The renewables challenge

Predicting the unpredictable

The renewables challenge

Yuya Ra - Junior Data Forecast Analyst avatar
Yuya Ra – Junior Data Forecast Analyst

Limejump’s Data Science and Forecasting team are always asked “what exactly is it that you do…and how can you possibly predict renewable generation?” In short, it isn’t a simple task, the team uses market data to ‘predict’ how much power our customers’ renewable assets could potentially generate, which in turn, informs our trading desk so they can accurately trade our customer’s renewable energy. Before diving into the detail, it is important to highlight the technologies and the challenges they impose on our Data Science & Forecasting team.

The energy challenge

According to IRENA, energy production and consumption contributes to two-thirds of greenhouse gas emission around the globe. For countries to be on track for their respective net zero goals, most nations have ambitions, some of which are already taking drastic actions, to decarbonise the generation and consumption of power. The grid in the UK is evolving, from how we generate electricity to how consumers utilise it, such as the way people are heating their homes (heat pumps) to how they travel (electric cars). As homes and businesses electrify, it has never been so important to generate as much of the UK’s demand from naturally replenishing resources, a major deliverable to achieve net zero targets.

Why don’t we just build more wind farms?

Simply building more renewable assets won’t solve the energy challenge straight away, it isn’t that simple. The two most prominent generation technologies in the UK are solar and wind, but technologies are heavily weather-dependent and deemed ‘unpredictable’ or ‘intermittent’. The temperamental nature of renewable generation means a reliable backup is necessary to keep households warm and, more crucially, to avoid a complete blackout on a national scale. Currently, that reliable source is either gas or coal which can be turned up and down quickly according to changes in demand or supply. This is a major reason why the last remaining coal plants are not quite ready to be retired in the UK, especially with the current ‘Energy Crisis’.
There is, however, a cleaner and faster replacement emerging – Battery Energy Storage Systems (BESS). The Battery’s role in today’s and future energy systems is very straightforward in principle – charge up with cheap electricity when the wind is blowing strong or the day is cloudless; discharge to the grid when the demand is higher than the supply, dispatching clean energy to the grid and a return on investment for the battery owner(s). However, coordinating thousands of assets to meet gigawatts of demand requires a lot of automation from both the asset optimiser and National Grid’s infrastructure.

The challenges

The role that the Data Science & Forecasting team play is crucial, both operationally and commercially. Limejump’s current portfolio involves a myriad of asset types, all bringing different forecasting challenges to the table. Limejump’s Data Science team’s main focus is on the long-term forecast of renewable energy generated, through our Power Purchase Agreement offering. As established, wind and solar are not deemed as ‘predictable’ and will never provide a baseload to the grid. Their output is highly dependent on capricious weather phenomena. While machine learning models can use these correlations to forecast generation in the future, the quality of weather forecasts, especially a year or two in advance, often fails to capture the variance across years. In addition, abnormal weather events are occurring more frequently, adding more unpredictability to long-term forecasts. However, what people assume to be ‘predictable’ or ‘consistent, is not quite as predictable when you dig a little deeper. ‘Baseload’ generation refers to assets with consistent energy generation such as anaerobic digesters and landfill gas generators. The term “consistent” here is misleading, because not only does the supply of source materials vary, but the rate of generation can also be influenced by temperature, waste, supply and more. While usually, the generation of digesters ramps up during hotter seasons, during the past summer when the UK observed historical heat waves, many generators had to be turned down or off to protect the machinery.

Forecasting the unpredictable

Limejump’s Data Science team hold several weekly meetings with various internal teams, all with the ultimate aim of ensuring generation forecasts remain accurate. The team also collaborate with the trading desk to review past forecasts with actuals, this enables the team to ensure that the data sources and their process remain best-in-class.
The team rigorously assesses the quality of our weather forecasts on a weekly basis. By observing the weather outturn retrospectively, they can tease apart the controllable elements from the ‘uncontrollable’ that influence the forecast accuracy. An obvious solution to improving the forecast is using more accurate weather data forecasts, as extreme weather can cause huge deviations. At Limejump, the current weather prediction employed in our forecasting is backed by a sophisticated interpolation of actual measurements taken all over the UK and updated on a regular basis.

Most deviations from the longer-term generation forecast are usually assumed to be caused by abnormal weather events as mentioned above – heatwaves, low wind – however, a proportion of consistent error can be attributed to human factors, namely lack of communication. For example, outages, due to scheduled maintenance of assets only when communicated in advance can be added to forecasts to inform the traders in time. Not knowing if an asset is due to be offline will result in a loss in revenue for the generator and the trading desk with a deficit, something that no parties wish to experience when the wholesale price is so high.

The Data Science & Forecasting team’s role is crucial to Limejump’s commitment to both our customers but also to the trading desk. They are relied upon to deliver accurate insight into how much our customer’s assets will generate to therefore enable the traders to then sell at the right time in the wholesale market, and not be short on committed volumes. The battle of predicting the unpredictable will continue, especially as more wind and solar generation is connected to the grid, and extreme weather events may become more common. One thing Limejump is confident in is the important role technology will play in helping our company, our customers, and the National Grid optimise our future energy needs.