Agents in the Smart Grid

Readings on smart grid futures

FigureEnergy: Understanding Energy Consumption using Interactive Feedback

leave a comment »

FigureEnergy: Understanding Energy Consumption using Interactive feedback

The FigureEnergy Logger

My colleague and I recently got our paper accepted at the Ubicomp conference and it’s to be presented in September. The paper presents our work on energy feedback interfaces that aims to differentiate itself from previous attempts, mainly by industry, to design energy feedback interfaces. The challenge we set ourselves was to create something that actually made sense to users rather than take an ‘engineer’s’ point of view of what feedback should be (both of us being electrical engineers did force us to think outside the box!).

In effect, the key, we believe, to understanding your energy consumption is to move away from the ‘graph’. Indeed, the graph can be the most interesting visualisation for daily consumption data from a smart meter measuring your electricity consumption from a economist’s, engineer’s, or mathematician’s point of view but not from the general public’s.

Hence, to change all this, we thought of a ‘change of representation’. Just as in AI, one would change the representation of a problem to try and solve it using a clever technique, by changing the representation we aim to make it easier for people to understand how they are using energy, compare energy consumption activities, and reflect on such activities. By so doing, they may be able to focus their energy reduction techniques on the right activities (e.g., did you know your set-top box drains 70W when on standby and consumes much more than your phone chargers you carefully remove every time you are not using them – why not unplug your set-top box as well?).

The FigureEnergy system allows you to find out these issues through the use of annotations of historical data, where basically, you annotate an energy consumption graph by dragging your mouse and selecting a period of time over which you performed some energy consuming activity. For example, you may annotate ‘Washing machine run with setting 1’ from 5:00 to 6:20 and ‘Dinner with Friends’ from 8:00pm to 11:00 pm on a Friday night.

The use of annotations is an extension of the work of Enrico’s on tangible interfaces (see website). It’s an interesting approach that’s based on ‘constructivism’ – the idea that you learn better when you practise reconstructing in your mind how to do something – here you are reconstructing your energy consumption activities by annotating them on a graph. Using these annotations, we then change this graph-based representation of your data to a ‘block-based’ one where each energy consumption activity is a rectangle whose size is proportional to the energy consumed by the activity. Users were found to like this representation as they could relate to it more easily than to numbers (e.g., my washing machine consumes 1kWh while my kettle consumed 0.6 kWh – is that really bad? Or are these standard consumptions? If the washing machine block is twice as big as the kettle and you can see them compared to your TV, then you know where most of your costs are).

The results reported in the paper do confirm our intuitions that the use of interactive feedback results in greater understanding (e.g., people start making sense of how much devices consume not just in terms of power -i.e, through spikes – but also in terms of duration – i.e., blocks of energy usage).

Given the good ‘feedback’ (no pun intended) on FE, we are planning several key improvements, one of which would be the extension to using CurrentCost  meters. At present we are using AlertMe meters to get aggregate energy readings and they’ve worked well for us.

Get the full paper at:

Costanza, Enrico, Ramchurn, Sarvapali D. and Jennings, Nicholas R. (2012)Understanding domestic energy consumption through interactive visualisation: a field study. In, Ubicomp 2012, 14th ACM International Conference on Ubiquitous Computing, Pittsburgh, US, 05 – 08 Sep 2012.10pp. (In Press)

Written by agentsinthesmartgrid

June 26, 2012 at 10:00 pm

On Electric Vehicle Usage

leave a comment »

EVs are likely to make  significant demands on how the grid operates but also on how people will  use them most efficiently while minimising their impact on their lifestyle. Indeed, across the spectrum of potential users of EVs, users’ mobility demands varies significantly. For example, while companies operating fleets of EVs (e.g., light-weight goods vans or cars) may need to serve specific routes all the time, professionals commuting to work will only do so on specific days (work week) and will drive on other, possibly longer distance, routes at other times. Also, families may need to have two cars to serve different sets of itineraries on different days (e.g., dropping off the kids, shopping trips, visits to parents, etc..). These usage scenarios exemplify the variety in energy demand that can be put onto the batteries of such EVs. Thus, while in the simplest case, fleets may only need enough energy for short routes and can afford to keep a stock of charged batteries, in the more complex and general setting (which we focus on here), individuals may only be able to purchase one or two electric vehicles and only charge them individually overnight or at specific charge points along their routes (e.g., shopping centre car parks or stadium).  Moreover, with the deployment of charge points (e.g., at home, en-route, or at service stations) at different locations and with different prices, users will be faced with a range of options as to where to charge their batteries, each with a price to pay. To this end, it is important to design information systems that will help them do so, while taking into account their travel needs.

To understand what types of systems will need to be designed, we consider some of the key features of EV usage in the smart grid as outlined by [1,2]. Key ideas include  the concept of ‘mobility-on-demand‘  whereby users will hire vehicles for only those trips they need and that of leaving it to those engineering the infrastructure to make sure users’ traditional mobility demands are served efficiently. Both approaches will require the design of decision support systems to enable a user to find the cheapest and most efficient way to get to her destination at all times.

Within the mobility-on-demand  vision, EVs would be stacked at various locations in a region and charged, ready to be rented out to users. This scenario requires that the user (or users) be informed about where these stacks are, how long it will take to get there, and where to deposit the vehicle and how long it will take to get from depositing the vehicle to her final destination. This assumes that each vehicle will have enough charge to take the user to the destination given the route taken and road conditions (e.g, blockages, hills). While existing satellite navigation systems can help estimate travel distances and hence the amount of energy (hence travel costs) required for a trip, users will need to be provided with real-time information about where EVs are available and an estimate of how much it is going to cost them (given the predicted route she will take) and how long it is going to take to get to their destination. Presenting that information to all users for them to  choose their cheapest and fastest option would need to factor in  the different time/destination preferences (and trade-offs between time and cost) of each user to minimise delays without overloading users with too much information. Hence, the challenge here is to minimise the complexity of providing overall routing cost and duration (pickup+travel+time from drop-off) information while ensuring that users make an informed decision that will minimise their costs and allow them to reach their destinations in time. This will be made more complex if different stacks are priced differently by the system, in an attempt to balance demand and supply of EVs across the region [2]. If users’ mobility demands are optimised accordingly,  more predictable behaviours in the system would result (as users act rationally), thus allowing for better system planning.

Now, if users instead own the EVs they use, they mainly need to worry about where and when to charge their vehicles to ensure they can make the trips they intend to do by the time they reach the next charge point/destination. To this end, users will be faced with a number of optimisation problems. First, similar to a vehicle routing problem, users will need to plan their trips around charge points in an attempt to minimise the overall cost of energy by either choosing the shortest routes or the cheapest charge points. The routing problem is further complicated by the fact that each charge points may have different power ratings, hence charging time, which contribute to the delays the user incurs in charging her vehicle. Second, if the user only travels short distances and charges her car at home, she would need to consider the costs to charge her EV at different times of the day (assuming a spot price for electricity), which means she would need to optimise the EV charging cycle  with respect to her travel needs in order to minimise her costs in the long term (over days or months). This may require predicting spot prices for a number of days ahead and, more importantly, the user’s travel needs. While spot prices are to some extent more easily predictable (various models exist for this [5]), travel needs may be harder to predict and could either be based on historical data or user input. Then, by combining such information with road condition predictions and, possibly, prices at other charge points on predicted routes, the charge cycle could be further optimised. Third, given the demand-side management and Vehicle-To-Grid (V2G) programmes  that user could have access to, she will need to optimise the types of price plans and V2G sessions she participates in order to maximise her profits. This may eventually mean trading-off certain trips to be able to sell back electricity to the grid.

Some of the above decision making processes could be automated but would need to ensure that users understand what they commit to as the impact of wrong charging cycles could negatively impact their general lifestyle and safety (e.g., if they need a car to go the hospital and the battery is flat). Initial work in the area of EV usage optimisation such as [3,4] do provide novel optimisation algorithms but these are limited to  the interaction mechanisms required to inform and receive input from users.

[1] UK Royal Academy of Engineering (2010) Electric Vehicles: Charged With Potential.

[2] Mitchell, W., Borroni-Bird, C. and Burns L. (2010) Reinventing the Automobile.

[3] Gerding, E., Robu, V., Stein, S., Parkes, D., Rogers, A. and Jennings, N. (2011) Online Mechanism Design for Electric Vehicle Charging. In: The Tenth International Joint Conference on Autonomous Agents and Multi-Agent Systems (AAMAS 2011), 2-6 May 2011, Taipei, Taiwan. (In Press)

[4] Hutson, C., Venayagamoorthy, G. and Corzine, K. (2008) Intelligent scheduling of hybrid and electric vehicle storage capacity in a parking lot for profit maximisation in grid power transactions. In Energy 2030 Conference, IEEE, pages 1-8.

[5] R. Weron (2006) Modeling and forecasting electricity loads and prices: A statistical approach.

Written by agentsinthesmartgrid

April 28, 2011 at 8:04 am

Posted in EVs

Data in the Smart Grid

leave a comment »

This is just a short post in response to this article in the IEEE Smart grid newsletter on ‘Slicing and Dicing Smart Grid Data’,  by Christos Polyzois which states that:

If a utility tried to collect detailed consumption data at a central site in real-time (interactively) once every 15 minutes (let alone more frequently), each customer would generate almost 100 transactions per day. This number is a multiple of the number of banking, credit card and airline transactions that a household generates per day, which implies that the smart grid would necessitate an IT infrastructure (computing and communications) larger than that of the banking, credit card, and airline industries together.

The author calls for the different types of data to be identified and managed differently:

Broadcast data. To balance the grid, data concerning grid status need to be communicated to market participants. That includes information on price changes, critical peak events, reliability signals, and–in the future-carbon content. Availability of this data makes the electricity market efficient by allowing consumers to make informed decisions, leading to better allocation and utilization of resources.

Comment: Some of this data is already broadcast over the internet in the UK at least – however tools are not readily available to filter down this data to users. AlertMe and GridCarbon are two apps that at least show the users how they fare against others near them and how much carbon the grid generates in real-time respectively. However, it is unclear what the benefits of broadcasting the data are – more user evaluations are needed to determine the impact of this information. Of course, if an agent (managing a storage device or a micro-generation facility) obtains such signals, it is likely to ‘react’ to them without consideration for what other agents ‘reacting’ to the same signals would do. In fact, if all agents react to a low price at the same time (as we’ve shown in our work – see my previous post), significant peaks could be generated in the system and hence, such signals could be detrimental to the system. Hence, the reaction to such signals need to be well studied before assuming that ‘better allocation and utilisation’ of resources will be achieved.

Billing interval data. For price signals to make markets efficient, some form of variable pricing must be in effect (time of use, critical peak pricing, peak time rebates, and so on). A utility needs to take readings at the beginning and end of a period during which a particular price is in effect, so that the utility can bill its customers properly.

The idea of billing a user for its periodic consumption aims to filter down the costs in the balancing market to the consumers. At the moment they are agnostic to such costs because of flat price/TOU pricing profiles. BUT will the users actually react to more accurate billing – will they care about a difference of 5 pence for half an hour? Again, on the utility’s side, accurate billing is good but it needs to be accompanied by consumer education or feedback (through the smart meter) to be of any long term value. It is to the advantage of the utility to do so as, making consumption flatter, will help it buy more energy in bulk in the forward market and possibly trade in futures as well.

Detailed consumption data. Energy customers need to know their current rate of electric consumption so that they can adjust it; they may also use automation controls to set their preferences so as to avoid having to constantly take actions.

Again, here the issue of human behaviour come up. It is not clear from existing results that users will indeed react to real-time prices or  want to automate the control of devices – they’d prefer to defer them themselves and might not let this deferment to be more than a certain number of hours (Note: I’ll add some references on this later).

Aggregate statistical data. Energy service providers may show customers their consumption by month, give them comparisons with neighbors or with historical time series, and other analytic results based on historic usage.

The impact of such data will probably help make users aware of their energy consumption and it is important to study the impact that the social network of a user can have on her energy consumption. Google PowerMeter and AlertMe try to do the same but, so far, no serious evaluation has been done in this space.


Written by agentsinthesmartgrid

February 22, 2011 at 10:09 pm

Posted in Data

On Demand-Side Management

leave a comment »

Since power systems specialists tend to be very good at controlling the grid and the suppliers on the grid (either via energy markets or through regulation), the consumer-side of the problem has always been quite elusive. The term ‘demand-side management’ (DSM)   aims to capture all the possible mechanisms by which demand can be contolled (effectively, ‘management’ seems to be the non technical term for the more controversial ‘control’ term that could have been used).  This may be an over-generalisation given that the ways in which demand can be controlled can be further broken down into a number of categories:

  1. DSM via pricing mechanisms – to incentivise consumers simply based on costs to reduce peak demand or shift consumption to later times.
  2. DSM via decentralised coordination mechanisms – the actors involved could be exchanging messages in order to agree to curtail demand (as in a Virtual Power Plant) or to provide energy back to the grid at times where it is needed (e.g., electric vehicles coalescing to provide energy back to the grid).
  3. DSM via centralised control of generation/consumption – the most drastic form of DSM.

The above categorisation shows that DSM, as it will be carried out in the smart grid, will have to take into accound that consumers will not just be consumers and would also be producers, hence the term prosumers. The term prosumer was coined by futurologist Alvin Toffler [1]. Toffler defined the prosumer as someone who blurs the distinction between a “consumer” and a “producer.”  In many other domains, a prosumer means various things (e.g., someone who works at home and at the work place [2] or even a professional consumer –see wikipedia entry ). However, in the energy domain, and in the smart grid in particular, prosumers will become complex first-class objects of their own. These prosumers are characterised by their:

  • Demand elasticity – how they respond to prices based on how they view their need for energy but also how they can shift consumption across any given time period by virtue of their energy storage capacity or the deferrability of their devices with respect to their impact on their comfort.
  • Supply capacity and availability – their micro-generation capability may widely vary – starting with individuals generating energy from their solar panels which are affected by the weather to VPPs of facebook friends banding together to sell energy from their combined CHP/solar panel installations. Note that supply = reduction in demand and hence supply capacity/availability is equivalent to how these actors reduce their demand according to their storage/deferrability of their loads.
  • Social context – as highlighted in the previous point the actors may wish to band together to form VPPs or even decide to run energy cooperatives/consumer groups keen on finding ways to reduce their consumption proactively or sell back to the grid their extra generation (curtailment) capacity. However, the conscious embedding of energy-related activities into the traditional social activities of consumer will require a significant paradigm shift. Similar to the shift caused by mobile phones which allows people nowadays to maintain their social relationships all the time (e.g., checking-in using Facebook places and FourSquare), electric vehicles or smart meters could ideally, one day, allow people to all coordinate to shift their energy and literally ‘check-in’ into some demand-reduction exercise in real-time. However, while the motivation to ‘check-in’ and form groups is natural when it comes to sharing interests in movies and books, it is not so much the case when it comes to energy. The benefit of switching off lights, turning down the thermostat by 2 degrees sounds greatly appealing on the first day you hear it but is quickly forgotten later on. Something (possibly an intelligent agent) has to keep this up.

The prosumer is a complex animal not yet fully understood and likely to morph into something most people (engineers in particular) are scared about. Building more predictability into the system is of immense importance. Some of the main reasons are as follows:

  • If prosumers decide to all charge their batteries or defer their loads at times when prices are low, they can and will cause higher peaks than currently exist. This is one of the main worries expressed in the power systems community [3]. Indeed, the grid makes significant savings in terms of generation capacity needed given that the energy consumption of typical consumers is so diverse – that is, they tend to consume their absolute maximum at different times of the day. This means that, when their demand is aggregated, it is never the case that the total demand at any point in time is equal to the sum of everyone’s peak demand! DSM may break this natural diversity.  To illustrate this see the figure below which shows the peaks that would have been generated in the grid if consumers were simply reacting to prices (i.e., choosing to consume at specific times of the day given that they could defer their usage of electricity at any time irrespective of the cost of comfort incurred) on the grid for example [4]. One of the key ways such peaks may be reduced is to engineer inertia into the shifting of loads using machine learning techniques as we showed in [4].

Figure showing the consequences of applying Time-of-use pricing, Spot-pricing (day-ahead prices), and a flat price (as it is currently).

  • If prosumers decide to buy new batteries and new CHP units/solar panels  in the home, their consumption/supply capacity and availability may change significantly throughout the day (depending on how users decide to use energy) and throughout the year (as new and old devices are used, fail, or serviced). Hence, the traditional mechanisms used to make sure supply capacity is available days, weeks, or months ahead may not apply anymore. The environment will be much more dynamic. One of the key ways this may be managed is to ensure that there is enough redundancy in the system, that is, making sure that the supply from multiple providers is contracted and used if needed.
  • Social activities will dictate how prosumers create peaks or contribute energy (or reduce energy consumption) to the grid if prosumers are given enough incentives to do so. Incentives, however, cannot simply come in the form of prices as, not all social activities can be matched to cost savings. For example, having dinner in a nearby restaurant instead of the sushi bar you prefer 3 miles away, may not be the action you decide to take in case going to the nearby restaurant will save you 5 dollars and help the grid reduce carbon emissions. People like to be together and will do anything to do so (e.g., fans of Justin Bieber travelling to see him in London from all corners of the UK). At this point in time, there is no obvious solution to this. The socio-technical challenge involved is immense.

The use of intelligent agents in managing devices, storage, and generation capacity is a foregone conclusion. Several major challenges, however, still lie ahead in terms of modelling energy usage optimisation problems and addressing the key human-computer interaction issues involved.

UPDATE: I saw this article on the Prosumer in the smart grid today! : http://smartgrid.ieee.org/news-smart-grid-newsletter/3027-top-level-european-planning-for-smart-grid?utm_source=IEEE+Smart+Grid&utm_campaign=fc3313adeb-February_2011_Smart_Grid&utm_medium=email

[1] A. Toffler, W. Longul and H. Forbes (1981) “The third wave”, Bantam Books New York.

[2] William Gerhardt (2008) “Prosumers: A New Growth Opportunity”, CISCO.

[3] G. Strbac (2008) “Demand side management: Benefits and challenges“, Energy Policy, Volume 36, Issue 12, December 2008, Pages 4419-4426, Foresight Sustainable Energy Management and the Built Environment Project.

[4] Ramchurn, S., Vytelingum, P., Rogers, A. and Jennings, N. (2011) Agent-Based Control for Decentralised Demand Side Management in the Smart Grid. In: The Tenth International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2011), 2-6 May 2011, Taipei, Taiwan. (In Press)

Written by agentsinthesmartgrid

February 19, 2011 at 11:27 am

On the deployment of Virtual Power Plants

leave a comment »

A Virtual Power Plant (VPP) is a concept derived from the ‘Virtual Utility’ metaphor used by Awerbuch and Preston [1] in the late 1990’s to describe the aggregation of distributed energy generators and consumers typically located in the distribution network of the grid. The key challenges identified by [2,3] are as follows (with my comments):

  • A clear definition of the key characteristics of the control interface exported by distributed energy resources (DER), which could be generators or controllable loads. How could such interfaces lead to portfolios of DERs be defined for the overarching CVPPs (Commercial VPPs) and TVPPs (Technical VPPs)?

    A possible answer to this would probably be to produce a clear ontology and a language to express the capabilities of the DERs and their feasible connections and conflicts. Does an ontology already exist to express DER capabilities and controllability? What would be the best formalism to express the feasible associations/conflicts among the DERs? Representations of plans (e.g., TAEMS) for multi-agent teamwork come to mind as these have been used to express such issues in very dynamic and uncertain settings (military – the COORDINATORS programme)
    as in the smart grid.
  • Distribution energy management – i.e., how to make sure the energy produced or used in the distribution network is properly managed to avoid overloading lines and keep voltage levels stable.

    The authors in [2] show, using optimal power flow equations, how the output from a VPP can be significantly affected by the conditions on the network. Power ouput may have to be severely curtailed by the distribution system operator (DSO) who would be mimicking the role of a TSO (transmission network operator) as it is on the tranmission network at the moment. Wouldn’t this unravel as the number of DERs scales to hundreds of thousands as consumers start using smart meters and solar panels in their home? The challenge would be to design distributed control mechanisms to ensure lines are not going to be overloaded as a result of (un)-coordinated actions (production/consumptions) by individuals.

  • Providing a commercial and regulatory framework – i.e., how to ensure the energy produced/consumed is paid for/billed for exactly.
    Aggregating resources technically is a different challenge from the commercial aggregation given that each individual resource is potentially owned by a different company which will try to maximise its profits. It seems, so far, that each DER will only get paid for what it produces at the end of the day, no matter what the others in the VPP produce. However, this is not entirely sane given that each DER is serving to compensate for others’ inability to produce at certain times as pointed at in [3]. Given the increasing heterogeneity of DERs, it can be expected that the negotiation power of some of these bigger/more reliable DERs will push others to share parts of their earnings with them. This points to the need to devise payoff sharing mechanisms that users are comfortable with. Cooperative game theory points to different ways of doing this through solution concepts such as the Core/Kernel/Shapley value but these tend to be intractable or rely on some simplifying assumptions about the actors’ utility functions. In the smart grid, solution concepts that are computable in real-time will be needed.

  • Providing the communication platform for all the components of the VPP to talk to each other.
    Since the VPP is likely to be composed of lots of individual actors it is important to design the communication infrastructure such that it is robust to failures but also to delays in comms. Conversely, the decision making algorithms of individual components would need to be able to manage any latency or failure in the communication infrastructure in order to guarantee stability.

 

[1] Awerbuch, S. and Preston, A.M., (1997) “The virtual utility: Accounting, technology \& competitive aspects of the emerging industry”, Kluwer Academic Pub.

[2] Pudjianto, D.; Ramsay, C.; Strbac, G.; , “Virtual power plant and system integration of distributed energy resources,” Renewable Power Generation, IET , vol.1, no.1, pp.10-16, March 2007 , URL: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=4159950&isnumber=4159947

[3] Dimeas, A.L.; Hatziargyriou, N.D.; , “Agent based control of Virtual Power Plants,” Intelligent Systems Applications to Power Systems, 2007. ISAP 2007. International Conference on , vol., no., pp.1-6, 5-8 Nov. 2007 URL: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=4441671&isnumber=4441582

Written by agentsinthesmartgrid

February 15, 2011 at 10:25 am

Posted in microgrids