Computational Performance Study on the Alternating Direction Method of Multipliers Algorithm for a Demand Response Peak Shaving Application

Abstract

This article quantifies and benchmarks the computational performance of the distributed alternating direction method of multipliers (ADMM) optimization algorithm applied to a slow-dynamics demand response peak shaving application. We propose a hierarchical demand response architecture, in which a commercial aggregator acts as the supplier and system-level flexibility service provider to a portfolio of residential prosumers. The prosumers are equipped with flexible distributed energy resources that we model based on convex operation constraints. The optimal day-ahead peak shaving control of the portfolio is realized in a distributed way through an entirely parallel implementation of the distributed ADMM optimization algorithm. In this scenario, we evaluate the ADMM algorithm's overall speed of convergence and conduct multiple large-scale numerical simulation experiments on a compute cluster to compare and assess the algorithmic performance of ADMM against a centralized reference benchmark optimization. Our simulation results show that the distributed ADMM algorithm is superior compared to a purely centralized optimization. The findings are of particular use to commercial demand response aggregators that show interest in the deployment and improvement of distributed coordination and optimal control concepts for future slow-dynamics demand response applications and services.