Abstract: Power load forecasting plays an important role in the development of power system and provides important guidance for power supply.Short-term load forecasting（STLF） can ensure the security and stability of a power system in a short time.In order to solve the problems of insufficient prediction accuracy and lack of reference factors in the dataset, the paper proposes a short-term power load forecasting method based on a multi-feature extracted Attention-BiGRU（Bidirectional Gated Recurrent Unit） network.The forecasting model is the basic structure of the Gated Recurrent Unit（GRU） network.According to the data set, the method extracts time characteristics and data distribution characteristics in advance.All characteristics are used as the influencing factors for load forecasting.Then, the paper used attention mechanism to allocate the weight of the input sequence.Finally, the BiGRU network is used to learn the characteristics and output the forecasted load values.The simulation results show that the Attention-BiGRU network based on multi-feature extraction outperforms the traditional Gaussian regression method, GRU network, BiGRU network with multi-feature extraction and BiGRU network.