DRIFT ANALYSIS ON NEURAL NETWORK MODEL OF HEAT EXCHANGER FOULING

Neural Networks (NN) provide a good platform for modeling complex andpoorly understood systems in many different fields. Due to the empirical natureof NN, it is typically valid only for small operating windows. As the processdrifts, the prediction accuracy of such models deteriorates very much renderingthe models unfit. An on-line mechanism to follow the drift in the process isnecessary in order to retrain the NN models. Information Criteria have beenreported to be used for the selection of relevant input variables anddetermination of optimal NN model structures. This paper proposes the use ofinformation criteria for tracking the model prediction accuracy and provides analgorithm for retraining the model. A heat exchanger in a refinery CrudePreheat Train (CPT) has been used as a case study. The operational problems ofheat exchangers in CPT are compounded by the varying nature of crude blendsand the complex fouling phenomenon. Fouling develops slowly and thereforethe drift in the process occurs on a slower scale. The performance of a NNfouling model, developed using industrial data is investigated for drift. Modelperformance at different operating conditions is evaluated and it has beenshown that drifts do occur in the process. An algorithm for retraining NN modelhas been proposed.