Abstract
M.Ing.
Eskom's power stations receive their main supply of coal from mines next to the power stations. The coal supply contracts only specify maximum allowable variations of some coal quality parameters. The quality of the supplied coal can, however, vary greatly within a few hours. The boilers in the power plant are optimized for a certain quality of coal, while the supplied coal is burnt as it is received from the mine. The variations in the coal quality can, therefore, have a negative impact on both the life expectancy and maintenance costs of the power plant as well as the controllability of the boiler.
The effects of short term variations in the coal qualities can be reduced by segregating the supplied coal into separate stockpiles according to coal quality parameters such as ash content and volatile matter, and then blending different portions from these stockpiles to a preferred coal quality before the coal goes to the boilers. A self organising feature map neural network was proposed in this research, to determine how to separate the supply coal, according to measured coal quality data. Furthermore, linear programming was proposed to determine the proportions to be taken from each stockpile in order to achieve a more consistent blended coal again. The segregating and blending systems are described in this thesis; and they were tested by means of a simulation based on measured coal quality data from a power station. It was shown that it is possible to successfully segregate coal from a single supply and then blend the different stockpiles to render coal with less short term variations in its quality parameters. The blending process uses stockpile size as its main driver to optimize the selection of the proportions, such that the most coal is taken from the largest stockpile, while the resultant coal quality remains within the specified constraints.