A distributed multi-agent MSBN: The implementation and a generic framework

A Multiply Sectioned Bayesian Network (MSBN) is an extension to Bayesian networks for flexible modeling and cooperative multi-agent probabilistic inference. It provides an approach for probabilistic reasoning in distributed systems. This solves the problem that current Bayesian networks cannot be used in distributed systems. During the construction or dynamic formation of a multi-agent MSBN, automatic verification of the acyclicity of the overall structure is desired.;The verification of an MSBN and the evidential inference have previously been implemented in a centralized environment; however, new issues, such as those related to communication, must be resolved in a distributed environment. For this purpose, distributed algorithms are proposed in this study for implementing the verification and inference, the processes of which involve communication. Besides, a generic framework for the implementation of a distributed MSBN is also constructed such that most modules can be shared when verification, triangulation and inference of distributed MSBN are implemented. This generic framework also provides a methodology for similar problems which involve communication.;In order to test the implementation of the verification and inference of a distributed MSBN, we constructed a multi-agent MSBN by modeling a piece of equipment. The experimental demonstration of the verification and the inference processes shows the effectiveness of the new algorithms.