Polynomial Trend Removal in Signals Using Back-propagation Neural Network

倒傳遞神經網路於非定常趨勢之濾除

張憲國(Hsien-Kuo Chang) ； 林立青(Li-Ching Lin) ； 林西川(Shi-Chuan Lin)

《海洋工程學刊》 6卷1期 (2006/04) Pp. 25-44

https://doi.org/10.6266/JCOE.2006.0601.02

摘要

本文應用類神經網路進行非定常趨勢的分離。使用不同的高斯亂數與一組人造二項式之非定常趨勢組成合成訊號，來測試最小二乘法、小波理論與類神經網路所求出二項式之係數，由結果顯示類神經網路與最小二乘法均能精確地分離非定常趨勢，且二者均優於小波理論之辨識能力，而且本文之模式可應用於時序列中，高階多項式趨勢之辨識。

關鍵字

類神經網路 ； 非定常序列 ； 迴歸分析

並列摘要

This paper presents a back-propagation neural network for detecting a quadratic trend involved in a time series of random signals. Synthetic data and real pressure data with a trend are used to elucidate the detection ability of wavelet-based method, regression analysis and the proposed model. Both the proposed model and the regression analysis have a comparably high accuracy to remove the quadratic trend from all sets of synthetic signals than the wavelet-based model. The present approach can be applicable for identifying a higher-order polynomial trend from a time series of real or synthetic data.

並列關鍵字

artificial neural network ； trend removal ； nonstationary time series ； regression analysis