Search

Multiple air separation units feed a compression network which supplies a major steelmaker with oxygen gas at two pressures. The project aim was to minimize the network running costs whilst reducing product spill and liquid backup usage during steady state and whilst reconfiguring the network. Previous work has modeled the supply network and minimized power consumption using a mixed integer nonlinear programming (MINLP) approach. This paper develops an MILP framework for improved solution speed and further application. Losses during transition between network configurations, such as liquid vaporization and product spill, were reduced by developing a multi-period optimizer and improving site control. The optimizer, when positioned within a production controller, will communicate set points to linear model predictive control interfaces and monitor flows to prompt operators when to alter the network. Coupled further with a scheduler for automated ramping of production units and load management of product liquefiers subject to power price variation, this paper suggests a control hierarchy where profit optimization is possible. [ABSTRACT FROM AUTHOR]

The inference of mathematical descriptions of (bio)chemical reaction networks from experimental data is a challenging problem whose solution would have significant commercial and academic impact. In this paper, integer linear programming (ILP) is used to develop a generic framework to recover sets of reactions comprising a chemical reaction network. Network search using ILP promotes sparse connectivity, incorporates elemental mass conservation and integrates both heuristic and experimentally identified structural properties using linear equality and inequality constraints. Two case studies are used to demonstrate the discovery of plausible chemical reaction network structures using ILP, the validity of which are verified by kinetic model identification studies. [ABSTRACT FROM AUTHOR]

A Genetic Algorithm (GA) is used to design an automated system that utilizes concentration profiles of measured chemical species obtained from experiments in a batch reactor to construct chemical reaction network (CRN). Ordinary differential equation (ODE) models are constructed to represent the rate of production / consumption of the chemical species that participate in the network of reactions and through the ODEs, the CRN is identified. This is achieved through a two level optimization approach. The first level instantiates the network structure using a GA to evolve network stoichiometry while the second level optimizes the kinetic rate constants associated with each reaction in each candidate network. The algorithm is designed to cope with unmeasured data and the diversity of potential CRNs being discovered is enhanced through the use of the multi-objective optimization algorithm. The ability to discover a set of plausible CRNs is demonstrated using experimental data from a reaction between trimethyl orthoacetate and allyl alcohol. [ABSTRACT FROM AUTHOR]