Have you ever wondered how well your system is performing or what needs to be tuned to improve its performance? Are you close to the limits of performance for your system? This blog post provides an overview of the types of optimizations that are available for you and your Modelica models.
Modeling projects often aim to simulate the system as accurately as possible before it actually has been built. This saves you both time and money.
In my opinion, however, simulations should show us more than what the system achieves; they should show what we actually could achieve when the system is operated at its limits of performance. State-of-the-art tools by Modelon for this type of analysis are available for you to use.
The types of optimization we can perform can be divided into three different categories, all of which are highly relevant for a system developer and modeler.
1. Parameter optimization is where you find optimal values for one or several system parameters such that a desired property is maximized or minimized. Examples of this could be:
a. Control system reference values at different operating points - minimizing, for example, energy consumption or usage of raw material or maximizing a production rate.
b. Component sizing - making sure that you are not over- or under-sizing your system while respecting operational constraints. This could be piping dimensions versus system weight or heat exchanger tube lengths versus economic cost and cooling capacity.
2. Trajectory optimization is the art of selecting your plant inputs, most often your control signals, over a time horizon such that your plant achieves your goals while respecting its constraints. Examples are:
a. Plant start-up trajectory generation - minimizing start-up times of plants while holding constraints on temperatures, pressures and flows.
b. Grade change optimization in production lines - saving money by reducing the number of products that don't meet specifications.
c. Production planning in district heating networks - making efficient use of environmental-friendly heat production units and avoiding unnecessary starts of oil boilers, for example.
3. Parameter estimation using data, also called model calibration, finds values of unknown parameters in your system model by using collected data from the physical system. Unlike the other optimization types, parameter estimation is performed when the system to be modeled is already built, at least partially.
Per-Ola Larsson holds a Ph.D. in Automatic Control and an M.Sc. in Electrical Engineering, both from Lund University and with focus on optimization, process control and signal processing. At Modelon he is a simulation consultant and project manager in several consultancy and research projects within the field of thermodynamics, containing both steady-state and dynamic models as well as several types of optimizations.