Abstract: Due to the large-scale integration of distributed power sources with uncertainty and intermittency in the distribution network, the complexity of the distribution network has significantly increased, making self-healing control more difficult. Therefore, this study proposes a self-healing control method for distributed power distribution networks based on multi task graph convolutional networks. Represent the distribution network containing distributed power sources as a heterogeneous graph and use it as input to a multi task graph convolutional network model. In the model, a graph convolutional layer is used to extract shared features of heterogeneous graphs, and combined with two sub task output layers: Fault area isolation and localization, and non fault area power restoration, self-healing control of distributed power distribution networks is achieved. The experimental results show that this method can prioritize improving the accuracy of fault location and the reliability of power supply recovery in distribution networks with distributed power sources.