摘要
介绍主成分变换和经规则化调整法进行泛化改进的BP神经网络在森林蓄积量建模估测中的应用,比较普通BP神经网络与泛化改进的BP神经网络对蓄积量预报的差异,分析直接用中心标准化的观测值建立仿真模型和进行主成分变换后再建立模型的效率问题。结果表明:泛化改进的BP神经网络比普通BP神经网络具有更高的预报精度,利用主成分得分作为仿真模型的变量比直接用观测值作变量具有更快的速度,并保证了预报精度。
The application of principal component transformation and Bayesian regularization hack propagation （BP）neural network in forest volume estimate was introduced through a specific sample in this paper. The difference of forest volume estimate between general back propagation neural network and Bayesian regularization back propagation neural network was compared and the efficiency of estimating forest volume by the means of using original data and transformed data set to establish emulating model was discussed. All the results showed that Bayesian regularization hack propagation neural network was more accurate than general BP neural network in estimating forest volume and using transformed data set stemmed from principal component analysis to establish simulating model is more efficient than using original data.
出处
《林业科学》
EI
CAS
CSCD
北大核心
2006年第12期59-62,共4页
Scientia Silvae Sinicae
基金
国家自然科学基金(30471424
30371159)资助。
关键词
BP神经网络
主成分变换
泛化
森林蓄积量
BP neural network
principal component transformation
generalization
forest volume
