基于属性选择的因果网络多传感器融合系统

Feature selection based causal network algorithm

针对粗集“简化”在实际应用中存在的问题提出了“统计简化”的定义和相应属性搜索算法。利用此算法对一个水域污染监测信息表进行属性简化 ,结果显示与常规算法相比 ,此算法得到的结果能够覆盖最大数量的对象 ,更不易失配。利用简化结果对上述数据融合系统建立了因果网络模型 ,实验表明 ,在保持原模型搜索正确率的同时 ,新模型压缩了搜索空间 ,提高了搜索效率。此外 ,为便于因果网络的建立导出了因果连接强度的粗集表达式。

A statistical definition of the reduct is propose and a RS feature selection algorithm upon the definition is developed. A water pollution multisensor fusion system is described by the causal network model.Comparativetestshowsthatwith the selected features, the computation time of the causal network searching algorithm is greatly saved, at the same time the classification accuracy is maintained. Also it shows that the causal strength of the causal network model can be derived from the information table by utilizing rough set theory.