GPU Optimisation of an Endogenous Peer-to-Peer Market with Product Differentiation

Abstract

With the increase of renewable energy and the consumer flexibility needed in the energy transition, grid management will become more and more complex. Solutions for a decentralized market and handling of the physical state of the network already exist in the literature. But when simulated on a single machine for simulation purposes, these implementations do not scale up well with the problem size. This makes it impossible for researchers and system operators to tune parameters and check for stability, robustness, and dimension in a large grid. In this contribution, we optimized an endogenous Peer to peer market on a Graphic Processing Unit to allow the scaling of these algorithms. The computation times between the Central and Graphic Processing Units are divided by more than 25 with 600 agents, and the speed-up increases with the problem size. Some algorithmic and implementation issues are highlighted for such extensive problem dimensions.