关键词: 近红外光谱, 灰狼算法, 支持向量机, 杉木密度

Abstract: [Objective] In order to explore the more efficient method of predicting the wood density of Cunninghamia lanceolata, the near infrared spectroscopy was used. It could provide the theoretical basis for quantitative analysis of wood properties.[Method] Firstly, wood density and the near infrared spectroscopy data of 109 C. lanceolata samples were normalized. Of which,88 C. lanceolata samples constituted training set and 21 samples constituted test set. Secondly, the principal component analysis was used to extract the principal components of 2 151 dimensions of the C. lanceolata near infrared spectrum. Thirdly, the principal components set as the independent variable and the C. lanceolata density set as the dependent variable were used to establish the multiple linear regression(MLR)model and the support vector machine(SVM)model. In order to improve the prediction accuracy of C. lanceolata density model, the grey wolf optimizer(GWO)algorithm was applied to the SVM model for parameter optimization. Therefore the prediction of the C. lanceolata density based on the GWO-SVM was proposed. The determination coefficient(R²), the mean square error(MSE)and the mean absolute percentage error(MAPE)were adopted to measure the prediction result of the three models.[Result] Five principal components were obtained from the near infrared spectrum, and their cumulative contribution rate was 98.7%. The significant value P of the MLR model and the partial regression coefficient was both less than 0.05, which indicated that the MLR model was effective and the model can be used for the C. lanceolata density prediction. The R² of MLR model was 0.771 4, the MSE was 0.000 282 1, and the MAPE was 3.009 23%. At the same time, the R² of SVM model was 0.923 8, the MSE was 0.000 233 1, and the MAPE was 2.794 50%. The parameters of the SVM were optimized by the wolf group algorithm, and the optimal parameters were C=18.366 6, σ=0.043 3. In the GWO-SVM model, the R² was 0.919 2, the MSE was 0.000 183 4, and the MAPE was 2.496 37%.The MAPE of the three models were all within the acceptable range, and the prediction of the GWO-SVM model was the best.[Conclusion] The GWO-SVM model is superior to the MLR model and the SVM model on the prediction accuracy. The GWO-SVM and the SVM model are superior to the MLR model on determination coefficient analysis. So the approach of GWO-SVM combined with the near infrared spectroscopy to predict the C. lanceolata density is reasonable and efficient.

Key words: near infrared spectrum, grey wolf optimizer(GWO), support vector machines(SVM), wood density of Cunninghamia lanceolata