On the Behavior of Rational Prosumers for a Holonic Smart Grid
Loading...
Files
Date
Authors
Publisher
Philipps-Universität Marburg
Abstract
Over the past century, the power grid has continuously changed in
order to meet evolving requirements for energy transmission and
power supply. But the biggest change is happening right now. The
transition from fossil to renewable energy resources, needed to achieve climate goals, is both driving and accelerating the transformation of existing grid infrastructure and management.
Simultaneously, the continuous societal digitization and the use
of computationally intense technologies increase overall electricity
demand and depend highly on a stable power grid. The power grid transformation is reinforced even more by the rise of prosumers. Distributed consumers, which are now also able to produce
electricity, are slowly causing a paradigm shift from centralized to
distributed power generation and grid infrastructure.
To guarantee stable grid operation and electricity supply with improved integration of highly volatile distributed energy resources, new data-driven methodologies are needed. In this thesis, we present the following three main contributions towards flexible and stable Smart Grid operation: First, we propose a conceptual framework for the next evolutionary step in power grid segmentation based on the holonic Smart Grid. Based on a bottom-up approach, we use supervised Machine Learning techniques for a tight monitoring and forecasting of prosumers’ electricity demand. Second, we transform a Micro Smart Grid into a Multi-Agent System, where prosumers are modeled as rational agents. Thereby, an agent is modeled in a flexible, an adjustable and an extendable way to represent different kinds of prosumers. We identify valid actions and derive a set of strategies with prosumer, grid- and market-centered goals, which are evaluated. Third, we develop, implement and evaluate a federated Machine Learning model for all prosumers within a Smart Grid with respect to data-privacy preservation. The proposed distributed approach based on a Federated Learning model addresses both data-privacy concerns and the need for sufficient data needed for developing accurate electricity consumption models. In short summary, these contributions form a flexible framework capable
of guiding and optimizing single- and multi-prosumer behavior
within a holonic Smart Grid.