COMBINED SMITH PREDICTION AND RECURSIVE FILTERING FOR OBSERVATION OF A POLYETHYLENE REACTOR

At present, the gas composition in a polyethylene reactor is only avai lable from a gas chromatograph with a 6-minute update. In this study, a mixed flow reactor model was developed to simulate the process in re al time, using measurements of the reaction conditions and feed rates. Initial open-loop tests revealed that the prediction was in advance o f the GC output by 450 seconds. A Smith predictor was then employed to establish the error between appropriately delayed composition predict ions and the corresponding measurements, and to correct current predic tions using this offset. Combination of this passive offset with recur sive filtering of predicted compositions, similar to Kalman filtering, proved advantageous. Dual-rate sampling provided fast model responses to the process inputs within the GC cycle period. (C) 1997 Elsevier S cience Ltd.