Issues in on-line optimisation

In general, on-line optimisation can be defined as the on-line process of finding the optimum set-points of the system. Several areas might be concerned in this procedure. This thesis evaluates algorithms for on-line Optimisation. Techniques for steady-state detection, static data reconciliation, gross error detection and steady-state optimisation are presented and implemented separately and within an on-line optimisation methodology. It has been acknowledged for some time now that the estimation of derivative information is probably the major drawback of the steady-state optimisation technique considered here: the ISOPE algorithm. This thesis investigates the requirements of these derivatives, methods proposed to estimate them, and presents some attempts to overcome some related problems. Also a modified version of the dynamic model identification method that uses a nonlinear model representation is proposed, and compared under simulation with other available techniques. In the same context, an alternative method based on Artificial Neural Networks to estimate the derivatives is also implemented and tested. Often, rigorous steady-state detection is crucial for process performance assessment, simulation, optimisation and control. In general, at steady-state data is collected for safe, beneficial and rational management of processes. A method for automatic detection of steady-state in multivariable processes is implemented and tested. The technique is applied on a dynamic model of a chemical reactor. The presence of errors in process measurements can invalidate the potential gains obtained from advanced optimisation and control techniques. Data reconciliation and gross error detection methods are used to reduce the inaccuracies of these measurements. The implementation and application of static data reconciliation and gross error detection techniques in this thesis show a noticeable improvement in the operation of the system, and general control system performance. The various algorithms mentioned above are successfully implemented and tested under simulation. It is illustrated that in some cases, it is possible to use steady state detection in conjunction with data reconciliation, gross error detection, parameter estimation and optimisation, to form an on-line optimisation methodology. The methodology was tested on a dynamic model of a chemical reactor