Process optimization by flowsheet simulation

TY - THES

T1 - Process optimization by flowsheet simulation

T2 - Dissertation

AU - Kaijaluoto, Sakari

PY - 1984

Y1 - 1984

N2 - In the work an algorithm for process optimization using flowsheet simulation has been developed.The algorithm is designed for use in procedure oriented simulation environment and it can be easily adapted to existing flowsheeting programs.The algorithm is based on a successive quadratic programming algorithm by Powell.Some enhancements have been added to the Powell algorithm to make it less sensitive to the scaling of the variables and the objective.The main emphasis in the work has been on the development of schemes for numerical evaluation of gradients required be the Powell algorithm.During the numerical testing of the algorithm development it was noted that the noise generated by the iterative unit procedures can, especially in large processes, seriously affect the accuracy of the derivatives calculated, if the perturbation factors are not properly selected. The problem was solved by developing an algorithm that can automatically select suitable perturbation factors for the iteration variables.The algorithm is based on the balancing of noise-induced and truncation errors.A novel forward difference algorithm has also been developed where the effect of truncation error has been reduced by introducing correction factors, based on approximated second derivatives.The correction factors, which are re-evaluated periodically, are calculated using the general difference formula.The optimization algorithm has been tested using realistic example problems.The performance of the algorithm has been adequate.In most cases the number of simulation time equivalents required to reach the optimum has been no more than ten.

AB - In the work an algorithm for process optimization using flowsheet simulation has been developed.The algorithm is designed for use in procedure oriented simulation environment and it can be easily adapted to existing flowsheeting programs.The algorithm is based on a successive quadratic programming algorithm by Powell.Some enhancements have been added to the Powell algorithm to make it less sensitive to the scaling of the variables and the objective.The main emphasis in the work has been on the development of schemes for numerical evaluation of gradients required be the Powell algorithm.During the numerical testing of the algorithm development it was noted that the noise generated by the iterative unit procedures can, especially in large processes, seriously affect the accuracy of the derivatives calculated, if the perturbation factors are not properly selected. The problem was solved by developing an algorithm that can automatically select suitable perturbation factors for the iteration variables.The algorithm is based on the balancing of noise-induced and truncation errors.A novel forward difference algorithm has also been developed where the effect of truncation error has been reduced by introducing correction factors, based on approximated second derivatives.The correction factors, which are re-evaluated periodically, are calculated using the general difference formula.The optimization algorithm has been tested using realistic example problems.The performance of the algorithm has been adequate.In most cases the number of simulation time equivalents required to reach the optimum has been no more than ten.

KW - optimization

KW - simulation

KW - process charting

M3 - Dissertation

SN - 951-38-2136-6

T3 - Publications / Technical Research Centre of Finland

PB - VTT Technical Research Centre of Finland

CY - Espoo

ER -