Due to the limitations of land, power generation resources and other conditions, modern large-scale cities are mainly powered by external power grids, forming a large receiving end city power grid. After the large-scale urban power grid is disconnected from the main system, the power imbalance will lead to a decrease in frequency, which may even cause the network to collapse. It is of great theoretical and practical significance to maintain stable switching to form local isolated network power supply. Coordination control of network-source-load is the key to ensure stable switching of local isolated networks. However, it is difficult to realize centralized coordinated control for local isolated networks, especially for unplanned local isolated networks. In view of this, this paper introduces a distributed multi-agent system and proposes a distributed coordination control method of network-source-load without centralized controller. Apart from the traditional centralized control method which relies too much on the centralized controller and complex communication structure, the distributed coordination control obtains the global parameters of the grid operation through the improved average consistency algorithm under the condition of only relying on the communication of the adjacent nodes to realize the network-source-load global collaborative control. The proposed control method can adapt to the changes of network structure and tap the coordination potential of network-source-load. Meanwhile, the improved average consistency algorithm has better convergence performance, and provides fast global data support for frequency and voltage control of local isolated network switching. The network-source-load coordination control technology is analyzed and verified under typical working conditions based on PSCAD/EMTDC and Matlab simulation platforms. The results show that the proposed improved average consistency algorithm and distributed network-source-load coordination control method can guarantee the stable switching of local isolated networks. This work is supported by National Natural Science Foundation of China (No. 51607036).