A Fast Distributed Algorithm for Large-Scale Demand Response Aggregation

TL;DR

This paper introduces a fast, scalable distributed algorithm for large-scale demand response aggregation that efficiently handles complex household energy models and converges quickly regardless of system size.

Contribution

The paper presents a novel nonconvex demand response algorithm decomposing at the household level, enabling efficient aggregation of large systems with mixed energy device types.

Findings

Algorithm terminates in 60 iterations regardless of system size

Converges to near-optimal solutions with minimal parameter tuning

Demonstrates scalability on systems with up to 2560 households

Abstract

A major challenge to implementing residential demand response is that of aligning the objectives of many households, each of which aims to minimize its payments and maximize its comfort level, while balancing this with the objectives of an aggregator that aims to minimize the cost of electricity purchased in a pooled wholesale market. This paper presents a fast distributed algorithm for aggregating a large number of households with a mixture of discrete and continuous energy levels. A distinctive feature of the method in this paper is that the nonconvex DR problem is decomposed in terms of households as opposed to devices, which allows incorporating more intricate couplings between energy storage devices, appliances and distributed energy resources. The proposed method is a fast distributed algorithm applied to the double smoothed dual function of the adopted DR model. The method is tested on systems with up to 2560 households, each with 10 devices on average. The proposed algorithm is designed to terminate in 60 iterations irrespective of system size, which can be ideal for an on-line version of this problem. Moreover, numerical results show that with minimal parameter tuning, the algorithm exhibits a very similar convergence behavior throughout the studied systems and converges to near-optimal solutions, which corroborates its scalability.