As the transformer top oil temperature is affected by many factors such as weather conditions and tidal current loads,it is difficult to improve the forecasting accuracy. To solve this problem,a method of transformer top oil temperature forecasting based on similar day and similar hour is proposed,which is to further select the similar hour corresponding to each hour of the day to be forecast within the similar days,and then use the similar hour to forecast transformer top oil temperature. Firstly,K-means clustering based on meteorological factors and the principle of'near big,far small' are used to select similar days from the historical samples. On the basis of the definition and description of similar hour,the calculation steps of the oil temperature forecasting method are given by using back propagation (BP) neural network and linear weighted method,which is applied to top oil temperature forecasting of a ultra-high voltage main transformer in Jiangsu. Finally,the results show that the proposed method has high accuracy on forecasting top transformer oil temperature,which verifies its feasibility and validity.