Abstract: As an important basis of power system operation planning，short-term power load forecasting is

great significant to the safe and economic operation of power system. A long-term and short-term time series

network（LSTNet）model was proposed to predict the short-term load variation of distribution area. The model

used convolutional neural network（CNN）to extract local dependencies between load data，and long and short term memory（LSTM）neural network to extract the long-term trend of load data，and then integrated the traditional autoregressive model to solve the problem that the neural network was insensitive to the extreme value of load data.Finally，the power load data of a distribution area was used in the network training and prediction process.Discovered by simulation experiment case，compared with LSTM，Bi-LSTM and CNN-LSTM prediction models，LSTNet model has more advantages and higher prediction accuracy in short-term load forecasting.

      Key words: short-term power load forecasting；long-term and short-term time series network（LSTNet）；long

and short term memory（LSTM）neural network ；convolutional neural network（CNN）；autoregressive model