Secure Multi-Party Household Load Scheduling Framework for Real-Time Demand-Side Management
Lilin Cheng, Haixiang Zang, Zhinong Wei, Guoqiang Sun
Abstract
Renewable power sources are being increasingly incorporated into distribution networks. Therefore, demand-side management (DSM) has become more critical for improving system reliability. Currently, decentralized real-time DSM is practicable based on home energy management system (HEMS). However, coordinating these HEMSs is difficult because DSM customers may not wish to communicate with each other due to competition and privacy contents. A new peak may even emerge in the aggregator if HEMSs shift their loads without proper coordination. Hence, a secure multi-party household load scheduling framework was proposed in this study to ensure encrypted data sharing between HEMSs based on homomorphic encryption (HE) technology. In order to solve decentralized real-time DSM by directly using additive HE data in this proposed framework, a reinforcement learning (RL) method, namely boosting tree-based deep Q-network, was developed to be trained on a distributed algorithm. The results of case studies revealed that the proposed data-sharing framework outperformed the conventional DSM in shaving peak loads of the aggregator, whereas the electricity cost of each customer did not increase. Moreover, the proposed RL method protected the privacy of users and obtained a similar result compared with no-privacy-preserving RL methods.