基于GA-NN的碳通量预测因素选择

Factors selection for prediction of carbon flux based on Genetic Algorithm—Neural Network

研究和选择碳循环的影响因素是预测碳通量的重要环节,也是研究碳循环机理的重要步骤。然而从众多的影响因素中选择重要的因素,依然存在着困难。提出利用相关分析、遗传算法和神经网络进行碳通量预测的主要因素选择的方法,首先用相关分析去处冗余的因素;然后利用遗传算法,以选择最小数目的因素时,最大碳通量的观测值和用神经网络预测值的相关系数为准则,来搜寻最优的影响因素。实验证明该方法能在不影响(或尽量小地影响)预测精度的前提下,有效地选择出碳通量预测的重要因素。

Selecting the driving factors for carbon cycle is critical step prior predicting carbon dioxide（CO2） flux and it also is the important step to study the machines of carbon cycle.But,how to select the driving factors among the plenty of factors is still a challenge problem.This paper proposes a method of driving factors selection based on correlation analysis, Genetic Algorithm（GA）—Neural Network（NN）.The redundant factors are reduced using correlation analysis firstly.And then, GA is used to select the driving factors according the criteria that maximizes the correlation coefficient between the Net Ecosystem Exchange（NEE） observed and the NEE predicted using Radial Basis Function Neural Network（RBFNN） as well as minimizes the number of driving factors.To evaluate the validity of the proposed method,it is used to select the driving factors in predicting CO2 flux of Duke Forest.The experimental results illustrate that the method can mine the main driving factors for predictive CO2 flux effectively without loss of precision.