Abstract：A fusion prediction method was proposed to predict and correct the calculation deviation of the top

transformer oil temperature model in IEEE guideline，so as to realize the more precise prediction of the transformer

top oil temperature（TOT）. Firstly，the characteristics of the transformer TOT model and the extreme learning

machine（ELM）prediction model was introduced. Secondly，in order to avoid the problem of slow operation speed

caused by double level intelligent prediction，the weighted multi-point extrapolation method combined with the

load curve clustering algorithm was used to obtain the future load coefficient of the transformer which introduced

as the load prediction level of the model. Finally，based on the calculation of thermal model，which the ELM was

used to predict the deviation between the calculated value of thermal model and the measured value，and finally the accurate predicted value of the TOT of the transformer was obtained.The simulation platform was built and the

simulation results show that the average prediction error rate of the proposed prediction method is only 0.59%，and

the root mean square error is only 0.47 ℃. Compared with the other three methods，it has higher prediction accuracy and stability. The model training speed and prediction speed are only 1.21 ms and 0.39 ms，respectively，which proves that the fusion prediction model proposed and established has high prediction accuracy，stability and

operation speed.

      Key words：transformer top oil temperature（TOT）；extreme learning machine（ELM）；thermal model；fusion prediction；load morphology clustering