基于免疫聚类与HMM的时序信息系统决策

本文针对时间序列数据的符号化问题，提出采用免疫聚类算法处理多维时间序列的符号化，利用克隆选择原理，生成能充分反映数据真实分布的记忆抗体作为符号集合. 时间序列信息系统中的决策问题的关键是有效地挖掘历史数据中包含的时序信息. 本文提出了一种改进的隐马尔科夫模型，运用最大熵原理对模型进行训练，求取熵最大化的概率分布，并将其应用于时序信息系统的决策. 通过实验验证了其有效性．

DECISION-MAKING IN TIME-SERIES INFORMATION SYSTEM BASED ON IMMUNE CLUSTERING AND HMM

For the problem of symbolization of time series, the algorithm of immune clustering is adopted to process the symbolization of time series with multi dimension. By using theory of clonal selection, the memory antibody set, which can reflect the real distribution of data, is obtained and used as symbol set. Furthermore, the key problem of decision making in time series information system is how to effectively mine the time order information in history data. Therefore a modified hidden Markov model (HMM) is proposed for decision making, and the maximum entropy principle is adopted to train the model and calculate probability distribution with maximum entropy. The effectiveness of these methods is proved by an experiment.