Estimation of Calorific Value and Grindability of Colombian Caribbean Coals by Multiple Regression and Artificial Neural Networks
- Type: Conference Presentation
- Conference Type:
AIChE Spring Meeting and Global Congress on Process Safety
- Presentation Date:
April 2, 2014
- Skill Level:
Coal is one of the most important raw materials in the world, it is used in a diverse applications such as water treatment, steel production and molecular sieves; however coal is mainly used as fuel in power generation plants. In the power industry the coal must have specific properties to make an affordable, efficient and environmentally responsible generation process, therefore large number of instrumental analysis are required to determine the value of those properties, spending time, equipment and qualified personal. Between many properties measured, there are two that are fundamental for combustion and grinding processes: Gross calorific value and Hardgrove Grindability index.
Mathematical models of coal properties – including Gross calorific value and Hardgrove grindability index– have been developed, mainly, to reduce the number of instrumental analysis, determining the mathematical relations of certain properties with parameters that are taken from more simple analysis. Although mathematical models generalizing performance of properties are the goal, in empirical models of Coals this is very difficult, due to coal heterogeneity; hence empirical models are often developed for coals of specific locations, where the estimation of properties becomes the main goal.
Due to the above, added to the lack of studies of Caribbean Colombian Coals, in this study we aim to find through multiple regression and artificial neural networks an empirical mathematical model to determine Gross calorific value and Hardgrove grindability index for Colombian Caribbean Coals using the data from proximate and ultimate analysis as regressors.
In the development of this research, we collected data from three different coal mines, we applied parallelly the regression method and the artificial neural network and finally we compared the two mathematical models to determine which of two methods showed a better performance. After that we made a comparison of our final model with another mathematical models published in the international literature applied to studied coals. Finally we established some criteria to determine which procedure – of the two utilized in this study - use when somebody wants to predict certain property.