Quantifying and reducing Input modelling error in simulation

This thesis presents new methodology in the field of quantifying and reducing input modelling error in computer simulation. Input modelling error is the uncertainty in the output of a simulation that propagates from the errors in the input models used to drive it. When the input models are estimated from observations of the real-world system input modelling error will always arise as only a finite number of observations can ever be collected. Input modelling error can be broken down into two components: variance, known in the literature as input uncertainty; and bias. In this thesis new methodology is contributed for the quantification of both of these sources of error. To date research into input modelling error has been focused on quantifying the input uncertainty (IU) variance. In this thesis current IU quantification techniques for simulation models with time homogeneous inputs are extended to simulation models with nonstationary input processes. Unlike the IU variance, the bias caused by input modelling has, until now, been virtually ignored. This thesis provides the first method for quantifying bias caused by input modelling. Also presented is a bias detection test for identifying, with controlled power, a bias due to input modelling of a size that would be concerning to a practitioner. The final contribution of this thesis is a spline-based arrival process model. By utilising a highly flexible spline representation, the error in the input model is reduced; it is believed that this will also reduce the input modelling error that passes to the simulation output. The methods described in this thesis are not available in the current literature and can be used in a wide range of simulation contexts for quantifying input modelling error and modelling input processes