基于RBF神经网络的非线性时间序列在线预测

On-line prediction of nonlinear time series using RBF neural networks

针对非线性非高斯时间序列,提出观测噪声服从隐马尔可夫模型(HMM)的径向基函数(RBF)神经网络(RBF-HMM)预测模型,其特点在于模型输入包含误差反馈项、RBF网络隐含层节点数的可变性和观测噪声的隐马尔可夫性;并采用序列蒙特卡罗(SMC)方法实现基于RBF-HMM模型的时间序列在线预测.最后采用太阳黑子数平滑月均值数据和CRU国际钢材价格指数月数据进行实证研究,结果表明该模型的有效性.

For the nonlinear and non-Gaussian time series, a novel predictive model--the RBF-HMM model is proposed based on the radial basis function（RBF） neural network with measurement noise being assumed to be of a hidden Markov model（HMM）. The characteristics of this model include： 1） the predictive errors of RBF neural network are associated with the input of RBF-HMM model; 2） the number of hidden neurons varies with time; 3） the measurement noise is assumed to be HMM distributed. Sequential Monte Carlo（SMC） method is then applied to the on-line prediction for time series in RBF- HMM model. Finally, the smoothed data of the monthly mean sunspot numbers and CRU（the Britain Commodity Research University） steel price index are analyzed. The experimental results indicate that the RBF-HMM model is effective.