A Load Aggregation Commercial Electricity Prediction Method Supporting User Privacy Protection

Abstract

With the development of the power industry, the number of two-way interactive power consumption units continues to increase, and the burden of power distribution on the grid continues to increase. The load aggregator is located between the power grid and the power consumption unit, and represents a certain range of users to request power from the power grid, and then performs power distribution and real-time scheduling to users, reducing the burden on the power grid. To better respond to the demand of electricity consumers, machine learning is used in electricity forecasting. However, the training data comes from power users, which will involve privacy protection issues. To this end, this paper proposes a load aggregator power prediction method that supports user privacy protection. This method can take into account the influence of user-related fixed factors and time-related variable factors on power consumption. The load aggregator is the aggregator. There is no need for power users to share their own data, and the necessary model parameters are passed to the load aggregator only when needed, and it can still be carried out for some participants without labels. Finally, the proposed method is evaluated through experiments, and the results show that the method in this paper can effectively protect user privacy, and has a considerable accuracy compared with the existing machine learning methods that do not protect privacy.