Divide and conquer strategies in large scale dynamic process simulation

Abstract

The homogeneous modeling of dynamic processes in chemical plants by large systems of differential-algebraic equations and their parallelized numerical solution on multiprocessor systems has been proven to be an effective way to handle large-scale process models efficiently but requires a substantial effort in the case of model extensions. In contrast to it the heterogeneous modeling approach distributes the solution of submodels of interconnected processes to a computer network. The objective is the independent treatment of each submodel within a time interval with its most appropriate numerical method and the subsequent iterative matching of the interconnecting variables of the over-all process flowsheet. This strategy of distributed simulation is investigated in the case that the submodels do not provide input-output sensitivities for solving the connecting equations between submodels.