Home > Research > Publications & Outputs > Methods for enhancing system dynamics modelling

Electronic data

  • 2015bellphd

    Final published version, 2.62 MB, PDF document

    Available under license: CC BY-NC-ND: Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License

View graph of relations

Methods for enhancing system dynamics modelling: state-space models, data-driven structural validation & discrete-event simulation

Research output: ThesisDoctoral Thesis

Published
Publication date2015
Number of pages325
QualificationPhD
Awarding Institution
Supervisors/Advisors
Publisher
  • Lancaster University
<mark>Original language</mark>English

Abstract

System dynamics (SD) simulation models are differential equation models that often contain a complex network of relationships between variables. These models are widely used, but have a number of limitations. SD models cannot represent individual entities, or model the stochastic behaviour of these individuals. In addition, model parameters are often not observable and so values of these are based on expert opinion, rather than being derived directly from historical data. This thesis aims to address these limitations and hence enhance system dynamics modelling. This research is undertaken in the context of SD models from a major telecommunications provider.
In the first part of the thesis we investigate the advantages of adding a discreteevent simulation model to an existing SD model, to form a hybrid model. There are few examples of previous attempts to build models of this type and we therefore provide an account of the approach used and its potential for larger models. Results demonstrate the advantages of the hybrid’s ability to track individuals and represent stochastic variation.
In the second part of the thesis we investigate data-driven methods to validate model assumptions and estimate model parameters from historical data. This commences with use of regression based methods to assess core structural assumptions of the organisation’s SD model. This is a complex, highly nonlinear model used by the organisation for service delivery. We then attempt to estimate the parameters of this model, using a modified version of an existing approach based on state-space modelling and Kalman filtering, known as FIMLOF. One such modification, is the use of the unscented Kalman filter for nonlinear systems. After successfully estimating parameters in simulation studies, we attempt to calibrate the model for 59 geographical regions. Results demonstrate the success of our estimated parameters compared to the organisation’s default parameters in replicating historical data.