Summary Description
Energy systems are undergoing fundamental changes, having to deal with increasing uncertainty, digitalization, and decentralization. This makes them a natural area for multi-agent systems research, given the increasingly decentralized way in which they are planned and operated. Decentralized AI-based solutions are playing an increasingly important role in energy systems management and control, yet building successful AI applications for this domain requires both an understanding of the fundamental AI techniques, but also of the features and constraints of energy systems. In this tutorial, we will look at the fundamental opportunities and challenges in applying AI, game theory and multi-agent techniques to this key domain. The tutorial is aimed at both AI researchers (with a particular focus on PhD students and young researchers) wanting to learn more about challenges in applying their work to this area, as well as practitioners from the energy sector, exploring the potential of multi-agent solutions.Overview
In recent years, energy systems have been experiencing rapid changes. These include the increasing generation coming from intermittent renewable resources, new and increasing sources of demand (such as electric vehicle charging, heat pumps, data centers, and AI software), and reinforcing the trend to increasing decentralisation, with many prosumers owning their own generation and storage units. Demand flexibility, with its inherent decentralized nature, is also gaining importance and reinforcing energy systems decentralised nature. For all these challenges, AI techniques play an increasingly crucial part in the solutions being developed. Examples include the use of machine learning for forecasting (e.g. for renewable generation, demand response and load), smart and distributed control (e.g. for charging electric vehicles and many other energy-consuming appliances), and the use of algorithmic game theory, agent and multi-agent techniques to deal with large systems, energy communities, and electricity markets with strategic agents. Yet, energy systems do not yield to a straightforward application of AI or multi-agent techniques, since they often involve understanding the underlying physical and economic constraints of energy networks, and their real- or near-real time operation. In this tutorial, we aim to cover both the fundamental challenges and techniques that enable the use of multi-agent systems in the energy field, as well as a range of practical research examples.From a network management viewpoint, recent technological advancements, including digital communication technologies, advanced metering devices and sensors, have dramatically changed the architecture and functionality of electrical networks. While these advancements have the potential to improve efficiency, reliability, and sustainability of electrical networks, they have also given rise to various technical challenges traditional methods were not designed to address. This promotes the need to develop novel decentralized and adaptable methods for energy network management at two related levels of abstraction. The first focuses on maintaining the power grid's stability while considering a large number of network entities and components, while the second level deals with the management of the electricity markets, in which the different agents trade energy and energy-based products, and with the regulation of prices within the market.
This half-day tutorial will start with an overview of the key technical and computational challenges that are involved in managing energy systems. We will then provide background on multi-agent AI, with a focus on agent-mediated electronic markets, and show how many of the challenges of current energy systems can be addressed by representing components of the system as autonomous or semi-autonomous agents. The second part of the tutorial will focus on the use of multi-agent modeling and game-theoretic methods (both cooperative and competitive) to provide efficient management solutions and decision-support in energy systems composed of many agents (generators, prosumers, EVs etc.) with their own goals and private information.
Outline and Topics Covered
The tutorial will cover a number of topics, divided among the two presenters depending on their experience. A list of potential topics is provided below, with the final presentation schedule to be agreed before the tutorial:- Introduction to challenges in decentralized energy systems and role of AI and MAS. Our energy systems have to deal with a number of new challenges they were not designed for. These range from increasing uncertainty (from renewable generation or new types of loads, such as EV charging), as well as increasing decentralisation, with individual prosumers acting autonomously, based on their own goals and private information. We start the tutorial with a brief overview of these challenges and discussing how techniques from machine learning, multi-agent systems and algorithmic game theory can address them.
- Simulation of electricity markets: participation in electricity markets involves high risks that should be mitigated using adequate tools. MASCEM - Multi-Agent Simulator of Competitive Electricity Markets is a simulation and modelling tool developed to study and simulate electricity market operations. It models the main market entities and their interactions. Medium/long-term gathering of data and experience is used to support players’ decisions in face of their specific and dynamic characteristics and goals. MASCEM includes day-ahead and intraday markets (symmetric or asymmetric, with or without complex conditions), bilateral contracts, and intraday markets. The Adaptive Learning strategic Bidding System (ALBidS) incorporates a large variety of market decision support strategies with different natures and perspectives, such as data mining and machine learning methods. It is a context-aware system able to deal with different scenarios, ensuring a large scope of approaches. The tutorial will provide participants with an overview of MASCEM as well as with an online tool for hands-on activity. Examples regarding real electricity markets in Europe and realistic scenarios up to 2050 will be discussed
- Design of incentive-compatible scheduling for electric vehicle charging under network capacity constraints. In this part, we will look at how algorithmic game theory and agent-based models with strategic agents can be used to incentivize EV owners (autonomous agents) to charge at times when the local distribution grids are less congested. We will also discuss mechanisms for coordination between vehicles charging at multiple charging stations, considering online/dynamic arrivals
- Design of incentives for the formation of virtual power plants from renewable energy resources. This part will discuss how mechanisms such as scoring rules and coalition formation can be used to enable multiple small generators to form virtual power plant coalitions to participate in the energy market. We will discuss how scoring rules, a technique from decision theory under uncertainty (also popular in the MAS community) can be used to reward accurate forecasts
- Design of incentives for distributed demand-side response. Demand-side flexibility is a key requirement for modern power grids, and increasing emphasis is placed on allowing small consumers (households, small industry) to participate in demand-response schemes. In this part we will discuss different aggregation techniques, as well as multi-dimensional incentive mechanisms that incentivise participants to offer not just lowest cost, but also reliability in meeting their reduction commitments (based on prior joint work with researchers from the EconCS group at Harvard).
- Market mechanisms and energy resource management for energy communities and micro-grids, formed by multiple prosumers (consumers with their own generation and storage). Such settings are a natural area of application for multi-agent systems, and solutions from the MAS community are very natural for this domain. Two main models are possible:
- Distributed peer-to-peer exchange mechanisms, in which each prosumer has its own generation and storage capacity, and then they negotiate energy exchanges with each other using autonomous agents. This part will discuss how the lively AAMAS sub-community working on electronic negotiation can contribute to the energy area;
- Coalition mechanisms, where the energy community invests in a joint asset (e.g. community wind turbine, larger jointly-owned battery capacity). Then, the challenge is smart control and division of the gains from these assets in a fair way. Here concepts from coalitional game theory and Shapley values are highly relevant, and we will discuss the computational challenges and implications in using them for energy systems.
- The challenges for real-time performance will also be discussed, considering market operation and energy resource management. Efficienty and effectiveness of the solutions delivered by multi-agent systems and the need to implement adequate efficiency/effectiveness mechanisms to cope with real-time constrainst will be addressed:
Length and Target Audience
The tutorial will take half a day, with roughly two equal slots of around 1.5 hours + breaks. The target audience of the proposed tutorial is two-fold:- Researchers with a background in AI or agent-based systems that want to learn more about what are the specific challenges/issues in energy systems, and how techniques from their domain of expertise can potentially be used to address them. Given that the proposed tutorial is to be held at AAMAS, we expect the majority of the audience will have some AI or agent systems background. This will likely include many starting PhD students and fresh postdocs, but also more experienced researchers.
- Researchers and practitioners with a power systems or energy systems background - or from the energy industry, that would like to learn more about how new developments in AI can help address some of the challenges faced by the energy sector.
Presenters
Zita Vale
Homepage: https://www.cienciavitae.pt/portal/pt/721B-B0EB-7141GECAD-LASI
Polytechnic University of Porto
Porto, Portugal
Email: zav at isep.ipp.pt
Valentin Robu (main contact)
Homepage: https://homepages.cwi.nl/~robu/Intelligent and Autonomous Systems (IAS) Group
CWI, Dutch National Research Institute
for Mathematics and Computer Science
Work Address Science Park 123, 1098XG
Amsterdam, NL
Email: v.robu at cwi.nl