关键词：
森林计测学;立木材积;量子粒子群(QPSO);最小二乘支持向量机(LSSVM);材积方程
摘 要：
基于材积方程建立的材积表是森林资源调查工作中重要的工具,估算立木材积的精度是编制材积表的关键。为了解决已有立木材积方程复杂多样、测算准确率低等不足,以北京地区侧柏Platycladus orientalis和落叶松Larix principis-rupprechtii为研究对象,提出利用量子粒子群优化最小二乘支持向量机(QPSO-LSSVM)算法建立材积方程的方法。通过伐倒解析法结合电子经纬仪无损立木材积精测法获取建模样本,对250株侧柏与300株落叶松数据分别建立一元与二元材积方程,计算得到侧柏与落叶松的一元材积方程测试集的决定系数(R2)为0.978 6和0.946 1,二元材积方程测试集决定系数(R2)为0.987 0和0.990 1,均在0.940 0以上,总体相对误差(TRE)依次为0.75%,-0.16%, 0.64%,-0.50%,均满足国家规程小于±3%的要求,表明QPSO-LSSVM模型估算效果良好。最后引用传统一、二元材积方程、 BP神经网络和粒子群优化最小二乘支持向量机(PSO-LSSVM)算法建立材积方程并与之进行对比分析。结果表明:QPSO-LSSVM材积方程在估测精度、收敛速度和稳健性等综合性能指标上优于其他材积方程。该方法在高精度材积估测中具有较好的应用前景。
译 名：
Tree volume estimates based on QPSO-LSSVM
作 者：
YANG Liyan;FENG Zhongke;LIU Yingchun;LIU Jincheng;Precision Forestry Key Laboratory of Beijing, Beijing Forestry University;School of Land Resources and Urban and Rural Planning, Hebei GEO University;Academy of Forestry Inventory and Planning,State Forestry Administration;
单 位：
YANG Liyan%FENG Zhongke%LIU Yingchun%LIU Jincheng%Precision Forestry Key Laboratory of Beijing, Beijing Forestry University%School of Land Resources and Urban and Rural Planning, Hebei GEO University%Academy of Forestry Inventory and Planning,State Forestry Administration
关键词：
forest mensuration;;tree volume;;Quantum-Behaved Particle Swarm Optimization(QPSO);;Least Squares Support Sector Machines(LSSVM);;volume equation
摘 要：
A volume table with calculations based on a volume equation, is an important tool for estimating tree volume in forest resources inventory. To overcome the large total relative error(TRE) that sometimes occurs,making the volume tables unable to meet the precision requirements of forest surveys and production, a Quantum-Behaved Particle Swarm Optimization(QPSO)-Least Squares Support Vector Machines(LSSVM) Model was used to estimate the volume of Platycladus orientalis and Larix principis-rupprechtii in Beijing. The traditional method of analyzing cut trees and the method of electronic theodolite standing timber volume were used to measure diameter at breast height(DBH), tree height, and volume of small diameter class(6-10 cm) and large diameter classe(12 cm or more) were used. A total of 792 trees were collected, including 344 P. orientalis and 448 L. principis-rupprechtii. Then, one-variable and two-variable tree volume equations of P. orientalis and L. principis-rupprechtii were established based on the QPSO-LSSVM algorithm. Also the volume model was established using the traditional empirical volume equation, BP neural network, and the PSO-LSSVM algorithm to test comprehensive performance indexes such as estimate accuracy, convergence speed, and robustness.The sample data set was divided into train set and test set according diameter class in the experiment. There were 250 train set samples of Platycladus orientalis, accounting for 72.67% of the total number, and 94 test set samples of it. Also, there were 300 train set samples of Larix principis-rupprechtii, accounting for 66.96% of the total number, and 148 test set samples of it. Results of the regression analysis on the improved one-variable and two-variable tree volume equations of P. orientalis and L. principis-rupprechtii showed R2(the coefficient of determination of the test sets) = 0.978 6, 0.946 1, 0.987 0, and 0.990 1, with the TREof 0.75%,-0.16%,0.64%, and-0.50%, all within ±1%. The QPSO-LSSVM has higher comprehensive performance indexes for the volume equation than the other three volume equations, including traditional empirical volume equation, BP neural network and the PSO-LSSVM algorithm. Thus, the proposed QPSO-LSSVM method should have a favorable application prospect with high precision volume estimates.
