Research on fault diagnosis method based on Temporal Bayesian Network

This paper is based on the traditional network of bayesian, introducing the concept of the time window, to construct the Temporal Bayesian Network (TBN) model. In addition, optimizing TBN model reasoning algorithm to process system data which has the temporal haracteristic. Then we can obtain the posterior probability of each node in the TBN and use importance ranking method to determine the fault retrieval sequence, optimize of fault diagnosis process.Combined with the state power line fault diagnosis example to verify the correctness of this method. Introduction With the continuous development of the complex system, we need to know the real-time state of the complex system. When the system occurs the failure, we can identify the failure location and maintain as soon as possible, but the traditional fault diagnosis methods have been unable to meet the need. The bayesian network has the ability to deal with uncertain information, coupled with the introduction of the concept of time window, real-time monitoring system status and when the system failure occurs, we can determine the fault location at the first time, and provide efficient fault retrieval sequence, to optimize the fault diagnosis process. Time window In the TBN, the nodes are intertwined, and the observation of any node or the interference to any node observations will affect the other nodes in the bayesian network. The short distance of the recent evidence has a greater impact on the reasoning results of the time slice, and vice versa. Thus, if you want to compute a posterior probability of a hidden variable or multiple hidden variables on a time slice, you can use the evidence of the time slice, the evidence before the time slice, and the evidence after the time slice to calculate its posterior probability. Because it is not using all the evidence of observation, the result is an approximate reasoning.In TBN, the reasoning is carried out only using TBN composed of successive time slices and the,forward information propagating to the network, the time window consist of multiple time slices. TBN model The TBN is developed from the static bayesian network, where each factor in the environment is represented by a random variable, and the ever-changing system environment is modeled in this way. The relationship between these variables can describe how the state of the system changes slowly over time. The process of the system state changing can be regarded as a set of random states of the system at each discrete point in time, and each time point is called a time slice, and each time slice can be regarded as a static bayesian network. In order to avoid assigning different conditional probability tables to each time slice, it is assumed that the environmental state of the system is evolved by a steady-state process, that is, the process of system changing is caused by the system itself, not dominated by the the regular pattern of the time factor. Under the assumption of steady state, the network structure in each time slice is same. Therefore, the research process only 5th International Conference on Mechatronics, Materials, Chemistry and Computer Engineering (ICMMCCE 2017) Copyright © 2017, the Authors. Published by Atlantis Press. This is an open access article under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/). Advances in Engineering Research, volume 141