应用神经网络和多元回归技术预测森林产量

Abstract

应用传统统计技术常会因样本小和测量数据不符某种分布而受到限制。本文评价一种前馈型神经网络算法以预测落叶阔叶林产量。另外,还介绍一种由定性变为定量的数据变换方法,以用相对小的样本建立多元回归预测模型。数据变换方法有助于改善多元回归模型的预测效果。在本实验的条件下,研究结果表明神经网络技术能够产生最好的预测效果.

Use of traditional statistical techniques is often limited by shortage of observation samples and difference in data measurement scales. Neural network techniques have been extensively explored in many fields for prediction and classification as an alternative to statistical methods. In this paper, a feed-forward neural network algorithm for predicting hardwood yield is introduced and evaluated. In addition, we report a data transformation method developed for converting qualitative variable data to quantitative data for use in multiple regression when relatively few samples are available for building prediction models. The method that converts qualitative data into quantitative data is helpful to improve hardwood yield prediction accuracy by multiple linear regression models. In this study, the best prediction results using the neural network technique are obtained.