级联相关算法在个体定税中的应用研究  

Application of Cascade-Correlation Algorithm to Individual Constant Tax

作　　者：李宏[1] 李满玲[1] 朱学军[2] 

机构地区：[1]中南大学信息科学与工程学院,湖南长沙410083 [2]株洲市地方税务局信息中心,湖南株洲412007

出　　处：《计算机仿真》2007年第12期257-259,277,共4页Computer Simulation

摘　　要：网络结构的选择是神经网络建模中的难题。BP算法必须在固定的网络拓扑结构下才能进行样本训练。纳税户信息变更频繁,当训练样本或参数发生改变都有可能导致网络结构的不适性。BP神经网络因其自身原因很难解决这个问题。此外,基于BP神经网络的个体定税预测的精度仍有待提高。文中提出一种具有自适应调整网络结构功能的级联相关算法,并将其引入到个体定税建模中。实验证明,该算法具有自适应性、预测精度高等特点,比BP算法更具建模能力,更适用于个体定税工作中。Choosing a network size is a difficult problem in neural network modeling.The training is generally carried out with the BP algorithm which performs gradient descent only in the neural network with static topology and the information of taxpayer is changed frequently.When training patterns or parameters are changed,the network structure may be changed.However,BP neural network is difficult to resolve the problem for its own reasons.Moreover,the precision of forecasting model of individual constant tax based on BP neural network should be improved.A method based on Cascade-Correlation which could adjust the network structure automatically is proposed in the paper,and applied to the experimentation of individual constant tax.The experimentation proved that the algorithm has the features of adjusting network structure automatically and high precision.It has stronger modeling ability and is more practical than BP algorithm for application to individual constant tax.

关 键 词：级联相关 个体定税 反向传播 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]