engleski

Soft Sensor Model Optimization for Continuous Toluene Estimation

From industrial facilities increasing need for continuous measurements of product properties and optimal process control are expected. This imposes the need for monitoring a large number of process variables using on-line analyzers. As an alternative for on-line analyzers soft sensors for toluene content estimation are developed in order to improve the control of the aromatic complex. Based on real-plant data extensive data analysis and preprocessing is carried out, and several type of the soft sensors are modeled: Finite Impulse Response (FIR), AutoRegressive model with eXogenous inputs (ARX) and Output Error (OE). The model structures (number of regressors) are optimized using global Simulated Annealing (SA) method. The models were evaluated based on Root Mean Square Error (RMSE), Mean Absolute Error (eMAE) and FIT criteria. Developed models are tested and validated on the real plant data. Obtained soft sensors serve as product quality continuous estimator as an alternative for on-line analyzers or laboratory assays.

soft sensor; identification; optimization; simulated annealing

Rad je objavljen i u Knjizi sažetaka ; str. 216-216.