摘要
结合数据特征及分布特点提出一种基于谱聚类的模糊时间序列自适应预测方法。首先基于谱聚类的思想,根据样本数据特征获取其所属论域的个数及范围,实现向模糊时间序列的自适应转化;然后基于Markov概率模型表示模糊时间序列中的模糊关系,从而对多步模糊关系、高阶模糊关系及模糊关系的稳态进行求解;最后获取预测值的可能模糊状态,进而利用去模糊化方法将其还原为预测值。在真实以及人工时间序列数据上的实验表明了所提方法的合理性与有效性。
A fuzzy time series self-adaption prediction method based on spectral clustering and data characteristics was proposed. First, based on spectral clustering and the characteristics of data, the number and scope of the discourses was obtained to convert into fuzzy time series self- adaptively. Then, fuzzy relationships based on Markov probability model was presented, and the multi-steps, high-order and steady fuzzy relationship are gotten. Finally, proposed meted obtained the probable fuzzy states, and got its predicted values based on defuzzification methods. Experiments on real-world and synthetic time series data indicate the rationality and effectiveness of the proposed method.
出处
《通信学报》
EI
CSCD
北大核心
2016年第2期106-114,共9页
Journal on Communications
基金
中央高校基本科研业务专项基金资助项目(No.HEUCF100603
No.HEUCFZ1212)~~
关键词
模糊时间序列
谱聚类
论域划分
Markov概率模型
模糊关系
fuzzy time series, spectral clustering, discourse partition, Markov probability model, fuzzy relationship
