摘要
针对BP神经网络的特点提出一种基于递阶遗传算法的四层BP神经网络的结构设计模型及应用。现有的BP训练方法只能训练BP网络的权重和阈值,网络的结构得预先用某种方法确定。利用很好设计的递阶遗传算法能够把网络的结构、权重和阈值同时通过训练确定。以经济系统中的人口时间序列数据进行训练和测试,与传统的BP网络预测模型相比较,结果证明该模型的预测精确度是令人满意的,提出的方法是可行的。
A nonlinear economic time series forecasting model, based on hierarchical genetic algorithm and four-layer BP neural network, was proposed. Different from the existing BP training method that can only lead to determination of connection weights and bias, a well-designed hierarchical genetic algorithm is used to train BP neural network with both connection weights, bias and numbers of neurons in two hidden layers determined at the same time. The model is then used to forecast the population. It is shown that the model based on hierarchical genetic algorithm and four-layer BP neural network is simple and effective.
出处
《系统仿真学报》
EI
CAS
CSCD
北大核心
2008年第9期2325-2328,2333,共5页
Journal of System Simulation
关键词
神经网络
BP神经网络
递阶遗传算法
经济系统预测
neural network
BP neural network
hierarchical genetic algorithm
economic system prediction