| This thesis proposes an algorithm for the on-line optimization of large-scale nonlinear dynamic systems. It combines the concepts of moving time horizon, augmented state estimation and discrete modelling, thus allowing formulation of the problem as a mathematical programming problem.; The moving time horizon reduces the effect of disturbances and model inaccuracy, and the augmented state estimator is used (in conjunction with the moving time horizon) in order to obtain a good approximation of the system state (at the time the computation of the control path is performed) and to account for nonstationary disturbances. Discretizing time allows the use of already developed and well tested optimization algorithms and the possibility of changing the problem (the objective function, the model and/or the constraints) with relative ease.; The algorithm's effectiveness was tested by using it to improve the operation of a commercially important system. The operation of these systems--pipeline networks for natural gas transmission--is a challenge for any control algorithm because pipeline networks have large dimensions, the gas flow is modelled by nonlinear hyperbolic partial differential equations, and the operation is subject to constraints on the manipulated, state, and output variables. It is mainly for these reasons that the published results on the control of these systems are restricted to small and very simplified examples.; The successful application of the proposed algorithm to pipeline networks of significant dimension exhibits its strength. This method also has the potential to achieve in a straightforward manner the optimal operation of the system when failure occurs in a compressor station. |
