时序环境中概率因果关系的定性表示及融合

Qualitative representation and fusion of probabilistic causalities in time-series environments

时序数据库各时间片的数据中蕴含着概率因果关系,多时间片概率因果关系的融合可为时态环境中因果关系的综合分析、全局考虑、决策支持及趋势预测等提供科学的依据,该文研究多时间片内同一变量集上概率因果关系的定性表示及融合.首先将各时间片的贝叶斯网(BN)抽象为定性概率网(QPN),以定性的方式高效、简洁地表示各时间片数据中蕴含的概率因果关系.然后,基于BN的推理方法及时序环境中的马尔可夫假设,构造多时间片QPN的一致表示,从而解决各时间片QPN的结构融合问题.基于证据叠加和QPN中定性影响加权的基本思想,给出基于区间值的QPN定性影响量化及相应的定性影响叠加方法,从而解决各时间片QPN的参数融合问题.实验结果表明所提出方法的可行性.

Probabilistic causal relationships are implied in the data of various time slices in time series databases. The fusion result of probabilistic causalities among multiple slices cbuld provide with the reasonable guideline for decision making, knowledge fusion, analysis, prediction of economic trends, etc. In this paper, we focus on the qualitative representation and fusion of probabilistic causalities on the same set of variables in multiple time slices. To represent the causalities efficiently and succinctly, we first construct QPNs abstracted from corresponding BNs in different time slices. Further,based on the reasoning approach of BN and the Markov assumption in time - series environments,we obtain the consistent representation of various network structures,and then propose the method for fusing the graphical structures of concerned QPNs. Consequently,we associate strengths to qualitative influences by interval values derived from BNs＇ probability parameters. Then we give a superposition method for fusing qualitative parameters of time - series QPNs. Experimental results verify the feasibility of our methods.