htmlend:)

The squared error objective is the most common form, used extensively in the literature. It is also the basis for the derivation of the Kalman filter and other well known estimators.

The l_{1}-norm objective is like an absolute value of the error but posed in a way to have continuous first and second derivatives. The addition of slack variables enables an efficient formulation (only linear constraints) that is also convex (local optimum is the global optimum). A unique aspect of the following l_{1}-norm objective is the addition of a *dead-band* or region around the measurements where there is no penalty. It is only when the model predictions are outside of this dead-band that the optimizer makes changes to the parameters to correct the model.

There are many symbols used in the definition of the different objective function forms. Below is a nomenclature table that gives a description of each variable and the role in the objective expression.

The following example problem is a demonstration of the two different objective forms and some of the configuration to achieve optimal estimator performance.

- 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. Article

<iframe width="560" height="315" src="https://www.youtube.com/embed/0Et07u336Bo?rel=0" frameborder="0" allowfullscreen></iframe> (:htmlend:)

<iframe width="560" height="315" src="https://www.youtube.com/embed/KuivI_QZ0IA?rel=0" frameborder="0" allowfullscreen></iframe>(:htmlend:)

With guess values for parameters (kr_{1..6}), approximately match the laboratory data for this patient.

With guess values for parameters (kr_{1..6}), approximately match the laboratory data for this patient as an initial solution. Use this initial solution to compute an optimal solution with dynamic estimation. Adjust parameters kr_{1..6} to match the virus count data. Start with different kr values to verify that the solution is not just locally optimal but also globally optimal.

The dynamic estimation objective function is a mathematical statement that is minimized or maximized to find a best solution among all possible feasible solutions. The form of this objective function is critical to give desirable solutions for model predictions but also for other applications that use the output of a dynamic estimation application. Two common objective functions are shown below as squared error and l_{1}-norm forms.

The dynamic estimation objective function is a mathematical statement that is minimized or maximized to find a best solution among all possible feasible solutions. The form of this objective function is critical to give desirable solutions for model predictions but also for other applications that use the output of a dynamic estimation application. Two common objective functions are shown below as squared error and l_{1}-norm forms^{1}.

The spread of HIV in a patient is approximated with balance equations on (H)ealthy, (I)nfected, and (V)irus population counts^{1}.

The spread of HIV in a patient is approximated with balance equations on (H)ealthy, (I)nfected, and (V)irus population counts^{2}.

- Hedengren, J. D. and Asgharzadeh Shishavan, R., Powell, K.M., and Edgar, T.F., Nonlinear Modeling, Estimation and Predictive Control in APMonitor, Computers and Chemical Engineering, Volume 70, pg. 133–148, 2014. Article

The dynamic estimation objective function is a mathematical statement that is minimized or maximized to find a best solution among all possible feasible solutions. The form of this objective function is critical to give desirable solutions for model predictions but also for other applications that use the output of a dynamic estimation application.

The dynamic estimation objective function is a mathematical statement that is minimized or maximized to find a best solution among all possible feasible solutions. The form of this objective function is critical to give desirable solutions for model predictions but also for other applications that use the output of a dynamic estimation application. Two common objective functions are shown below as squared error and l_{1}-norm forms.

#### Squared Error Objective

#### l_{1}-norm Objective

#### Nomenclature

The spread of HIV in a patient is approximated with balance equations on (H)ealthy, (I)nfected, and (V)irus population counts^{2}.

The spread of HIV in a patient is approximated with balance equations on (H)ealthy, (I)nfected, and (V)irus population counts^{1}.

- 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. Article

#### Exercise

**Objective:** Estimate parameters of a highly nonlinear system. Use an initialization strategy to find a suitable approximation for the parameter estimation. Create a MATLAB or Python script to simulate and display the results. *Estimated Time: 2 hours*

^{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.

With guess values for parameters (kr_{1..6}), approximately match the laboratory data for this patient.

#### Solution

(:html:) <iframe width="560" height="315" src="https://www.youtube.com/embed/0Et07u336Bo?rel=0" frameborder="0" allowfullscreen></iframe> (:htmlend:)

#### 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. 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.

(:title Dynamic Estimation Objectives:) (:keywords objective function, MATLAB, Simulink, moving horizon, time window, dynamic data, validation, estimation, differential, algebraic, tutorial:) (:description Objective function forms for improved rejection of corrupted data with outliers, drift, and noise:)

The dynamic estimation objective function is a mathematical statement that is minimized or maximized to find a best solution among all possible feasible solutions. The form of this objective function is critical to give desirable solutions for model predictions but also for other applications that use the output of a dynamic estimation application.