The characteristics of the near infrared（NIR） spectra from the electricity coal samples were investigated.During the whole process of the study,with principal components score,Mahalanobis distance and Partial Least Squares cross validation for picking out outliers,the first three principal components and six discrete Fourier transformation（DFT） coefficients were obtained,and finally partial least squares regression（PLSR）,grid-support vector regression（G-SVR）,genetic algorithm-support vector regression（GA-SVR） and particle swarm optimization-support Vector Regression（PSO-SVR） quantitative analysis models were constructed.The results indicated that when the particle swarm optimization-support vector regression generation was 300,the population was 20,when c1 was 1.5 and c2 was 1.7,the calibration correlation coefficient was 0.990,the prediction correlation coefficient was 0.954,the calibration standard error was 0.366 and the prediction standard error was 0.128.The method was accurate and reliable,and was applied in the near infrared electricity coal calorific value detection system,which could also be used in other extremely complex near infrared online systems.