engleski

Development of soft sensors for refinery advanced process control

Industrial facilities show an increasing need for continuous measurements, monitoring and controlling a large number of process variables, strict process and products standards and advanced process control. On-line analyzers are the vital measuring devices in today’s industry. They are frequently unavailable or malfunction. Laboratory assays are irregular and therefore not suitable for process control. Inevitable alternative for unavailable on-line analysis, irregular laboratory assays and more effective process control are soft sensors and inferential control. Based on the plant history data detailed data analysis and preconditioning is carried out. Soft sensor models are developed and the model structures are optimized using global optimization methods. Linear and nonlinear models are identified using finite impulse response (FIR), auto-regressive models with exogenous inputs (ARX), output error (OE) models, nonlinear FIR, nonlinear ARX and Hammerstein-Wiener (HW) models. Global differential evolution (DE) and simulated annealing (SA) methods are used for the model regressor number optimization. Developed models are evaluated based on the model validation criteria, residual analysis and correlation test. The models are tested in refinery production for toluene content estimation with the purpose to design aromatics complex advanced control. Overall results indicate that the application of DE and SA as global optimization methods is suitable for the regressor number estimation of polynomial dynamical models with multiple inputs, especially in a case of very large optimization pool. This makes the development of soft sensors easier and more systematic. By applying proposed research in the plant effective monitoring, diagnostics and advanced process control are expected.

aromatics complex; differential evolution; simulated annealing; soft sensor; system identification

nije evidentirano