Abstract

Collective photovoltaic self-consumption is an extension of traditional, single user self-consumption, whose objective is to maximize the share of local generated energy that is consumed at the generation point with multiple consumers sharing a photovoltaic system. With demand-side management, consumption profiles can be adapted to generation profiles which can be posed as an optimization problem. In this work, the problem of scheduling uninterruptible shiftable devices is addressed from a multiagent perspective considering two different approaches. On one hand, a centralized architecture assumes that a single agent has the capacity to solve all agents problems. On the other, a new partially decentralized architecture is proposed to solve the same problem by distributing decision-making among agents using an heuristic algorithm with a virtual dynamic tariff as coordination mechanism. A computational experiment is set up in order to test increasing numbers of agents. Results show that the centralized approach, although finding the optimal solution, shows running times that are two orders of magnitude greater than the decentralized method, which is able to find high quality solutions at lower time cost but with increased monetary cost. This trade-off becomes more relevant as the number of agents increase.