The coal quality test data of about 150 thermal power enterprises are selected, and the coal calorific value prediction model based on L-M algorithm is built by analyzing the ubiquitous coal quality test data information. The experimental results show that: (1) the linear relationships between calorific value and carbon, ash are significant respectively, and the correlation coefficients are 0.8768 and 0.6880; (2) principal component analysis(PCA) method excavates the information of the principal component eigenvalue, characteristic matrix and score, which could realize the dimension reduction effect from six-dimensional matrix to four-dimensional matrix, and enhance the convergence stability of neural network in the training process; (3) based on the L-M algorithm, the training coefficient , verification coefficient and test coefficient of the Levenberg-Marquardt back propagation neural network prediction method(LMBP) model are 0.9957, 0.9942 and 0.9963 respectively, and the overall coefficient is 0.9931. Through the further verification of 20 groups of data to be tested, the LMBP prediction model is more reliable, the prediction accuracy is higher, and better to meet the actual forecast demand.

coal quality test data  /  Principal component analysis  /  L-M algorithm  /  training set coefficient