Proceedings Article | 13 December 2024
KEYWORDS: Laser induced breakdown spectroscopy, Data modeling, Principal component analysis, Spectroscopy, Mining, Carbon, Plasma, Statistical analysis, Machine learning, Combustion
China is rich in coal resources, and the detection of carbon content in coal in different mining areas can also meet the requirements of coal quality in various industries. At the same time, due to the different coal-forming years and coal seam quality, the carbon content of different coal mining areas is also different, which directly affects the combustion efficiency of coal and environmental pollution emission. Laser induced breakdown spectroscopy (LIBS) has gradually become a research hotspot in soil detection due to its advantages of fast, real-time, non-destructive detection, no need for sample pretreatment, simultaneous multi-element analysis, and remote operation. However, the traditional LIBS technology has some shortcomings in detection and application, such as low sensitivity, high detection limit, poor repeatability, strong self-absorption effect, and large matrix effect, which affect its research accuracy. Therefore, this paper proposes a coal classification method based on principal component analysis (PCA) LIBS technology combined with long short-term memory network algorithm (LSTM) to classify and identify coal samples from 10 different regions. After laser ablation of coal samples and corresponding data collection, data standardization and PCA preprocessing, LSTM optimization model is used to train and continuously generate analysis network of test sets. The final results show that the accuracy of coal classification by PCA-LSTM machine learning model can reach 99.58%, which proves that LIBS technology combined with PCA-LSTM can realize fast and accurate classification of soil in different mining areas. Therefore, this method can provide a new method for coal classification in different mining areas.
