The applications of prior-knowledge-based differential evolution algorithm in PTA crystallization process

Multi-layer feed forward networks have been proved that they can approximate a wide class of functional relationships very well in modeling chemical processes. Given finite sample data, it is important to encode the prior knowledge in neural networks to improve the fit precision and the prediction ability of model. Differential evolution (DE) algorithm has the advantages of its conceptual simplicity and ease of use, besides its good convergence properties and suitability for parallelity. In this paper, multi-layer feed forward networks and differential evolution algorithm have been analyzed at first. Then prior-knowledge-based differential evolution algorithm (PKDE) is studied for modeling so as to conform the known process mechanism, such as the monotone character, and improve the model precision as well. This method is compared with the conventional back-propagation neural networks when applied to modeling the granularity of pure terephthalic acid (PTA). And simulation results show that the proposed methods are good at modeling complex chemical industrial process.