Neural network models of the combustion derivatives emanating from a chain grate stoker fired boiler plant

This paper is concerned with the use of neural networks as a means to represent the formation of pollutant emissions, resulting from the combustion of coal with a 1 MW_th chain grate stoker test facility located at CRE Group Ltd. (CRE), Stoke Orchard, Cheltenham. The resultant `black-box´ models of the pollutant emissions, namely the nitrogen oxides and carbon monoxide emissions, were able to represent the dynamics of the process and delivered reasonably accurate estimates over a wide range of unseen data. This system identification approach is in many ways a simplistic approach to that of the more daunting route of mathematically modelling the physical processes. The neural network model, although lacking in model transparency and elegance, is able to produce estimates of the derivatives of combustion with acceptable accuracy considering its relative simple model design. This has been demonstrated not only with data sets that were obtained from the same series of experiments but also from data with a temporal separation of 8 months from the training data. The work presented here addresses some of the deficiencies in the modelling of lump coal combustion on grates reported in the literature and the method adopted in this paper could be used to develop a `software sensor´ that when linked to a combustion control system could help to minimise pollutant emissions of chain or travelling grate stokers