优选时间序列数据模型的人工智能算法

Study on proper model for mining time series data based on artificial intelligence method

为了解决大量文献对时间序列数据的建模、仿真及预测的研究存在不做模型选择,直接应用某种模型进行分析的局限,针对经常应用于时间序列分析的3种人工智能模型:隐马尔可夫模型、人工神经网络模型和自回归移动平均模型进行基于仿真比对方法的模型选择研究。简述3种模型的基本原理,对各类模型进行数值仿真,考察各类模型生成时间序列的特征,以该特征为依据,提出模型选择的理论与算法,应用本文的模型选择理论和算法,进行实证分析。实验结果表明,各类模型生成的时间序列数据有不同的数理特征,其提供了模型选择的依据,同时,该选择理论是实用的,应用该理论选择的模型有较好的拟合和预测效果,为时间序列类型的数据挖掘提供了依据。

Lots of traditional papers have the flaws on selecting a proper model for given time series data.To prevent abusing the wrong model,a study based on the simulation method is operated which aims at comparing three influent time series data models——hidden Markov model,artificial neuro-network model and auto-regression and moving average model.Firstly,the principles of these models are reviewed briefly.Secondly,the theory and algorithm of selecting the proper model are given after the first step for grasping the data characteristics.Finally,a stimulation empirical analysis is given as a use of the above theory and algorithm.The simulation results show that different models can produce different time series data,and based on the different properties the theory and algorithm is useful and successful.