基于相关向量回归的非线性时间序列预测方法

Nonlinear Time Series Forecasting Based on Relevance Vector Regression

针对非线性时间序列预测建模的复杂性和不确定性,提出一种基于相关向量回归的非线性时间序列预测方法。该方法在传统的核函数基础上,融入Bayesian推理框架,得到具有概率特性的预报结果,无须对误差/边界参数进行预估计,具有学习算法简单、易实现的特点。仿真计算表明,该方法能反映非线性时间序列的内在特性,预测结果较好。

A forecasting method for nonlinear time series based on Relevance Vector Machine（RVM） is proposed for the purpose of dealing with the complexity and uncertainty during engineering modeling. Based on the traditional kernel functions, RVM using a sparse kernel representation can directly provide probabilistic forecasting results under Bayesian frame. The method is simple and easy to be realized without pre-calculation of error/margin parameters. Simulation instance shows that the method reflects inherent characteristics of nonlinear time series, exhibits high model efficiency and provides satisfying forecasting precision.