In order to forecast short-term electric heating distribution transformer load quickly and accurately, a method in which the Back Propagation (BP) neural network algorithm is intelligently combined with ensemble empirical mode decomposition(EEMD) is proposed considering the influence of weather type and temperature on the residents' heating behavior. Firstly, the daily load sequence is decomposed into four series of low-to-high frequency sub-sequences and a remnant sub-sequence by EEMD method. Secondly, each sub-sequence, temperature data and meteorological data are input into the BP neural network to predict. Finally, sum the predicted components to obtain the final prediction result. EEMD-BP combined method is applied in Coal-to-Electricity Project and forecast a certain distribution network load with a large proportion of electric heating. Simulation results show that EEMD-BP combined forecasting method can effectively reduce the prediction error.