Improved analysis of inorganic coal properties based on near-infrared reflectance spectroscopy
Abstract
Near-infrared reflectance spectroscopy (NIRS) is a fast and convenient analytical tool, and it has become the preferred choice for online coal property analysis in recent years. Organic molecules have much stronger absorption ability than inorganic molecules. Therefore, better analysis accuracy can be achieved for organic properties of coal such as volatile matter and fixed carbon than that for inorganic properties such as ash and sulfur. This paper used a much better algorithm (least squares support vector machine) than a previous study and proposes a new method to improve the analysis accuracy of inorganic properties by utilizing the analysis results of volatile matter and fixed carbon to enhance the regression models for ash and sulfur. Four types of coal (i.e. fat, coking, lean and meager lean) were considered in our experiments. Individual models for each type of coal have been established and predicted values of volatile matter and fixed carbon based on NIRS were added with the relevant PCA components during the modeling. The experimental results have shown that our proposed method which utilizes information of the organic properties could improve the analysis results of the inorganic properties by around 35% using NIRS.