Hot spot temperature prediction model of transformer based on GP-NLMS

Abstract

An accurate estimation model of winding hot spot temperature is the key to assess the thermal state and insulation life of oil-immersed transformers. Based on the hot spot temperature and load current monitored by the substation,the genetic programming algorithm is applied to train the basic structure of the hot spot temperature estimation model. Then,the parameter identification of the hot spot temperature estimation model is performed by the normalized least square mean (NLMS) algorithm. Finally,an explicit prediction model of the hot spot temperature is established for oil-immersed transformers. The explicit winding hot spot temperature estimation model can effectively reflect the relationship between the load factor and the winding hot spot temperature. Moreover,the goodness of fit of the model under the test set is 0.998 8,and the maximum absolute error is only 1.36 ℃,which verify the correctness and effectiveness of the model. Furthermore,the strong generalization performance of the proposed model is proved by estimating the winding hot spot temperature for oil-immersed transformers with the same capacity and model in the same area.

Keywords

• oil-immersed transformer
• hot spot temperature
• genetic programming
• normalized least mean square (nlms) algorithm
• insulation life
• explicit prediction model