Assessment of NNARX structure as a global model for self-refilling steam distillation essential oil extraction system

Abstract

This paper investigates the performance of neural network autoregressive with exogenous input (NNARX) model structure and evaluates the training data that provide robust model on fresh data set. The system under test is a self-refilling steam distillation essential oil extraction system. Two PRBS signals with different probability band were tested at different operating points and conditions. A total of three data sets will be used to evaluate the model. NNARX model was estimated by means of prediction error method with Levenberg-Marquardt algorithm. It is expected that the training data that covers the full operating condition will be the optimum training data. All data are separated into training and testing data by interlacing technique. For each data, the model order selection is based on ARX structure and MDL information criterion. These data are cross-validated between each other and the validation results are presented and concluded. The model performance is based on the R2, adjusted-R2, RMSE and NMSE. The histogram is also used to evaluate the distribution of the one-step-ahead residuals. Overall results have shown that the NNARX model trained with data of full operating condition is the most robust when it is validated on a fresh data set.