Complex-demand Scheduling Problem with Application in Smart Grid

TL;DR

This paper addresses the complex-demand scheduling problem in smart grids, proposing approximation algorithms with theoretical guarantees based on demand angles, and a practical greedy solution for single-slot scenarios.

Contribution

It introduces approximation algorithms for complex-demand scheduling with bounds depending on demand angles and offers a practical greedy algorithm for real-world applications.

Findings

PTAS for demands with angle up to π/2

Bi-criteria FPTAS for demands with angle up to π-ε

Efficient greedy algorithm with 1/2 cosine ratio for single time slot

Abstract

We consider the problem of scheduling complex-valued demands over a discretized time horizon. Given a set of users, each user is associated with a set of demands representing different power consumption preferences. A demand is represented by a complex number, a time interval, and a utility value obtained if it is satisfied. At each time slot, the magnitude of the total selected demands should not exceed a given generation capacity. This naturally captures the supply constraints in alternating current (AC) electric systems. In this paper, we consider maximizing the aggregate user utility subject to power supply limits over a time horizon. We present approximation algorithms characterized by the maximum angle $\phi$ between any two complex-valued demands. More precisely, a PTAS is presented for the case $\phi \in [0,\tfrac{\pi}{2}]$, a bi-criteria FPTAS for $\phi \in [0,{\pi} \mbox{-} \varepsilon]$ for any polynomially small $\varepsilon$, assuming the number of time slots in the discretized time horizon is a constant. Furthermore, if the number of time slots is part of the input, we present a reduction to the real-valued unsplittable flow problem on a path with only a constant approximation ratio. Finally, we present a practical greedy algorithm for the single time slot case with an approximation ratio of $\tfrac{1}{2}\cos \frac{\phi}{2}$ and a running time complexity of only ${O}(N\log N)$, $N$ standing for the aggregate number of user demands, which can be implemented efficiently in practice.