Neural network based model reference adaptive control (NMRAC) for steam temperature regulation in steam distillation process

This paper presents a neural network controller in model reference adaptive controller (MRAC) for steam temperature regulation in steam distillation process. Steam temperature is one of the most significant parameters that can influence the composition of essential oil yield. Due to parameter variations and changes in operational conditions during distillation, a robust steam temperature controller becomes nontrivial to avoid the degradation of essential oil quality. Initially, the PRBS input is fed to the system and output of steam temperature is collected and modeled using ARX model. Model reference for adaptive control is designed using first order plus dead time (FOPDT) transfer function. The weights of neural network are updated using Recursive Prediction Error Method (RPEM). The performance of neural network is compared using different number of hidden layer, momentum rate and learning rate. The experimental result demonstrates the proposed neural network using 9 hidden nodes achieves a better control to the system.