一种增加先验知识库的贝叶斯网络推理模型

An Inference Model for Bayesian Network with Prior Knowledge Base

贝叶斯网络作为一种不确定知识表示网络,由网络结构和各节点的条件概率表组成,在解决系统决策问题方面具有先天的理论优势。目前在大多数贝叶斯网络的应用中,各节点条件概率表的产生均是以人工输入的方式完成,这在一些拥有较多网络节点的复杂背景中,需要巨大的人工消耗,效率低下。针对这一问题,提出一种增加先验知识库的贝叶斯网络推理模型。根据具体的建模问题创建先验知识库,在该先验知识库下对网络节点进行类别标记,然后根据局部马尔可夫性自动生成各节点的条件概率表。在贝叶斯网络推理任务中,使用在精确推理任务中处理速度快、应用最为广泛的联结树算法,并使用Hugin算法完成消息的传递。最后通过一个贝叶斯网络实例验证了整个模型的处理流程。

As an uncertain knowledge representation network,Bayesian network is composed of network structure and conditional probability table of each node and has an innate theoretical advantage in solving decision problems.At present,in most applications of Bayesian network,the generation of conditional probability table of each node is completed in the form of manual input,which requires huge labor consumption and low efficiency in some complex backgrounds with many network nodes.To address this problem,an inference model for Bayesian network with prior knowledge base is proposed.A prior knowledge base is created for the specific modeling problem,based on which the network nodes are labeled with classes,and then the conditional probability table for each node are generated automatically according to local Markov properties.To infer Bayesian network,the clique tree algorithm with efficiency and popularity in precise inference is adopted,and the Hugin algorithm is utilized to make message transmission.Finally,the entire processing flow of the model is verified by a Bayesian network example.