The Research On Coal Constituents Based On Functional Data Analysis

With the development of the electric power industry, demand for coal is expectedto increasing dramatically over the next decade, and thus, more attention has beenpaid on the problem of coal quality. As the coal qualities for different industries arediverse, it is becoming quite urgent for fast coal constituents analysis.Although many prediction models of coal constituents have been proposed in thepast few years, the research on them is still at the preliminary stage. Numerousmethods have been applied in coal quality analysis, including chemical analysis,regression analysis, peak area algorithm, inverse matrix approach, neural networkmethod, etc. However, the limitations of conventional methods, such as too muchtime-consuming, high computational complexity, inefficiency of data utilization andlow accuracy, can not be ignored. Functional data analysis has been attached greatimportance in the field of statistical theory in recent years, but the method forapplications in industries is now rarely used.In order to address the problems of conventional methods of coal quality analysisand further improve the ability of the prediction model, this paper attempts toimplement coal quality analysis based on functional data analysis. Based on thecharacters of coal data, we construct the prediction model by function linearregression method. The fundamental assumption is that, coal data can be consideredas the observational values of a function. First we converse original discrete data tocontinuous functions, then train the linear regression model, and finally smooth theconstructed function by roughness penalty and make predictions by its parameters.Based on the experiments performed on43coal data,39of them as trainingsamples and4of them as prediction samples, prediction results of many elementstotally meet or approach to the requirements of the enterprise. Experimental resultsshow that the new method provides high prediction accuracy of coal constituents.