光谱学与光谱分析
光譜學與光譜分析
광보학여광보분석
SPECTROSCOPY AND SPECTRAL ANALYSIS
2014年
1期
175-179
,共5页
焦有权%冯仲科%赵礼曦%徐伟恒%曹忠
焦有權%馮仲科%趙禮晞%徐偉恆%曹忠
초유권%풍중과%조례희%서위항%조충
粒子群算法(PSO)%支持向量机(SVM )%活立木材积%光电经纬仪
粒子群算法(PSO)%支持嚮量機(SVM )%活立木材積%光電經緯儀
입자군산법(PSO)%지지향량궤(SVM )%활립목재적%광전경위의
Particle swarm optimization(PSO)%Support vector machine(SVM)%Living tree volume%Photoelectric theodolite
为了建立立木材积模型，每年有数十万棵优质活立木被伐倒，这是一种破坏性较大的实验。应用光电经纬仪自动量测与手工量测活立木地径、胸径相结合，通过活立木材积计算软件批量计算，获得中林系107杨（Zhonglin aspens No.107）的胸径、树高、树干材积值400组数据集。采用粒子群算法嵌入支持向量机（PSO-SVM ）建立了非线性智能活立木材积预报模型，并以400组实验数据集作为研究资料，随机抽选300组数据的胸径和树高作为输入值，材积为输出值，用MATLAB软件运行PSO-SVM工具箱，训练得到PSO-SVM 模型，用100组数据进行检验预测。研究表明，PSO-SVM算法模型预测值与实测值间复相关系数 r为0.91， r值比Spurr二元材积模型计算值高出2％，平均绝对误差率提高0.44％。引用经典Spurr二元材积模型计算值和PSO-SVM模型预报值进行对比，认为将PSO算法引入到SVM参数优化中，使活立木材积预报具有自学习能力和自适应能力，PSO-SVM模型对样本数量要求较小、预报准确率高、学习速度快，具有很好的推广价值与应用前景。
為瞭建立立木材積模型，每年有數十萬棵優質活立木被伐倒，這是一種破壞性較大的實驗。應用光電經緯儀自動量測與手工量測活立木地徑、胸徑相結閤，通過活立木材積計算軟件批量計算，穫得中林繫107楊（Zhonglin aspens No.107）的胸徑、樹高、樹榦材積值400組數據集。採用粒子群算法嵌入支持嚮量機（PSO-SVM ）建立瞭非線性智能活立木材積預報模型，併以400組實驗數據集作為研究資料，隨機抽選300組數據的胸徑和樹高作為輸入值，材積為輸齣值，用MATLAB軟件運行PSO-SVM工具箱，訓練得到PSO-SVM 模型，用100組數據進行檢驗預測。研究錶明，PSO-SVM算法模型預測值與實測值間複相關繫數 r為0.91， r值比Spurr二元材積模型計算值高齣2％，平均絕對誤差率提高0.44％。引用經典Spurr二元材積模型計算值和PSO-SVM模型預報值進行對比，認為將PSO算法引入到SVM參數優化中，使活立木材積預報具有自學習能力和自適應能力，PSO-SVM模型對樣本數量要求較小、預報準確率高、學習速度快，具有很好的推廣價值與應用前景。
위료건립립목재적모형，매년유수십만과우질활립목피벌도，저시일충파배성교대적실험。응용광전경위의자동량측여수공량측활립목지경、흉경상결합，통과활립목재적계산연건비량계산，획득중림계107양（Zhonglin aspens No.107）적흉경、수고、수간재적치400조수거집。채용입자군산법감입지지향량궤（PSO-SVM ）건립료비선성지능활립목재적예보모형，병이400조실험수거집작위연구자료，수궤추선300조수거적흉경화수고작위수입치，재적위수출치，용MATLAB연건운행PSO-SVM공구상，훈련득도PSO-SVM 모형，용100조수거진행검험예측。연구표명，PSO-SVM산법모형예측치여실측치간복상관계수 r위0.91， r치비Spurr이원재적모형계산치고출2％，평균절대오차솔제고0.44％。인용경전Spurr이원재적모형계산치화PSO-SVM모형예보치진행대비，인위장PSO산법인입도SVM삼수우화중，사활립목재적예보구유자학습능력화자괄응능력，PSO-SVM모형대양본수량요구교소、예보준학솔고、학습속도쾌，구유흔호적추엄개치여응용전경。
In order to establish volume model ,living trees have to be fallen and be divided into many sections ,which is a kind of destructive experiment .So hundreds of thousands of trees have been fallen down each year in China .To solve this problem ,a new method called living tree volume accurate measurement without falling tree was proposed in the present paper .In the meth-od ,new measuring methods and calculation ways are used by using photoelectric theodolite and auxiliary artificial measurement . The diameter at breast height and diameter at ground was measured manually ,and diameters at other heights were obtained by photoelectric theodolite .Tree volume and height of each tree was calculated by a special software that was programmed by the authors .Zhonglin aspens No .107 were selected as experiment object ,and 400 data records were obtained .Based on these data , a nonlinear intelligent living tree volume prediction model with Particle Swarm Optimization algorithm based on support vector machines (PSO-SVM) was established .Three hundred data records including tree height and diameter at breast height were ran-domly selected form a total of 400 data records as input data ,tree volume as output data ,using PSO-SVM tool box of Mat-lab7.11 ,thus a tree volume model was obtained .One hundred data records were used to test the volume model .The results show that the complex correlation coefficient (R2 ) between predicted and measured values is 0.91 ,which is 2% higher than the value calculated by classic Spurr binary volume model ,and the mean absolute error rates were reduced by 0.44% .Compared with Spurr binary volume model ,PSO-SVM model has self-learning and self-adaption ability ,moreover ,with the characteristics of high prediction accuracy ,fast learning speed ,and a small sample size requirement ,PSO-SVM model with well prospect is worth popularization and application .
