1: \begin{abstract}

2: Distributed power dispatch algorithms enable privacy preservation for power system agents: each agent optimizes its local optimization by means of primal-dual communication without disclosing its sensitive data, e.g. generation cost or load profile. However, sensitive information can be inferred by a potential adversary from responses of agents to communication signals. To ensure information integrity, this paper leverages the concept of deferential privacy to develop privacy-preserving distributed algorithms for optimal power flow (OPF) problem. We first decentralize the OPF problem using consensus alternating direction method of multipliers (consensus-ADMM), and then introduce two methods to provide differential privacy: dual and primal variable perturbations. These perturbations are Lalpalcian and integrated into local optimization of each agent, such that inferring agent data from primal-dual communication conforms to random guessing. We show that dual perturbations incur smaller optimality losses over primal ones. Furthermore, dynamic perturbations over iterations further reduce optimality loss, though the convergence with static perturbations is stronger. To suggest informed privacy-preserving designs of distributed OPF, we study the trade-off between privacy and optimality loss induced by perturbations.

3: \end{abstract}

4: