基于平均包络线匹配算法的EMD端点效应分析及在股价趋势分解中的应用

经验模式分解(EMD)能够有效获得非平稳非线性信号的时频特征,但传统的EMD分解算法存在严重的端点效应. 在深入研究和分析EMD算法的基础上,提出了一种基于波形匹配的端点效应处理方案,通过计算波形匹配度, 在平均包络线内部寻找与其端部变化趋势最为接近的子波,并用这段子波代替平均包络线的边缘部分, 使处理后的平均包络线极大地接近真实包络线,并把这种端点效应处理方案的EMD分解算法应用到实际的股票市场价格趋势分解中.实验结果表明,与经典的EMD边界延拓算法相比,本文提出的算法能更有效地抑制EMD分解时的边界效应, 分解得到的固有模式函数更能体现模拟信号真实的频率、幅值信息.应用实验表明:与现有方法相比,该方法更能提高预测精度.

The end effect of EMD based on matching mean envelope and its applications in trend decomposition of stock price

Empirical mode decomposition (EMD), which can extract real time-frequency characteristics from non-stationary and nonlinear signals, however, has an involved end effect in the course of getting the envelops of the signal by the spline interpolation. In this paper, a new method based on wave matching to deal with the end effect is proposed, which replaces the ends of the mean envelop with the most suited sequence in the inner envelop making the post-dealing mean envelope have the most similarity of tendency with the real one. Compared with the classical boundary extension algorithm, the improved algorithm can increasingly suppress the end effect in EMD and reflect the true original signal frequency information as well as amplitude value, and it is well used to forecast the trend of stock market prices. The experiments show that the presented method can more effectively improve the prediction accuracy.