Neural networks for on-line prediction and optimization of circulating fluidized bed process steps

In this paper, a methodology is presented to use neural networks for on-line prediction and optimization of two CFB processes. Neural networks are trained from experimental data to correlate process outputs with inputs, and Multiplier and Lagrangian Methods are used to optimize operation of the processes using these trained models, with reasonably good results. The time needed for neural network training and execution, as well as optimization, is acceptable for on-line purposes. Some discussions and experiences are provided for the appropriate use of neural networks in practical engineering, specifically on how to get the data for neural network training and validation.