Multivariable correlation analysis with low sampling rate in output and its application in an LNG plant

Abstract

Multivariable correlation analysis (MVCA) is a powerful tool in the field of system identification, especially where only normal operational data with simultaneous disturbance and input variations, all with some degree of correlation, are present. This paper presents the application of a MVCA technique for the prediction of C3 concentration in the outlet stream of the deethanizer tower (outlet stream of a liquefied gas natural plant). This is a variable whose prediction and inference is of great importance in the correct control of the unit. Good results are obtained with a non parametric MISO (Multiple Input Single Output) model (impulse response) provided by the multivariable correlation technique using real data from a commercial plant. This paper also suggests a procedure to handle output data whose sampling rate is lower than input sampling.