基于小波分析与支持向量机的时间序列预测

时间序列广泛存在于工业、经济、军事等各个领域,时间序列预测是数据分析处理的一个重要方面。目前提出的预测模型大多基于＂原始时间序列是无噪的＂这一假定,而实际应用中,对时间序列去噪处理的好坏将直接影响预测的准确率,针对这一事实,使用小波分析对原始时间序列去噪。利用小波变换对时间序列进行多尺度分解,对各尺度上的细节序列使用阀值法去噪;使用支持向量机对重构后的各组小波系数进行预测并将结果融合,得到预测结果。实验结果表明,用于时间序列预测,能及时反应序列的变化趋势并具有较高的预测精度。

Time Series Prediction Based on Wavelet Analysis and Support Vector Machine

Time series is widespread in the industrial,economic,military fields and so on.Predicting the time series is one of the important aspects of data analysis and treatment.For the moment,most predicted models are based on the assumption that the original time series doesn′t contain noise,but in the practical application,if the original time series couldn′t be denoised properly,the accuracy of the prediction would be affected greatly.This paper uses wavelet analysis to denoise the original time series.Wavelet analysis could be utilized to analyze the time series in multiple scales and then the threshold method is used to denoise the detailed sequence in each scale;support vector machine(SVM) is applied to predict the reconstructed wavelet coefficient of each group and fusing all the predictions of them,the predicting results are got.Experimental results show that this method for time series prediction could timely response to the trend of time series and has high prediction accuracy.