Abstract: The accuracy of short-term electricity load forecasting is important for the smooth and efficient operation of smart grids, but the strong non-smoothness and random fluctuation of the load data under the influence of various factors make the high-precision short-term power load forecasting challenging. To fully explore the trend features and the periodic features in the load sequence, accurately extract the auxiliary information features that have a potential correlation with the power load, and improve the short-term power load forecasting accuracy, this paper proposes a short-term power load forecasting model based on the N-BEATS and the auxiliary encoders. The model consists of two parallel encoders, a load feature encoder based on the neural base extension analysis model(N-BEATS) and an auxiliary information encoder based on the multi-headed attention mechanism, which are used to learn the temporal features and the auxiliary information features in the load data respectively. At the same time, a feature fusion module is constructed to construct the temporal and the auxiliary information features into a joint feature vector, and a prediction decoder module based on GRU units is designed for short-term electricity load forecasting. Experiments are conducted on the GEFCom2014 public dataset, and the results show that the proposed method has significant advantages in terms of prediction accuracy compared with the baseline models such as the long and short-term memory(LSTM) network model, the CNN-LSTM network model, the Seq2Seq network model, the seasonal autoregressive differential moving average (SARIMA) model, and the support vector regression(SVR) model, and the MAPE index is improved by 24.16% on average.