Predictive Dynamic Modelling for Next Generation Processing

EPSRC Mathematics Case Project (Ref. Case 2006/034, Sept. 2006—Feb. 2010)

Supervisors: Dr. J Q Shi (Maths & Stats), Prof. A. J Morris (CPACT, SCEAM), Dr. Z Rawi (BP Refining Technology)

CASE research studentship is available now! A phd studentship is available now for 3.5 years: standard minimum EPSRC stipend (£12,300 for 06/07) plus £3,000 pa from BP.

Outline

This project aims to develop novel statistical theory and methodology building on the results already achieved (Kamnik et al, 2005, Shi et al, 2003, 2005 a, b and c) for dynamic non-parametric multi-step-ahead prediction focusing in particular on complex engineering systems that exhibit varying batch data quantities from very large to small, and large dimension of input covariates. The statistical (hybrid) modelling methodologies developed in this project will be applied the modelling of both continuous and batch nonlinear dynamic systems operated by BP. For example, in large production plants detailed mechanistic models are used for real time optimisation with challenging operational issues during change-over between different product grades (recipes) to meet market demands, might also be regarded as a series of batch operations embedded within a continuous operation. A major requirement is to be able to predict future process performance and detect and diagnose potential process malfunctions. A major research challenge is to be able to combine known mechanistic process modelling approaches with data based approaches (process measurement data with analytical data from spectroscopic instruments, and in future image data) within a hybrid mechanistic and Bayesian / Gaussian modelling framework.