基于模糊逻辑关系组的时间序列模型改进

Improvement of Time Series Model Based on Fuzzy Logic Relation Group

1993年Song首次建立基于模糊逻辑关系组的时间序列预测模型,从而有效地解决了语言数据或具有模糊不确定性数据的预测问题,但至今在论域划分及模糊逻辑关系的阶数确定问题上依然存在不足.为此本文引入模糊熵确定最优聚类数目来划分论域,其次借助时间序列自相关函数解决了模糊逻辑关系的阶数确定问题,最后引入灰色理论于模糊时间序列模型中,利用灰色残差模型对模糊时间序列模型的预测值进行了修正.研究发现本文方法的预测精度均优于现有模型,并利用台湾机械行业产品价值数据进行了实证检验,效果显著.

The traditional time series prediction model is dependent on a large number of historical data, but the historical data is often incomplete, inaccurate and vague due to the widespread presence of uncertainty in practical problems, in order to solve the problems, Song first proposed time series model based on fuzzy logic relation group in 1993, but these methods are still inadequate. This paper proposed to introduce the concept of fuzzy entropy to determine the optimal number of clusters which effectively divided the domain at first, then used the concept of correlation function of traditional time series to determine the order of the fuzzy logical relationships in fuzzy time series, considering that hybrid algorithm can significantly improve the prediction accuracy of the overall model, therefore, we use the residual model to amend the prediction value on the basis of fuzzy time series forecasting results. Finally our method is used for Taiwan machinery industry product value of 1998/01- 2001/12 forecast, the results of our method and the results of existing models are compared and find that the proposed model with higher prediction accuracy.