基于极大熵和小波的时间序列数据预测研究

在极大熵准则的基础上,以苏州虎丘塔的历史形变观测数据为例,进行时序数据的预测分析。研究不同的样本数据选择原则对预测效果的影响。然后在一致的样本基础上,得到基于不同的模型参数估计方法的时序数据预测曲线,并通过小波的去噪分析进行不同曲线之间优劣的比较。结果表明,在1985年到2000年期间,塔体的变形周期在不断的延长,但总体而言,塔体的变形是在不断的加剧。而基于极大熵参数估计法的AR(p)模型能对实际的观测曲线有很好模拟和预测效果,选择均匀步长的样本数据对提高模型预测精度是很重要的。

Research on time series prediction based on maximum entropy and wavelet

Based on maximum entropy principle, the recorded deformation data of Tiger Hill pagoda of Suzhou is taken as an example to predict time series data. This paper analyzes the influence of different strategies of data sampling on the precision of prediction. According to the same sampling data, this paper concludes different prediction curves based on different methods of parameter estimation for various models, and compares these curves through de-noised disposal by wavelet analysis. The result indicates that the periods of deformation for pagoda between 1985 and 2000 have extended gradually, and the whole situation of deformation has being intensified. The prediction curve of AR(p) model based on maximum entropy principle can fit real curve very well, symmetrically sampled data is key to enhance the precision of model prediction.