Optimisation techniques for advanced process supervision and control
This thesis is concerned with the use and development of optimisation techniques for process supervision and control. Two major areas related to optimisation are combined namely model predictive control and dynamic data reconciliation. A model predictive control scheme is implemented and used to simulate the control of a coal gasification plant. Static as well as dynamic data reconciliation techniques are developed and used in conjunction with steady-state optimisation and model predictive control schemes. The inaccuracy of process data due to measurement errors can be considerably reduced by data reconciliation techniques. This in turn improves process knowledge and control system performance. The static and dynamic data reconciliation techniques developed in this thesis are tested using dynamic models of process plants. In the steady-state case, a static data reconciliation algorithm that uses a static model of the process is implemented. This algorithm has capabilities of estimating measured variables, unmeasured variables, systematic bias and unknown physical parameters. The technique is applied to static optimisation to show the improvements in performance of the optimiser when using reconciled data. In order for static data reconciliation to be applied, it is necessary to employ a steady-state detection scheme since the underlying assumption is that the process is at steady-state. An algorithm for steady-state detection is implemented and tested in conjunction with the static data reconciliation technique. In the dynamic case, a moving horizon estimator that employs a dynamic model of the process is used to reconcile dynamic process data. An algorithm for the detection, identification and elimination of gross errors is implemented and tested. Furthermore, an algorithm for the detection and identification of systematic bias is developed and implemented. These techniques are then applied in combination to the dynamic model of a process. The effect of dynamic data reconciliation on the performance of model predictive control is observed by means of applying the above techniques to such a scheme. The various algorithms outlined above are implemented in software and tested using appropriate simulations. It is shown that it is possible to implement a steady-state detection algorithm and to successfully use it in conjunction with static data reconciliation. The application of static data reconciliation to steadystate optimisation shows a marked improvement in the performance of the optimiser. It is further shown that it is possible to combine bias and gross error detection and identification algorithms and to successfully apply them to dynamic data reconciliation procedures. The application of dynamic data reconciliation techniques to model predictive control shows improvement in the performance in cases where the objective is not purely economic.