Rejecting Method of Abnormal Samples in Analysis Model of Coal Quality of Near-infrared Spectrum

In building of quantitative analysis model of coal quality of near-infrared spectrum, abnormal samples in training set seriously influence forecast precision of the model. So the paper proposed a new method of rejecting abnormal samples by twice diagnoses: get suspicious samples by clustering training samples based on fuzzy C-mean algorithm, and taking suspicious samples as validation set, reject abnormal samples by secondary diagnosis through PCA-GA-BP model. An experiment was done to compare predicting ability of the model with and without abnormal samples in training set by 15 groups of samples, and the result showed that the method can reject abnormal samples accurately, and improve forecasting precision effectively.