Date of Award

Winter 12-14-2017

Degree Type


Degree Name

Doctor of Philosophy (PhD)


Computer Science

First Advisor

Xiaolin Hu


Assimilating real-time sensor data into simulations is an effective approach for improving predictive abilities. However, integrating complex simulation models, e.g., discrete event simulation models and agent-based simulation models, is a challenging task. That is because classical data assimilation techniques, such as Kalman Filter, rely on the analytical forms of system transition distribution, which these models do not have. Sequential Monte Carlo methods are a class of most extensively used data assimilation algorithms which recursively estimate system states using Bayesian inference and sampling technique. They are non-parametric filters and thus can work effectively with complex simulation models. Despite of the advantages of Sequential Monte Carlo methods, simulation systems do not automatically fit in data assimilation framework. In most cases, it is a difficult and tedious task to carry out data assimilation for complex simulation models. In addition, Sequential Monte Carlo methods are statistical methods developed by mathematicians while simulation systems are developed by researchers in particular research fields other than math. There is a need to bridge the gap of theory and application and to make it easy to apply SMC methods to simulation applications. This dissertation presents a general framework integrating simulation models and data assimilation, and provides guidance of how to carry out data assimilation for dynamic system simulations. The developed framework formalizes the data assimilation process by defining specifications for both simulation models and data assimilation algorithms. It implements the standard Bootstrap Particle Filtering algorithm and a new \emph{Sensor Informed Particle Filter}, (SenSim) to support effective data assimilation. The developed framework is evaluated based on the application of wildfire spread simulation, and experiment results show the effectiveness of data assimilation. Besides the framework, we also developed an open source software toolkit named as Data Assimilation Framework Toolkit to make it easy for researchers to carry out data assimilation for their own simulation applications. A tutorial example is provided to demonstrate the data assimilation process using this data assimilation toolkit.