Rapid analysis of coal characteristics based on deep learning and visible-infrared spectroscopy

Abstract With the development of society, the global demand for coal quality is getting higher and higher.High-quality coal has a major impact on theindustry and the environment.Therefore, it is essential to analyze the characteristics of coal before using coal.This study proposes a method for quickly analyzing coal characteristics.The method can detect the volatile matter,ash, moisture, and fixed carbon of coal.First, coal samples were collected and the visible-infrared spectral data of these samples were measured byan SVC HR-1024 spectrometer. Then, advanced features of spectral data are extracted through a deep belief network (DBN). A derivative function with regularization two-layer extreme learning machine (DF-RTELM) algorithm is proposed and uses the algorithm to construct an analysis model for coal characteristics.The results show that the proposed model can effectively predict the compositions of coal.For the testing cost of 158 coal samples, our method only takes about 5 h and $ 100 USD, while the traditional proximate, ultimate analysis methods take about 300 h and $ 6,000 USD.Therefore, compared with these methods, our proposed method can greatly reduce the cost.