Data Protection of AMI Based on Oblivious Inference and Deep Learning

In recent years, deep learning has been widely used in the analysis and mining of electricity information data. However, the power consumption data input to the deep learning model contains the private information of the power user. If the power consumption data is directly exposed to the server, there will be a security risk of leakage of power user information. In order to solve this problem, inadvertent inference was proposed. Inadvertent inference allows users to obtain neural network inference results without exposing input data to the server, thereby protecting user privacy. Based on the confusion circuit, this paper implements an inadvertent inference deep learning method based on XNOR-Net to analyze data of advanced metering infrastructure system. Our method improves the operation speed while ensuring the accuracy of network inference as much as possible.