Classification of Indian power coals using K-means clustering and Self Organizing Map neural network

The present study reports results of the classification of Indian coals used in thermal power stations across India. For classifying these power coals a classical unsupervised clustering technique, namely “K-Means Clustering” and an artificial intelligence (AI) based nonlinear clustering formalism known as “Self-Organizing Map (SOM)” have been used for the first time. To conduct the said classification, five coal descriptor variables namely moisture, ash, volatile matter, carbon and gross calorific value (GCV) have been used. The classification results thereof indicate that Indian power coals from different geographical origins can be classified optimally into seven classes. It has also been found that the K-means and SOM based classification results exhibit similarity in close to 75% coal samples. Further, K-means and SOM based seven coal categories have been compared with as many grades of a commonly employed Useful Heat Value (UHV) based Indian non-coking coal grading system. Here, it was observed that a number of UHV-based grades exhibit similarity with the categories identified by the K-means and SOM methods. The classification of Indian power coals as provided here can be gainfully used in selecting application-specific coals as also in their grading and pricing.