Abstract: Five levels of Huangshanmaofeng tea with 6 varied storage time were detected by electronic nose. Firstly， the original feature vectors presenting the tea odor were acquired. Then， the first 5 principal components were extracted as the principal feature vectors by principal component analysis （PCA）. With the principal feature vectors used as BPNN input， a new model called PCABPNN for storage time analysis of Huangshanmaofeng tea was built. After experimental test， it was shown that for the tea of 0 d storage， the maximum prediction error （MPE） was 11 d， and the samples of prediction error exceeding 10 d was 5 （6.67%）; for the tea of 60 d storage， MPE was 13 d， and the samples of prediction error exceeding 10 d was 4 （5.33%）; for the tea of 120 d storage， MPE was 16 d， and the samples of prediction error exceeding 10 d was 7 （9.33%）; for the tea of 180 d storage， MPE was 19 d， and the samples of prediction error exceeding 10 d was 8 （10.67%）; for the tea of 240 d storage， MPE was 21 d， and the samples of prediction error exceeding 10 d was 8 （10.67%）; for the tea of 300 d storage， MPE was 14 d， and the samples of prediction error exceeding 10 d was 6 （8.00%）. In conclusion， PCABPNN model could be used to analyze the storage time of Huangshanmaofeng tea， and the proposed model was better than BPNN， which used original feature vectors as the input.

Key words: electronic nose, PCA, BPNN, prediction model