✍ Scribed by Miloslav Hajek; M.Robin Judd
No coin nor oath required. For personal study only.
The analysers used for the determination of CO, COz, Hz and O2 in gas from coal gasifiers may not be completely selective, so for process control it is necessary to predict the interferences or interactions by calibration and modelling. The use of a neural network to model these interactions is compared with standard linear or non-linear models. Only a small number (seven) of standard gas mixtures were used in this comparison, since the objective was to evaluate a neural network model under industrial conditions, where cost and time never permit an extensive calibration study. For this limited situation it is shown that a neural network with a carefully chosen architecture can be significantly better than standard models.
📜 SIMILAR VOLUMES