Abstract: In order to solve the problem of uncertainty and improve the prediction accuracy, a probability prediction of residential load based on deep mixed density network is proposed. An end-to-end convolutional neural network combined with gated recursive unit was designed. A loss function was reconstructed to prevent large errors caused by indirect structure propagation and improve the calculation efficiency. Further, the designed depth model was combined into the hybrid density network to directly predict the probability density function. The case results show that compared with other methods, this method has certain advantages in residential load probability prediction.