AN ADAPTIVE SOFT SENSOR DESIGN BASED ON A NEURAL NETWORK WITH A PREDICTIVE FILTER IN THE FEEDBACK LOOP FOR A REACTION DISTILATION TECHNOLOGICAL PROCESS

Rectification and reactive distillation columns are the main of all units in petrochemical refining industry. Soft sensors consist mathematical models that estimate quality an output product real time used for technological processes control. In general, changes composition raw materials, catalyst deactivation, etc. result inconsistency between obtained data current state process. sensor design on such will loss accuracy estimating necessary parameters product. An adaptive soft based a neural network with predictive filter feedback loop solve mismatching process problem is proposed. “Moving window” conception size window adapting to actual Parameter estimation using matching A improve parameter at cost predicting error designed comparative analysis several networks "moving window" by-product concentration reaction-distillation column effectiveness proposed approach shown. Application allows by 12.94 % (coefficient determination) 39.81 (mean absolute error) comparison without filter.