Energy Systems Catapult has explored new techniques for electricity price forecasts using supply-demand data to support the development of local energy market business models and innovation.

The Probabilistic Day-Ahead Wholesale Price Forecasting report outlines a proof of concept for three potential forecasting models to the GB wholesale market and compares them to better understand their capabilities and limits for probabilistic price forecasting.

Delivered by the Energy Systems Catapult’s digital and data team for the Energy Revolution Integration Service (ERIS), which brings together learnings from across the Government’s Prospering from the Energy Revolution (PFER) programme to support the development of Smart Local Energy Systems.

Many of these local energy projects are exploring new market arrangements and practices which can enable a more energy-efficient energy system and enable the UK to meet its net zero targets. In this proof-of-concept project, the Catapult builds on the insights from models for generating day-ahead, probabilistic, wholesale electricity price forecasts to support this ambition.

Key points

There are several motivations for the models explored in this report:

• Data-driven, statistical forecast models are being investigated rather than simulation-type models since these are more likely to generate accurate short term (day to week ahead) forecasts. These models utilise the most recent information and more closely replicate what potential information could be available in real-time market operations. Simulation models of the wholesale electricity price are typically more suitable for making longer-term estimates.
• The focus of this report is on probabilistic approaches since pricing information is relatively volatile and understanding this information will help the local energy projects challenge projects better, optimise their market-based algorithms and help reduce the risk in their business solutions.
• One focus of this report is a relatively new and novel approach called the X-model, which aims to accurately predict price spikes in the wholesale electricity market. Such price spikes can be particularly disruptive, and hence forecasting them accurately can provide a valuable opportunity for better mollifying their negative effects or maximising the opportunities they provide.

The report concludes that:

• Relatively accurate (compared to standard benchmarks) probabilistic forecast models can be created which capture the regular wholesale electricity price behaviour. Several of the methods tested can be implemented relatively quickly and with minimal historical information. However, vast improvements can be made by using larger training datasets.
• There is evidence that the X-model for electricity price spike forecasting is able to better model the extremes and jumps in the price. However, this method appears to have bias towards larger price spikes and hence predicts larger values more frequently than they would otherwise occur.

The methods here can be used in multiple ways to support market-focused projects. This includes the following:

• Forecasts can be used to optimise any market-based solution which uses advanced planning and hence require accurate estimates of the future prices.
• The probabilistic forecasts can be useful for risk-focused applications which require understanding of the uncertainty in the future prices. The range of possible values can help make better decisions as well as understand the risk and opportunities.
• Probabilistic forecasts can help with modelling a range of possible future scenarios with which applications can test their innovative solution. A single point estimate will only simulate one eventuality and will therefore not test the robustness and viability in scenarios with more uncertainty.
• The spike-based forecasts can help support innovations focused on worst-case scenarios. The spikes in price are infrequent but extreme and hence their estimation through the models presented here can simulate intermittent events and test the robustness of innovations under system stress.