Distributed-MPC with Data-Driven Estimation of Bus Admittance Matrix in Voltage Control

Abstract

This paper presents a distributed model-predictive control (MPC) design for real-time voltage stabilization in power systems, allowing the bus admittance matrix Y = G + jB to be not known a priori (it may be time-varying), and so is estimated online. The prevalent control designs are either centralized and optimal but lack scalability and are subject to network attacks, or decentralized that are scalable and less vulnerable to attacks but are suboptimal. The proposed distributed solution offers the attractive features of both types of schemes, namely, optimality, scalability, as well as enhanced security to network attacks. In addition, since acquiring the exact knowledge of the line conductance and susceptance, which are required to form Y, is in general challenging, the presented framework integrates data-driven estimation of Y to circumvent this challenge. We first introduce the centralized version of the formulation, and next transfer it to a distributed version for efficiency, scalability, and attack-resilience, leveraging the graph structure of the power system. Only local computation and communication are used for (i) computing local control via distributed optimization (which is solved by the alternating direction method of multipliers ADMM), as well as (ii) data-driven estimation of Y. The performance of the proposed methodology is validated using numerical examples of the IEEE-30 Bus and IEEE-57 Bus systems.