Integrated transformer condition assessment based on SOM neural network evidence cloud element model
Lingzhi Yi, Huang Yu, Yahui Wang, njian Xu, Haixiang She, Xuanli Su, Ganlin Jiang · International Journal of Advanced Mechatronic Systems · 2024
Aiming at some problems in the process of transformer evaluation, a comprehensive transformer state evaluation method based on SOM neural network evidence cloud object element model is proposed. Firstly, the random factor optimisation combination weights are adopted to solve the problem of expert randomness and incomplete data characteristics of current weight calculation method. Secondly, filtering algorithms and threshold learning mechanisms are introduced to optimise the SOM neural network for clustering. It can solve the problems of randomness and subjectivity in the previous transformer evaluation class intervals. Thirdly, improving the D-S evidence theory based on Pearson's correlation coefficient. It can fuse different characteristic indicators of transformers to avoid contradiction when fusing high conflict evidence. Lastly, the cloud entropy optimisation algorithm applies to improve the cloud object element model to determine the final assessment results. Take 6 actual transformers as an example. It proves the effectiveness and accuracy of the method, and applies to oil-immersed power transformers of different service periods.