• 首页关于本刊期刊订阅编委会作者指南过刊浏览
卢志宏,武晓东,柴享贤,杨素文,李燕妮,叶丽娜.2017.应用BP神经网络对荒漠啮齿动物种群数量的预测研究.动物学杂志,52(2):227-234.
应用BP神经网络对荒漠啮齿动物种群数量的预测研究
Prediction of the Population of Rodent Community Based on BP Neural Network in Alasan Desert
投稿时间:2016-02-14  修订日期:2017-02-23
DOI:10.13859/j.cjz.201702006
中文关键词:  BP人工神经网络  啮齿动物  标志重捕法
英文关键词:BP neural network  Rodent  Mark recapture method
基金项目:国家自然科学基金项目(No. 30760044,31160096),公益性行业科研专项经费项目(No. 201203041)
作者单位E-mail
卢志宏 内蒙古农业大学生态环境学院 luzhihong305@163.com 
武晓东* 内蒙古农业大学生态环境学院 wuxiaodong_hgb@163.com 
柴享贤 内蒙古农业大学生态环境学院  
杨素文 内蒙古农业大学生态环境学院  
李燕妮 内蒙古农业大学生态环境学院  
叶丽娜 内蒙古农业大学生态环境学院  
摘要点击次数: 2095
全文下载次数: 1944
中文摘要:
      群落的格局与动态是群落生态学研究的核心内容,种群数量预测是研究群落动态的主要途径之一。本研究尝试采用2006 ~ 2014年阿拉善荒漠区啮齿动物数量数据建立BP神经网络模型,对啮齿动物群落全部组成物种的总个体数量进行模拟与预测。BP神经网络通过模拟学习,建立模型,能够实现对啮齿动物群落数量动态规律进行模拟与预测。本研究以阿拉善荒漠为试验区,以啮齿动物个体数量为研究对象,采用标志重捕法,监测2006 ~ 2014年每年4 ~ 10月的数量,建立BP神经网络预测模型,利用2006 ~ 2013年的数据建立训练网络,以2014年的数据进行验证与测试,比较单层隐含层、双层隐含层和三层隐含层BP神经网络模型。结果表明:单隐含层模型的隐含层节点数为6时,最大误差百分比为16.13%,决定系数0.998 0(P = 0.006 0)。双隐含层模型的两层隐含层节点数均为6时,最大误差百分比为8.58%,决定系数0.999 5(P = 0.002 3)。三层隐含层模型的三层隐含层节点数分别为1、10和7时,最大误差百分比为5.87%,决定系数0.999 2(P < 0.000 1)。不同隐含层网络模型的预测效果均取得了满意效果,通过比较最大误差百分比、平均误差百分比、决定系数及拟合优度,三层隐含层优于单隐含层及双隐含层的BP神经网络模型。本文认为三层隐含层的BP神经网络模型更适合于阿拉善荒漠区啮齿动物群落全部组成物种的总个体数量的预测研究。
英文摘要:
      Community structure and dynamics is the core content of community ecology, and the prediction of population is one way to study the community dynamics. This study attempted to establish a neural network model based on BP neural network model in the rodent communities in Alasan desert area from 2006 to 2014. Through the simulation study and the establishment of the model, BP neural network could achieve the simulation and forecast the dynamic law of the number of rodent communities. Taking Alasan Desert as a test area and the number of rodents as the research object, this study used mark recapture method to monthly monitor the catches from April to October from 2006 to 2014, count the minimum alive number, and set up BP neural network prediction model. Then, built the training network by the data from 2006 to 2013 minimum survival, and used 2014 year′s data for verification and testing. Through comparing hidden layer, double hidden layer and triple hidden layer BP artificial neural network model. The results showed that: 1) When the nodes' number of single hidden layer were 6, the maximum error percentage of single hidden layer model was 16.13%, and the determination coefficient was 0.998 0 (P = 0.006 0, Table 1). 2) When the nodes' number of the two hidden layer were both 6, the maximum error percentage of double hidden layer model was 8.58%, and the coefficient of determination was 0.999 5 (P = 0.002 3, Table 2). 3) When the nodes' number of the triple hidden layer is 1, 10, and 7, the maximum error percentage of triple hidden layer model was 5.87%, and the determination coefficient was 0.999 2 (P < 0.000 1, Table 3). 4) The forecasting effect of the different hidden layer network model has been achieved. By comparing the maximum error percentage, the average error percentage, the decision coefficient, and nonlinear fitting rate, the triple hidden layer was better than other two BP neural network models (Table 4). In this paper, the BP neural network model with triple hidden layer was most suitable for the prediction of rodent population in Alasan desert area.
附件
查看全文  查看/发表评论  下载PDF阅读器