engleski

On-line optimisation of integrated chemical plants

Chemical processes are complex systems with interconnected mass and energy balances and high degrees of freedom enabling optimisation. When mathematical models and parameters are known, the optimisation task becomes a mathematical problem for which nowadays are available robust methods and software. For optimisation of an industrial plant under production conditions a process engineer has to deal with numerous uncertainties stemming from imprecise models and errors in parameter estimation. Usually in this case a primary goal is optimisation of a steady state, i.e. a working point, and its stabilisation by a control system. Steady state models can be approximated with simple response surfaces and optimum is found by analytical evaluation. The same procedure is repeated around a new steady state. However, the method requires for a process to be conducted through a series of steady states which is usually time consuming, leads to long optimisation procedure and disables optimisation under continuous process perturba-tions. In order to cope with perturbations on-line dynamic mod-elling techniques are developed. Such models may have simple input-output structure defined with linear discrete relations or complex nonlinear structures based on artificial intelligence methods. The aim of this work is to study application of adaptive linear models for optimisation of nonlinear dynamic processes such as are typically used in chemical and biochemical reaction engineering. Optimised is a system Fig. 1. of two nonisothermal chemical reactors interconnected in series with a recycle loop. The recursive least square estimation algorithm for adaptive pa-rameter estimation is applied. Tested are various linear discrete input-output models used for determination of local gradient which is adapted in time due to process dynamics and perturba-tions of input variables. Compared are gradients determined by pseudo steady state assumption and the dynamic gradient tracking method. Enhancement of stability and accuracy of RLS by harmonic and PRBS perturbations is studied.

process optimisation; adaptive control; RLS parameter estimation

nije evidentirano