Neural networks for process analysis and optimization: Modeling and applications

Abstract

This paper describes the development and application of neural networks for analysis and optimization of industrial production data. Artificial neural networks based on a feedforward architecture and trained by the backpropagation technique were applied to analysis and improvement of a separation process. Various neural network topologies have been tested and compared. The Powell method was used to train the network by minimizing the sum of squares of residuals as well as the generalized delta rule (GDR) algorithm. The performance of the network have been analyzed. The obtained results show the applicability the neural networks structure with and without hidden layers. These results illustrate the feasibility of using a neural network as a data analyzer and as an optimization tool.