Analysis and evaluation of a distributed optimal load coordination algorithm for frequency control.

Highlights • The modified Distributed Gradient Projection algorithm with a constant step size can handle non-strictly and convex consumer costs. • We establish a bound on the step size such that the algorithm behaves in a stable manner. • Numerical tests with the IEEE 39-bus system show that predictions are still accurate even when consumer costs are non-convex. • The algorithm is robust to model mismatch and time delays. • Its performance suffers if the dynamic response of the loads is slow. Abstract The Distributed Gradient Projection (DGP) algorithm was proposed in prior work to allow loads to provide contingency service to the grid using local noisy frequency measurements by varying their demand. Convergence of DGP was established in prior work for a decaying step size. In this paper we modify the algorithm to using a constant step size—constant step size being much more useful for practical implementation. We provide a convergence analysis of the modified algorithm, which we call DGP-C (DGP with Constant step size) and perform extensive simulations in the IEEE 39-bus test system. These studies (i) demonstrate that the DGP-C algorithm is robust to several assumptions made in the analysis and (ii) reveal which factors among the many tested (measurement noise, loads' response time, etc.) have significant effect on the algorithm's performance. [ABSTRACT FROM AUTHOR]