Dynamic soft-sensing model by combining diagonal recurrent neural network with levinson predictor

Dynamic soft-sensing model of diesel oil solidifying point (DOSP) in crude distillation unit (CDU) is proposed based on diagonal recurrent neural network (DRNN). Because of long time-delay of the DOSP measurements, multi-step-ahead predictions are obtained recursively by Levinson predictor and then used as input of DRNN. Simulation results on the actual industrial process data show that the proposed dynamic soft-sensing model took good effects practically and significantly diminished the time-delay of output value.