论文部分内容阅读
基于正规正交分解(Proper Orthogonal Decomposition,POD)提出一种适用于非线性时间序列预测的径向基函数(Ra-dial Basis Function,RBF)神经网络模型-POD-RBF神经网络模型.该模型在选取中心时考虑了时间序列数据之间的时序关系,并且使得中心的选取具有并行性.股票价格预测问题的模拟结果表明,POD-RBF神经网络可以有效地用于非线性时间序列预测问题.与基于硬C均值(Hard C-means,HCM)聚类的RBF神经网络(HCM-RBF)和基于正交最小二乘(Orthogonal LestSquare,OLS)的RBF神经网络(OLS-RBF)相比,POD-RBF神经网络不仅具有更好的训练、预测精度,而且具有更好的收敛稳定性、更好的泛化能力和抵抗噪声干扰的能力.
Based on the Proper Orthogonal Decomposition (POD), this paper proposes a neural network model of Radial Basis Function (RBF) neural network, which is suitable for nonlinear time series prediction, The center of the time series is taken into consideration when considering the time series of the data, and the simulation results of the stock price forecasting problem show that the POD-RBF neural network can be effectively used in the nonlinear time series forecasting problem. RBF neural network (HCM-RBF) based on hard C-means clustering (HCM-RBF) and OLS-RBF neural network based on OLS (OLS-RBF) RBF neural network not only has better training and prediction accuracy, but also has better convergence stability, better generalization ability and resistance to noise interference.