A Transformer Latent Fault Warning Strategy Based on Self-Adaptive Threshold Values

Dissolved gas analysis(DGA) is considered as the most effective method to detect latent fault of transformers. But the stationary threshold values are tend to decrease the sensitivity of the fault early warning. In this paper, a transformer latent fault early warning strategy is proposed based on self-adaptive threshold values. First, upper bound and lower bound of historical gas concentration data is obtained by kernel – smoothing method. Then, the future threshold values could be obtained through a self-adaptive prediction model based on PSOGSA-kELM. Moreover, the warning strategy is made based on the comparison between new detected data and predicted threshold values. By the comparison between other existed warning methods, the feasibility of proposed warning strategy is testified.