A Privacy-Preserving Online Learning Approach for Incentive-Based Demand Response in Smart Grid

Abstract

Incentive-based demand response (IDR) programs enable smart grid customers to participate in the demand reduction, triggered by system contingencies or peak load, to improve reliability, sustainability, security, and efficiency of the power grid. However, the deployment of smart meters and the increasing number of customers in IDR programs make the data generated in the smart grid at a large scale. Meanwhile, fine-grained data from smart meters can deluge customers’ lifestyle and usage pattern, posing threat to customer privacy. Therefore, privacy-preserving demand source management techniques that support increasingly large-scale datasets are in urgent need. In this paper, we propose an online privacy-preserving IDR management system, in which social welfare is maximized through recommending the optimal consumer to the utility company. Since the contexts of electricity curtailment offers from the utility company are different, an adaptive context partition method is proposed to enable the system context awareness. In addition, we cluster the customers in a tree structure to make the analyses of the customers in the cluster level and thus enable the algorithm to support the large-scale system. Furthermore, a tree-based noise aggregation method is applied to guarantee both the differential privacy of customers’ sensitive information and the utility of the data. Theoretical analysis shows that our proposal guarantees differential privacy of customers, while converging to the optimal policy in a long run. Numerical results validate that our proposed algorithm supports the large-scale dataset while striking a balance between the privacy-preserving level and social welfare.