D.2 Real-time optimising control

Project Details


Issues in real time optimisation are relevant in many instances of process operation. In this project two instances are investigated. One is in scheduling plant operations in accord with market demands and raw material availability thus providing set points for the control loops. The other deals with operating semi-continuous or fed-batch processes where the feeding profiles are determined to optimise the process productivity under the loads of changing raw material and catalyst properties.
A methodology for simultaneous dynamic optimisation of gasoline production and blending in response to market demands of a refinery is being developed. The optimisation decides the set points for the refinery operations as well as the blending recipe such that profit is maximised several time periods into the future. As such the optimisation program has strong similarities to the receding horizon controller employed in model predictive control. The main model requirements are that the gasoline blending model is piecewise linear and that reactors and fractionators in the gasoline production plant are restricted to a finite set of possible operation points. Using these simplifying but realistic assumptions the model can be formulated as a large-scale MILP model. This model is solved using CPLEX. The actual refinery operation profitability of this approach remains to be explored.
In control of semi-continuous or fed-batch processes the first engineering principle models are time varying and non-linear thus the resulting control problem becomes a rather complicated non-linear optimisation which however often can be solved within reasonable computational time. In this project optimising control of filamentous fungi cultivations and of Saccharomyces cerevisiae cultivations will be investigated. The applied cultivation models will be structured models for the organisms.
Effective start/end date01/05/1997 → …


Explore the research topics touched on by this project. These labels are generated based on the underlying awards/grants. Together they form a unique fingerprint.