基于人工蜂群算法优化SVM的NIR杉木弹性模量预测  

Elastic Modulus Prediction of Cunninghamia lanceolata Based on Artificial Bee Colony Algorithm SVM and NIR

作　　者：陈芳 程献宝[2] 黄安民[2] 王学顺[1] Chen Fang;Cheng Xianbao;Huang Anmin;Wang Xueshun(School of Science,Beijing Forestry University,Beijing 100083;Research Institute of Wood Industry, CAF Beijing 100091)

机构地区：[1]北京林业大学理学院,北京100083 [2]中国林业科学研究院木材工业研究所,北京100091

出　　处：《林业科学》2021年第1期161-168,共8页Scientia Silvae Sinicae

基　　金：中央高校基本科研业务费专项资金(2015ZCQ-LY-01);国家自然科学基金项目(31670564)。

摘　　要：【目的】利用近红外光谱分析技术,提出一种基于人工蜂群算法优化支持向量机(ABC-SVM)的木材弹性模量预测模型,为木材弹性模量无损预测提供科学参考。【方法】以294个杉木样本为试验材料,采用常规力学方法测量样本弹性模量,采集样本近红外漫反射光谱,选择350~2500 nm光谱段,对原始数据进行15步指数平滑和一阶导数预处理,并利用主成分分析降维处理后的数据,建立偏最小二乘回归(PLS)模型、支持向量机回归(SVR)模型和人工蜂群算法优化支持向量机(ABC-SVM)模型预测杉木弹性模量,采用决定系数(R2)、均方根误差(RMSE)、平均绝对百分比误差(MAPE)和平均绝对误差(MAE)对所建模型进行对比分析。【结果】PLS模型的R2为0.726700、RMSE为6.7449、MAPE为0.0635、MAE为5.0656;SVR模型的R2为0.935305、RMSE为3.5281、MAPE为0.0237、MAE为1.8409;将人工蜂群算法用于SVM参数寻优,获得的最优参数c=5.67051、g=0.03125,ABC-SVM模型的R2为0.935371、RMSE为3.5260、MAPE为0.0237、MAE为1.8400。3种模型均可对杉木弹性模量进行有效预测。【结论】1)根据决定系数(R2),SVR和ABC-SVM模型的预测性能优于PLS模型,ABC-SVM模型的预测性能最佳;2)根据均方根误差(RMSE)、平均绝对百分比误差(MAPE)和平均绝对误差(MAE),3种模型的MAPE均在可接受范围内,ABC-SVM模型关于误差的各项指标均最小,基于ABC-SVM模型预测杉木弹性模量高效、科学。【Objective】Using near-infrared spectrum(NIR)analysis technology,a wood elastic modulus prediction model based on support vector machine optimized by artificial bee colony algorithm(ABC-SVM)was proposed,which was expected to provide a scientific reference for non-destructive prediction of wood elastic modulus.【Method】Taking 294 samples of Cunninghamia lanceolata as experimental materials,the elastic modulus of the samples was measured by conventional mechanical method,and the near-infrared diffuse reflection spectra of the samples were collected.The 350-2500 nm spectrum was selected and the original spectrum was preprocessed by first-order derivation and 15-step exponential smoothing.Using principal component analysis to reduce the data,then the partial least squares regression(PLS)model,support vector machine regression(SVR)model and the ABC-SVM model were established to predict the elastic modulus of Cunninghamia lanceolata.The determination coefficient(R2),the root mean square error(RMSE),the mean absolute percentage error(MAPE)and the mean absolute error(MAE)were used to evaluate the advantages and disadvantages of the three models.【Result】The R2 of PLS model is 0.726700,the RMSE is 6.7449,the MAPE is 0.0635,and the MAE is 5.0656.As for SVR model,the R2 is 0.935305,the RMSE is 3.5281,the MAPE is 0.0237,the MAE is 1.8409.Using artificial bee colony algorithm to optimize SVM parameters,the optimal parameters are c=5.67051,g=0.03125.The R2 of ABC-SVM model is 0.935371,the RMSE is 3.5260,the MAPE is 0.0237 and the MAE is 1.8400.All three models could effectively predict the elastic modulus of Cunninghamia lanceolata.【Conclusion】1)Observing the index of determination coefficient(R2),both SVR and ABC-SVM models have better prediction performances than PLS model,and ABC-SVM has the best prediction performance.2)According to the obtained error results(RMSE,MAPE and MAE),the MAPE of the three models is within the acceptable range,and the ABC-SVM model has the smallest indicators for the error,indicating

关 键 词：近红外光谱 弹性模量 支持向量机 人工蜂群算法 

分 类 号：S781.2[农业科学—木材科学与技术]