The Control REACT project is featured in National Grid ESO's innovation update. Ciaran Gilbert and I are working together with TNEI and The University of Edinburgh to help the ESO benefit from probabilistic forecasts by demonstrating how they can be produced with available data, how they can be used to support decision-making, and what the value of changing practice could be.

Exciting and challenging times ahead for forecasters at the ESO and across the energy sector better managing variability and forecast uncertainty becomes increasingly important (not to mention an economic necessity!).

For the past few months Naomi and I have been on a domestic electricity tariff with dynamic pricing, Octopus Energy's Agile tariff. At 4pm we can access the prices for each half-hour of the next day. The pricing features a premium for consumption between 4pm and 7pm, but is generally much lower than a fixed rate tariff at other times.

Did we look at the prices every day? No! Did we avoid using energy intensive appliances between 4pm and 7pm? Yes! Below you can see the impact on our effective unit rate each day. From a fixed rate around 15p/kWh we were down to just over 12p/kWh once we adapted our routine. See graph below.

​Could this kind of tariff help reduce the evening peak in national demand if enough consumers signed up? Or would we end up with a sharper peak at 7:01pm?

This post was originally published on the University of Strathclyde Centre for Energy Policy Blog here.

The decarbonisation of electricity generation has been the major driver of wider decarbonisation in the UK over the past 15 years (See: Ofgem). Increasing the share of low-carbon electricity generation will continue to be an effective means of decarbonising over the next 15 years, but it brings with it some challenges in terms of managing the grid. Electricity networks need more than a supply of energy to function – they need a wide range of services to ensure that supply and demand match on a second-by-second basis, and to manage the flows of power around the network. A wide range of ancillary services play a vital role in keeping the lights on. These services regulate frequency and voltage under normal conditions, and also provide contingencies for faults and other risks.

If we want a truly low-carbon electricity system, we’ll need low-carbon ancillary services as well as low-carbon electricity. The good news is that renewables such as wind and solar can provide many of these services – and at highly competitive rates in many cases – but the market for these services needs to be updated to match the emerging needs of the electricity system and to allow renewables to compete on a level playing field with the fossil fuelled power plants which provide the majority of services today. The University of Strathclyde has been working with the energy industry to help make low-carbon ancillary services a reality.

What are ancillary services and who provides them now?

Just like other commodities, the electricity that is delivered to consumers is expected to meet certain standards of quality. In this case, that means a voltage of around 230V (or perhaps higher for industrial consumers), a frequency of 50Hz and with almost no interruptions in supply. Electrical devices may be damaged or even pose a safety risk if power quality drifts too far from these targets. Furthermore, generators - from large power stations to domestic solar - may disconnect themselves from the grid if the voltage or frequency become extreme, potentially causing further disruption.

To date, ancillary services have mainly been provided by conventional power plants, those with large synchronous generators that rotate at the grid frequency. In fact, these generators help maintain the grid frequency automatically by providing ‘inertia’: the sheer mass of their rotating generator slows down any change in grid frequency caused by a mis-match of supply and demand. They can also automatically change their output to rebalance supply and demand, thus providing a frequency response service. Similarly, this type of generator can also provide reactive power services by producing and absorbing reactive power, which regulates voltage.

It wasn’t long ago that these conventional generators provided almost all of our electricity too, but today a significant portion comes from wind and solar power. In 2017 wind and solar provided 18.7% of electricity on the GB grid, and as much as 47% during some half-hour periods (Data from National Grid and Elexon). During these periods of high renewable penetration, there is both a reduced pool of conventional plant able to provide ancillary services and an increased demand for those same services.

The increased demand is in part a consequence of wind and solar not having any inertia, meaning that the grid frequency can change more quickly and therefore more and faster frequency response is required. Another driver is the size of the largest unexpected generation unit outage that National Grid must prepare for, currently up to 1320MW, depending on the schedule of specific generators. With the completion of Hinkley Point C (1800MW) and planned new interconnectors to other countries this could increase significantly, and with it, so will demand for frequency response.

The need for fast frequency response in GB today is largely met by batteries, which have disrupted the market, reducing prices for some ancillary services by over 50% since 2016 (Data: National Grid). The role of batteries in future electricity systems is a hot topic in research as the technology continues to develop and reduce in cost. Their integration is also a challenge as they exhibit characteristics of both demand (when charging) and generation (when discharging), but also have limited storage capacity so can’t only generate (or consume) power continuously for extended periods of time.

Low-carbon ancillary services

While wind turbines and solar panels may not provide inertia, they connect to the grid via highly controllable electronics which convert the direct current electricity generated by the turbines and panels into the alternating current with a frequency of 50Hz to match the grid. This allows the power output of wind and solar farms to be changed very quickly to help keep the grid frequency at 50Hz and provide reactive power services. Using this technology, wind turbines can even emulate inertia by extracting kinetic energy from their rotating parts to briefly increase their electrical power output.

The best part is that the lack of fuel costs mean that this can all be done at extremely low cost. However, providing frequency response requires power from the wind or sun, which is variable and can only be accurately forecast a few days ahead. This is a problem in the current market, as the majority of ancillary services are procured a month or more in advance, meaning that wind and solar can’t join in. Furthermore, the economics of providing upward response aren’t appealing: the market value of supplied electrical energy is far greater than the value of frequency services, and leaving head-room to be able to increase output quickly in in response to a frequency drop would mean lost revenue from energy generation. This isn’t the case for wind turbines’ emulated inertia but there is no route to market for this capability in GB at present.

Near future of Ancillary Services in GB

The electricity landscape is changing rapidly and ancillary services is part of that change. A wide range of changes to the ancillary service market are being explored. Among the most exciting for renewables is the move to day-ahead procurement via an auction. From summer 2019 some frequency response products will be bought in a competitive auction only a day before they are required. On these time scales weather forecasts are accurate enough for wind and solar to compete with the likes of batteries, biomass, combined cycle gas and coal plants. New products are also being designed to combine the capabilities of multiple technologies, including emulated inertia, to provide super-fast frequency response in regions where it is most needed under the name Enhanced Frequency Control Capability (EFCC).
The University of Strathclyde has been working with the energy industry on collaborative initiatives such as EFCC and directly with wind farm operators on taking frequency response from wind turbines to market. EFCC is in advanced stages of testing and in the second half of 2018 wind farms have begun competing for provision of High Frequency Response and are being called upon to provide this service to the grid and applying price pressure on competitors. This is important work in developing our low carbon electricity system and contributing to building a reliable, cost-effective and low-carbon grid.

Ciaran Gilbert and I have won an international wind power forecasting competition run by TU Dresden in conjunction with the 14th European Energy Market Conference. The competition required us to forecast the total generation for a portfolio of wind farms from 2 to 38 hours-ahead every day for two weeks. Our method, which was based on regime-switching autoregression, produced forecasts with an error score 4.5% lower than the second place finishers. Ciaran will be attending the EEM conference in Dresden in June to receive a cash prize and present the approach, which is detailed in this paper and available in open-source software (R code).
 
Energy forecasting is a vital component of modern power system and electricity market operation, and improving the production and use of forecasts is becoming increasingly important as the penetration of weather-dependent renewable generation increases. Improved forecasting will help keep consumer energy bills down by reducing the amount of reserve power required to manage fluctuations in generation and demand. Strathclyde has a strong and growing capability in energy forecasting with active projects in wind, solar and electricity demand forecasting, and engagement from leading utilities including National Grid, SSE and Iberdrola.

I'm participating in the Global Energy Forecasting Competition 2017. This edition of GEFcom is a hierarchical load forecasting challenge, entries take the form of quantile forecasts for the zones of ISO New England, total-Massachusetts, and total-ISONE. I though this was a good excuse to play with ployly... see below for my total-ISONE entry to the first task.

I'm working with a great consortium of industry leaders to develop decision-support tools for wind O&M. Recently, we've been focused on heavy-lift operations which as subject to strict wind speed safety limits. Using scenario forecast (some times called trajectory or statistical ensembles) we can estimate both the expected cost and risk associated with different scheduling options. This kind of information can help reduce operational costs for on- and offshore wind, lowering the cost of wind energy for all.

Thanks to our partners Romax, Datalytics, ScottishPower, and SSE!

There were many interesting presentations of forecasting work at this year's EEM conference, headlined by an electricity price forecasting competition. Competitors were required to forecast the MIBEL spot price for each hour of the next 5 days on a rolling basis for two weeks. The top three placing competitors were:

1. Csaba Megyeri, Uniper Global Commodities, Germany
2. Carlos López Llopis, Alcanzia, Spain
3. Sanjeev Aggarwal, University of Mulana, India

The top two entries came from professional energy analysts who both cited a degree of expert judgement, in addition to their mathematical modelling, when presenting their work.

My attention was recently drawn to this project examining the communication of complex climatological data. Under the umbrella of the European Provision Of Regional Impacts Assessments on Seasonal and Decadal Timescales, and FP7 project, project Ukko visualises the probabilistic forecast of mean seasonal wind speed, compared to climatology. The user can get an overview of whether wind speed is expected to be higher or lower than average (mean from 1981-2014) with brighter markers indicated a region with historically higher skill. Clicking on a specific region reveals more detailed forecast information, including historic measurements and the forecasts from individual ensemble members.

I really like this tool. It successfully presents a lot of complex information in a clear way, one you've got your head around the key or watched the video below. It has inspired me to think about how I present probabilistic information and I hope to see more innovation in this area in the not-too-distant future. It's a shame, however, that the skill is insufficient to offer forecasts for most of Europe and the North Sea where there is significant wind power capacity, though that is of course and issue with the forecast, not the visualisation.

Ukko is the result of a collaboration between the Barcelona Supercomputing Center and FutureEverthing using ECMWF seasonal forecasts, and well worth a play with!
​

This post is a little overdue, but I wanted to thank EWEA for running this fantastic event. The biennial wind power forecasting workshop was fantastic! The technical content was excellent and being able to meet and discuss wind power forecasting with so many forecast vendors, practitioners, users and researchers in one place was an invaluable experience. I will be back! Sadly the proceedings are only available to attendees (but I'm sure most of them would send you their slides if you send them an email :p).

​Thanks again, I will be back in 2 years!

It's a couple of weeks since the IEEE Power & Energy Society's General Meeting, and I thought it was worth highlighting some of the society's forecasting activities. IEEE PES co-sponsors a number of IEEE journals that frequently publish energy forecasting research, such as Trans. Power Systems, Smart Grid and Sustainable Energy, they also sponsor the the Global Energy Forecasting Competition along with the International Institute of Forecasters.

At this year's general meeting there was a large range of forecasting activities lead by experts in various aspects of energy forecasting beginning with a whole-day tutorial on the Sunday before the main conference. Introductory lectures were given by Dr Tao Hong, Dr Shu Fan and Prof Hamidreza Zareipour on load, price, wind and solar power forecasting to a diverse audience of both forecasters and forecast users for both industry and academia. During the conference proper almost every session seemed to include a few posters on one aspect of energy forecasting or another! While the quality of papers varied the number of contributions from, or sponsored by, industry is evidence of the important role forecasting plays in modern power systems. All papers should be available on IEEE Xplore shortly.

The highlight of the week for me was the day of presentations by winners of the 2014 Global Energy Forecasting Competition. Many of the top performing entrants in all four competition tracks were present to discuss their work and receive their prizes. The winning entries will be published in a forthcoming special issue of the International Journal of Forecasting. Something that I had under-appreciated before the meeting was the strength of the machine learning community in the energy forecasting space: many of the top performing entrants were not energy experts or pure statisticians but computer scientists.