The research topic addressed in this thesis is the development of new ideas and techniques for acceleration and automation of processes involved in the design of chemical products with predefined properties. In particular, we demonstrate techniques that address the shortcomings of the existing methods and take a bird's-eye view over the new possible directions for chemical product development necessitated by the integration of bio-feedstocks into the existing supply chain. Futhermore, we introduce an approach for sequential, on-line multi-target product/process optimization in a scenario where: automation of the overall design process is sought; adequate physical models are not available; unknown constraints on the decision space may be present; and resources are limited or costly. We test the approach on a number of simulations. The results indicate that the approach is able to, in a modest number of iterations, find solutions associated with the targets to a satisfactory degree of accuracy. In addition, for supervised problems where categorical data are available, we introduce an approach that allows one to perform categorization of a given product composition according to a particular property. We test our solutions empirically on real data. The results show that the approach compares well with existing state of the art techniques. We also investigate the application of a variety of nonlinear dimensionality techniques to the visualisation of chemical product data.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:606177 |
| Date | January 2013 |
| Creators | Peremezhney, Nicolai |
| Publisher | University of Warwick |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Source | http://wrap.warwick.ac.uk/62032/ |
Page generated in 0.0036 seconds