Differentiated warning of transformer based on data mining techniques

Transformer is the key electrical apparatus in power system. The running status and defects of the transformer could be monitored on time by on-line monitoring system of dissolved gases content and producing rate. In IEC and Chinese GB standards, the attention value in warning system is calculated by statistical method, considering voltage level without other factors. In order to fill the gap, this paper calculates attention and warning value of transformer with different intrinsic properties (e.g., running period, manufacturers) by data mining techniques. According to larger amounts of data collected by DGA sensors installed in field application, this paper elevates the quality of the data firstly, and then builds construct distribution models to associate defect/fault rate with cumulative probability. Finally, the attention value and warning value could be calculated by using the inverse cumulative distribution function. In this paper, the calculation method to calculate the threshold of dissolved gases content and producing rate is proposed, and the differentiated warning method of transformer with different intrinsic properties is built up. Applying the warning method to the running transformers, the occurrences of false positive and false negative reduced obviously, and warning accuracy increased significantly.