摘要
从数学和系统科学角度研究了经济预测中的神经网络泛化问题,主要结论包括:(1)利用实分析理论,证明了从有限到无限的不完全归纳以及经济系统的不连续性是造成泛化性能不良和预测效果不佳的根本原因.(2)从系统科学角度分析了系统噪音、随机性等对神经网络泛化能力的负面影响,证明了单纯的前馈神经网络不能对混沌经济系统作出可靠的预测.
Combining the view of mathematics with systems,the paper studied the the generalization of Feedforward Neural Network（FNN）：（1）By real-analysis,it is proved that：incomplete-induction of concluding infinity from finity and the possibility of discontinuity are two essential causes caused the failure of generalization of FNN.（2）From the systems point of view,it is clearly showed that how the system noises,randomness,inner randomness affact the generalization of FNN.We also proved that a chaotic system cannot be accurately approached by building any ANN model.
出处
《陕西科技大学学报（自然科学版）》
2011年第6期108-111,共4页
Journal of Shaanxi University of Science & Technology
基金
国家863计划项目:地质应用系统建设与典型应用示范研究(2008AA121103)
关键词
经济预测
前馈神经网络
泛化问题
混沌经济系统
初值敏感性
economic forcasting
FNN
generalization problem
economic chaotic system
initial value sensitivity
