Enhanced Run-Time Experiments for Continuous System Simulation Languages

François E. Cellier
Institute of Computational Science
ETH Zürich
CH-8092 Zürich
Switzerland
Email: FCellier@Inf.ETHZ.CH

Abstract

In recent years, CSSLs have seen a drastic improvement of their capabilities. Most of these enhancements are concerned with:

modeling: enhanced power of model building and model structuring facilities including specification of submodels, model bases (data bases to store models and submodels), and proper discontinuity handling (combined continuous/discrete simulation)
compilation: enhanced compilation techniques including separate compilation of submodels, and direct executing languages (faster response in an interactive execution environment)
execution: enhanced integration techniques including new algorithms for stiff systems, highly oscillatory systems, linear systems, and techniques for state-space partitioning (faster execution of large-scale systems).

However, very little has happened with respect to the flexibility in experimenting with the simulation model. Basically, only two "experiments" are available in the classical CSSL:

simulation from time 0 to time t_f (expressed by: FINTIM, TMAX, ...)
simulation from time 0 until something happens (expressed by: FINISH, TERMINATE, ...).

This is not what we like to find. Ideally seen, we should be able to design our experiment entirely separate from the model itself. With the modeling stage already completed, we would like to decide on the experiment to be performed, such as the nature of the variables to be monitored during simulation, evaluation of steady-state points, derivation of linearized models and their eigenvalues, execution of complete sensitivity analyses, or replication analyses, etc..

DARE-INTERACTIVE is a new dialect within the DARE family of simulation languages that has been designed and implemented as a testbed for enhanced and more intelligent run-time experiments. With one single command, it is already possible to perform a complete range analysi (sensitivity analysis for nonlinear models), or a complete replication analysis. The graphics postprocessor allows to view families of trajectories either by use of envelope graphics or three-dimensional graphics with hidden lines removed. Envelope graphics allow to view the range of a variable as a function of parameter uncertainties, instead of a particular trajectory, displayed with fourteen digits accuracy out of which possibly not a single one may be significant (!) Other experiments such as automated steady-state finding, and curve fitting are currently under development.

In this paper, we wish to summarize what possibilities exist with respect to enhancing the capabilities of simulation run-time experimentation. At the same time, we shall show practical examples of implementation of some of these ideas in the DARE-INTERACTIVE software system.

Interested in reading the full paper? (6 pages, 504,340 bytes, pdf)

Homepage