摘要
自组织理论是基于神经网络和计算机科学的迅速发展而产生和发展起来的。它将黑箱思想、生物神经元方法、归纳法、概率论、数理逻辑等方法有机地组合起来。其主要思想是通过简单的初始输入(局部变量)的交叉组合产生第一代中间候选模型,再从第一代中间候选模型中选出最优的若干项组合而产生第二代中间候选模型,重复这样一个产生、选择和遗传进化过程,使模型复杂度不断增加,直到选出最优复杂度模型为止。本文利用自组织方法进行数据筛选和建立税收预测模型,并在数据筛选基础上建立线性回归预测模型和BP神经网络预测模型,然后结合时间序列的预测模型,利用自组织方法建立组合预测模型。通过预测结果比较得出了组合预测模型比其它单个模型具有更高的预测精度。
The methodology of self-organization is originated from development of neural network and computer science. It is a logical combination of Black Box Theory, Biological Neuron Methodology, Induction Method, Probability Theory and Mathematical Logic. Its basic idea is： combination of simple initial input （local variables） generates the first generation of intermediate candidate model from which the best will be selected for the second generation, after several rounds of such cycles involving generation, selection and evolution, models are in- creasingly complicated until the best one is selected. The methodology of self-organization is adopted to select data as basis for establishing a linear regression model of forecasting and a forecasting model based on BP neural network, the self-organization model will then be adopted to establish a combined forecasting model with forecasting model of time series being taken into consideration. Comparison between different models proves high accuracy of the combined forecasting model.
出处
《技术经济与管理研究》
北大核心
2010年第2期38-40,共3页
Journal of Technical Economics & Management
关键词
组合预测
自组织方法
神经网络
线性回归
时间序列
combined forecasting
methodology of Self-organization
neural network
linear regression
time series.
