Traditional trend forecasting methods are rarely applicable to the recently-installed transformers and their nearby old ones for lack of years of the former’s historical data while the load trend of the latter is distorted by the permanent load transfer. For this reason

a concept of customer group’s load is presented

which consists in decomposing a transformer’s load into components of an original customer group and several switched-in and switched-out customer groups. The least square method is first applied to fit the load trends of the various customer groups

the results of which being further used to build the transformer load forecasting models according to the load switching records. Case studies are carried out on the 110 kV/35 kV main transformers in the Songjiang distribution network. Results show that the method presented can yield better prediction accuracy than the seasonal time series method and direct regression method based on cubic polynomial functions. Furthermore

for the application of the method

historical load switches between the main transformers are not required to be exactly collected

except for three years of historical data. So the method presented is fit for load forecasting of main transformers in the substations following permanent load switching.