Grade Level of Lignite Coal datas in the different areas with Decison Tree, Random Forest, and Discriminant Analysis Methods

Abstract

Lignite is one of the most important energy sources. An important problem in the economic and technical evaluation of lignite reserves is to measure lignite quality. The quality of lignite depends on some parameters such as moisture, ash, sulfur, and calorific values. The assessment of the parameters has a critical importance. The lignite data obtained from Kalburcayi area of the Sivas-Kangal Basin (SKKB) and the dataset in the Turkey Lignite Inventory (TLI) were used in this article. In addition to the average values given in TLI, another set (SKKB), which beyond the inventory, has been employed. By this way, comparable data were created for performing the modeling and classification work. To make lignite quality classification, a study was performed in five steps. In the first step, the calorific values have been used for verification by the k-means method. The coal lignite data are seperated into two groups, low and high quality. In the second step, wavelet families have been applied to the properties of moisture, ash, and sulfur regulated in the first step. The applied wavelet families such as haar, daubechies, symlet, biorspline, and reversebiorspline were used and the approximate coefficients produced by wavelet families have been obtained. In the third step, the features obtained in the second step have been given to random forest, discriminant analysis, and decision tree classifiers as input. In the next step, the quality classification performances have been compared for lignite coal data derived from SKKB and TLI. While the highest quality classification performance of lignite coals in the SKKB area has been found as 93.75%, the highest quality classification performance for lignite coals obtained from TLI has been found about 100%. In the final step, the success rates provided in this study have been compared with the conventional applications in literature. The results showed that the success rates of classification recorded by the proposed method better performs than the studies used for the comparison. Because this study addresses a hybrid work, more transparent and flexible classification structures can be provided. Making an effective and reliable classification between high and low lignite calorifics can provide some possibilities for decision-makers.