Online energy disaggregation is an advanced technology that can help both consumers and the utility to implement load components analysis, enhancing the reliability of demand-side management. However, the online system requires continuous access to personal electricity data to train an online machine learning model, which would infer personal information. In this paper, we introduce a privacy-preserving online energy disaggregation system, the online model and training data can be protected by adding Gaussian noise to the model during the training process.