MODELING OF A FIXED-BED REACTOR USING THE K-L EXPANSION AND NEURAL NETWORKS

摘要

Karhunen-Loeve expansion and feedforward neural networks are combined together in modeling a wall cooled fixed-bed reactor for its on-line performance prediction. The K-L expansion is employed as a preprocessor of neural networks while the latter is used to generate the coefficients of the K-L expansion. The performance of the KL-NN model is investigated by both experimentation and simulation with benzene oxidation as a working system. It is shown that the method is effective for on-lint prediction of the bed temperatures. Our conclusions are more important than just that one term can be used. Sometimes it might be two or three, but the method described in the paper is still powerful. [References: 17]