RTO: An Overview and Assessment of Current Prac-Tice

1. Real-time optimization (RTO) implements business decisions in real time based on a calibrated nonlinear steady state model to maximize plant profit. 2. RTO model is updated at each cycle to minimize the deviation between model variables and plant measurements. 3. There is potential for conflict between RTO and model predictive control (MPC) due to differences in their objective functions and models (nonlinear for RTO vs linear for MPC).

Uploaded by

Arvind R ch15d401

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

30 views

RTO: An Overview and Assessment of Current Prac-Tice

Uploaded by

Arvind R ch15d401

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 3

Literature review R.

Arvind

RTO: An overview and assessment of current prac-

tice
1. Plant decision hierarchy
planning(weeks) ⇒ scheduling(days) ⇒ RTO(hours) ⇒ MPC(mins) ⇒ Distributed
control systems(secs)

2. Varying spatial time scales from plant-wide optimization to regulatory control

via single PID loops.

3. Planning is based on economics and forecasts and deals with feedstock purchase,
desired output/product, quantity to be produced to maximize profit.

4. Scheduling deals with timing of delivery of feed stocks, product liftings and ideal
time scale for operational mode changes targeted to avoid inventory related issues.

5. RTO implements business decisions in real time based on a calibrated nonlinear

steady state model with details that are not available in planning model. Typically
formulated based on a profit function of a plant.

6. Rate at which RTO can be implemented depends on the frequency of unmea-

sured disturbances and the time required for the MPC to move the process to a new
steady state.

7. RTO model is updated(calibrated) at each cycle to minimize the deviation be-

tween model variables and plant measurements.
8. Nonlinear MPC can theoritically allow RTO and MPC to be combined in a single
layer.

9. Since RTO and MPC have their own objective function, there is a potential for
conflict arising due to differences in model (Ex. nonlinear RTO vs linear MPC).

10. RTO operation is not required when optimal operation lies at constraints that
can be consistently determined by MPC. If the optimum lies at constraints, then con-
straint set could possibly see a change which is determined by RTO.

11. If variability is minimal, the optimum policy can be determined offline and RTO
becomes redundant.

12. Trade off between a rigorous model and their contribution in long run is con-
sidered over a simplified model.

Page 1
Literature review R.Arvind

13. Model update :

Z - vector of noisy measurements

X - model variable
x = xm (has measurement value) xu (does not have any measurement value and ap-
pears only in the model)
α - fixed model parameters
β - updated model parameters

min aT V −1 a s.t f(z+a,xu , α, β) = 0

α,β,xu
f - vector of model equations
a - measurements adjustments/biases
V - covariance matrix of measurement errors

14. Economic optimization.

P - profit expression of RTO

max P(x, α, β) s.t f(x, α, β)

x
max (xLB , x − δ) ≤ x ≤ min (xU B , x + δ)
min/max limits of optimization variables

δ - max change in the optimizer value (from the current value) that x is allowed
to move in a single step
Z opt = xm
opt − a

15. MPC design issues - which optimizer decisions to be passed to the MPC and
which d.o.f should be left to the MPC.

16. RTO can also set lower and upper limits of manipulated or controlled value
of MPC.

17. RTO challenges - performance of MPC is critical.

18. RTO pushes the operation to a different operating region, therefore the MPC
linear model needs updation or a linearizing transformation.

19. Good estimates of initial input parameters are critical for RTO (Ex. feed compo-
sition).

Page 2
Literature review R.Arvind

20. Addressing the issue (in model) of number of components to be tracked by RTO.

21. Consistency and coordination are required for setting of economic values, identi-
fying limiting plant constraints, model consistency.

22. Instrument accuracy is of greater importance in a RTO project.

Model consistency issue between MPC and steady state RTO models.

23. Many organizations use RTO offline to address certain process issues.

24. Rigorous offline model can be used as reference model for development and vali-
dation of simplified model for online use.
Linear dynamic models have been used to reduce steady state wait time.
Hybrid/greybox model - combines experimental process data with known or approxi-
mated nonlinear structure.

Page 3