Abstract: Due to the fluctuation and complexity of non-linear and non-stationary time series, it is difficult to use a single forecasting method to accurately describe the moving tendency. So a novel hybrid intelligent forecasting model based on Empirical Mode Decomposition（EMD） and Support Vector Regression（SVR） is proposed, where these Intrinsic Mode Functions（IMF） are adaptively extracted via EMD from a non-stationary time series according to the intrinsic characteristic time scales. Tendencies of these IMF are forecasted with SVR respectively, in which the kernel functions are appropriately chosen with these different fluctuations of IMF. These forecasting results of IMF are combined to output the forecasting result of the original time series. The proposed model is applied to the tendency forecasting of non-linear and non-stationary time series, and the results show that the forecasting performance of the hybrid model outperforms SVR with the single-step ahead forecasting or the multi-step ahead forecasting.

Key words: time series, empirical mode decomposition, support vector regression, forecasting

摘要： 针对非平稳、非线性时间序列变化复杂、难以用单一智能方法进行有效预测的问题，提出一种新的基于经验模式分解和支持向量回归的混合智能预测模型。经验模式分解能将非平稳时间序列按其内在的时间特征尺度自适应地分解为多个基本模式分量，根据这些分量各自趋势变化的剧烈程度选择不同的核函数进行支持向量回归预测，对各预测分量进行加权组合，得到原始序列的准确预测值。实证研究表明对于非平稳、非线性时间序列的预测，不论是单步预测还是多步预测，该模型均能取得很好的预测效果。

关键词: 时间序列, 经验模式分解, 支持向量回归, 预测