Smart grids are electricity networks that intelligently monitor, communicate, and control the actions of generators and consumers in order to deliver sustainable, economic, and secure electricity
supply. This vision poses technology challenges in core areas such as powersystems engineering, telecommunications, network security, and artificial intelligence (AI).
The AI challenge is to deliver the “smarts” into smart grids – algorithms and software that solve complex problems involving the interaction between multiple self-interested actors (e.g. generators and consumers). These tools must be capable of efficient and reliable operation despite the significant amounts of uncertainty and dynamism that exist throughout the grids.
Following a brief introduction to AI, in particular the sub-field of AI known as Multi-Agent Systems (MAS), this article presents four smart grid technology challenge areas: smart homes, electric
vehicles, smart distribution networks, and energy co-operatives. The article outlines how MAS can address these challenges and in doing so provide opportunities for the benefit of smart grids to be realised by consumers, suppliers, and society at large.
AI and Agents
AI is the branch of computer science concerned with the study and development of intelligent agents and software, an agent being a system that perceives its environment and takes informed actions.
Within AI, the sub-field of multi-agent systems (MAS) concerns multiple actors, distinct stakeholders, and interactions that involve co-operation, co-ordination and negotiation. These concerns are particularly aligned with the distributed intelligence, automation, and information exchange needs of the smart grid. In particular, MAS research focuses on the relevant problems of how to provide automated assistance to users in complex decision making tasks, how to allocate resources efficiently under competing demands, and how to co-ordinate decentralised systems. The next sections make an explicit connection between MAS and smart grids by considering how MAS could be used to solve specific challenges arising in four key components of smart grid systems.
Smart Homes
Smart homes, equipped with smart meters, will provide domestic energy users with real-time cost and carbon emission feedback. This presents consumers with the opportunity to better understand and
control their energy consumption. However, home-owners already grapple with the complexity of the feedback and advice provided by such meters and this issue is likely to be exacerbated when more complex tariffs are introduced.
Software agents can improve the utility of smart meters by autonomously modelling the home and providing energy advice. For example, the agents can learn a thermal model of the home and combine it with a prediction of external factors (e.g. air temperature, carbon intensity of grid electricity) to optimise heating and provide feedback to consumers. MAS technology can also help consumers to break down their total energy consumption into individual appliances, empowering them to take steps toward reducing their consumption. For example, one step a consumer may take in this regard is to partially shift the daytime loads of certain appliances (washing machine, dishwasher, or tumble drier) to night time.
In practice there must be interaction between software agents and human agents (householders) to deliver benefit through the automated management of home appliances to take advantage of real-time pricing tariffs. The software agents need to explicitly model the householders they interact with, knowing when to act autonomously and when to seek approval for their decisions to shift loads or to turn them off. This is particularly important if the aim is to elicit behaviour change in householders by appropriately tailoring the feedback that agents provide.
Electric Vehicles
The projected numbers of electric vehicles (EVs) on UK roads in coming years will place a significant extra load on the electricity grid infrastructure. Smart approaches will be needed to schedule the charging of EVs given the fluctuating demands on the grid imposed by the movement and charging of vehicles within it, and the variable supply of renewable energy.
Agent technology can help to meet this challenge by drawing on methods for prediction under uncertainty and online mechanism design. Uncertainty arises in the behaviours of large numbers of EVs and
their impact on the grid. Software agents can mine data such as network load information and traffic information to predict future movements of consumers to specific locations. Given these predictions, MAS techniques can be used to elicit users’ travel requirements, schedule the charging of their EVs, and suggest peak and dynamic pricing to shift demand across a city.
In more detail, agents can predict individual users’ EV charging needs based on data about their daily activities and travel needs. For example, they may predict aggregate EV charging demands at different points in the network given the continuous movement of EVs, the available charge in their batteries, and the social activities of the users. The MAS techniques coordinate where, when and for how long the EVs are charged, in order to satisfy the predicted needs of users and maximise profits generated from participating in discharging back into the grid.
Smart Distribution Networks
The widespread deployment of renewable generators (e.g. solar panels, wind turbines, and biomass combined heat and power), and the constrained capacity of existing
infrastructure, suggest that flows within the grid will have to be actively managed to make optimal use of renewable generation without overloading the network. Moreover, this resource management problem must be solved on a scale that can accommodate millions of loads and generators.
This challenge can be met by representing or managing each generator in terms of its own autonomous agents within the physical network and the energy markets. Largescale resource management problems have long been the focus for MAS research and particularly efficient. scalable solutions have been developed that employ local computation and communication.
These MAS solutions can be applied in the smart grid to manage the power flow within the networks. where a particular objective might be to minimise the carbon emissions of the generators subject to a constraint that ensures the capacity of no distribution line is exceeded.
Energy Co-operatives
The widespread adoption of renewable generation at the level of individual homes and businesses will lead to the creation of markets composed of millions of ‘prosumers: These both produce and consume energy and will need to optimise this balance in order to make real-time trading decisions to maximise the profits they can make by buying and selling. Moreover. individual stakeholders could form groups for cooperative energy purchasing.
In this setting. the stakeholders can be represented by software agents that employ automated negotiation technologies to settle micro-contracts for energy purchasing.
These agents would be equipped with efficient algorithms that predict prosumers’ consumption and generation profiles and exploit their predictions to maximise profits in the electricity trading market. Moreover, MAS technology can be used predict whether multiple individuals could maximise group profits by banding together at scale within the smart grid.
Concluding Remarks
This article has discussed a number of smart grid technology challenge areas. What they have in common is a requirement for autonomous systems to solve complex prediction and resource allocation problems in the presence of uncertainty and multiple stakeholders. The article has noted that agents and MAS technology are very well matched to this requirement.
The smart grid challenges also have in common the need for software agents to interact and cooperate with humans (and vice versa) in a meaningful and trusted way. The authors believe that a fundamental new science is needed to design and build such human-agent collectives (HACs).
In the context of smart grids. this science should address questions relating to: the flexibility of autonomy; the agile teaming of human and software agents; the engineering of incentives (e.g. on consumers to reduce personal energy use); and the accountability of the information infrastructure (i.e. to generate reliable decisions). Rigorous and practical answers to these questions are required for the full potential of smart grids to be realised.
Further Information
The authors are involved in a number ofprojects that are pursuing the vision of a smart grid in which AI technology (and specifically multi-agent systems) plays a central role.
The following project websites may be
consulted for further information:
iDEaS: Intelligent Decentralised EnergyAware
Systems. www.ideasproject.info
Intelligent Agents for Home Energy
Management. www.homeenergyagents.info
ORCHID: Human-Agent Collectives.
www.orchid.ac.uk