Learning-to-optimize infused decentralized disaggregation for multi-entity technical VPP considering equity and privacy

The growing renewable energy integration challenges grid stability, highlighting demand response as a critical solution. 5G base stations emerge as a valuable demand-side response resource due to their inherent flexibility potential. Virtual Power Plants, especially Technical VPPs (TVPPs), play a crucial role in effectively aggregating these base stations and other resources to enhance grid flexibility. However, TVPPs encounter significant obstacles in achieving rapid disaggregation, concerning real-time requirements, energy equity and privacy security. This paper presents a novel instruction disaggregation algorithm addressing these issues through a decentralized coordination mechanism. Firstly, a multi-entity TVPP instruction disaggregation model is established by incorporating both grid security and energy equity, and distinct in considering energy equity regarding the disaggregation process. Then, a decentralized disaggregation method based on learning-to-optimize is proposed where self-supervised learning is embedded to train surrogate models so that the computational time due to traditional optimization can be significantly reduced. Moreover, a lightweight privacy-preserving scheme is integrated to avoid privacy breaches without introducing excessive computational burdens. Finally, theoretical guarantees for the proposed algorithm are established, including solution quasi-feasibility, convergence and generalization properties. Case studies show that the proposed method significantly decreases computational demands, achieving speedup ratios of two orders of magnitude compared to the traditional decentralized method while ensuring privacy security.