Amoss is a generic equation-orientated stochastic simulation platform speci cally designed
to facilitate the development and simulation of stochastic models, using Sasol's
in-house MOSS methodology. In the current situation in Sasol, modelling stochastic processes
with recycle streams (feedbacks) using the MOSS methodology is a laborious task.
To simulate these models with acceptable accuracy and simulation speed the model equations
were derived and ordered manually. This human input leads to long development
times and makes it hard to alter an already created model. Amoss aims to automate the
development of MOSS simulation hence the name automatic-MOSS and is an extension
of the MOSS methodology.
The automation was achieved by automatically creating the bulk of the model equations
and leave only the inputs that make each model unique as user inputs. The aspects
that make each model unique were identi ed as the characteristics of the unit operations,
how these operational units are connected, the heuristic rules that govern how the process
is operated and the stochastic elements. Given these inputs, a stochastic model is
automatically created and ordered to the bordered lower triangular form.
The created models are simulated using Euler's algorithm for integration, coupled with
automatic di erentiation and multidimensional Newton's method to nd the roots of the
system of equations at each time step. The models created using MOSS are embarassingly
parallel, and simulation speed was increased by employing parallel processing to exploit
the decoupled nature of these simulations.
Amoss consists of all the necessary building blocks to create and simulate stochastic
simulations and provides a good platform from which improvements in usability and
expansions to the MOSS methodology can be made.
Dissertation (MEng)--University of Pretoria, 2018.