Skip to Main Content
This paper presents a Bayesian network approach to dissolved gas analysis (DGA) problems for power transformers, which is easy to construct and able to interpret with formal probabilistic semantics. The effectiveness of the traditional IEEE/IEC coding scheme is validated using the proposed approach, which is also able to handle the missing codes in the traditional coding scheme. Firstly, the essential concepts of Bayesian networks are introduced, which are graphical representations of uncertain knowledge. The methodology of combining knowledge in the DGA domain with statistical data used to learn new knowledge is then described. A specific Bayesian network is designed to diagnose transformer faults based upon the IEEE/IEC DGA ratio method. An applicable solution to a transformer DGA problem, using the Bayesian network approach, is illustrated highlighting the potential of Bayesian networks. It can be seen from the results developed that the proposed approach is capable of tackling the DGA problem for power transformers as a supportive tool along with the IEEE/IEC DGA coding scheme.