EVALUATION OF ANNS AND REGRESSION METHODS FOR SULPHUR RECOVERY UNIT AND OTHER REFINERY APPLICATIONS

Abstract

Three prediction methods namely linear regression, non-linear regression and neural networks were used to build prediction models for the tail gas quality of a sulphur recovery unit. It was found that the neural networks give the most accurate prediction and linear regression the least accurate. Generally, the accuracy of the neural network model improves with a more complicated architecture, that is, a network having more hidden layers and nodes. However, building a neural network model is tedious and time consuming. One of the reasons is the lack of scientific rules regarding the architecture of the network and thus has to be found by trial-and-error. Furthermore, the neural network model takes significantly more memory and computational resources compared to the other two methods. It was shown that if these problems can be circumvented, the neural network method is good for the above refinery applications.