基于相似性最优模块神经网络的股票预测

该文提出一种最优模块化神经网络的模型.BP网络存在学习后面的样本而"遗忘"前面的样本,以及训练速度很慢的问题,但具有泛化能力强的优点,同时网络的结构不会随数据增加而变的庞大.而RBF网络随着输入维数增加其隐藏层的神经元个数呈指数增加,并且其泛化能力不强,但RBF网络具有训练速度比较快,逼近效果好等优点.于是提出最优模块化神经网络的模型,综合BP和RBF网络的优点.使学习样本能力,运算速度,网络规模得到改善.该模型适合于较多的样本训练.

The Stack Predication Based on Comparability and Optimum of a Modular Neural Network

In this paper a best modular neural network model is introduced.The defect of BP network is the problem of training speed and forgetting the sample feature learnt before,but BP network's generalization is well and the network's size is not very large.The defect of RBF network is generalization and the network's size is very large with increasing the input dimension,but its training speed is very fast and the approximation is very good.We construct the modular neural network model integrated with BP and RBF ...