The invention belongs to the technical field of cloud computing and distributed neural networks, and relates to a low-overhead smart power grid data aggregation method based on a distributed neural network, which comprises the following steps: (1) extracting features of terminal equipment; (2) collecting local data; and (3) processing cloud data. According to the method, the edge equipment is allowed to carry out rapid local inference, the communication cost is greatly reduced, the time cost is reduced, and meanwhile, through distributed calculation, the privacy problem in data aggregation canbe ensured, and the fault-tolerant capability of the system is improved.