At times, it feels like summer and winter are the only seasons that get any love from energy forecasters. Although there are higher stakes in peak seasons, shoulder seasons have their own complexity when it comes to forecasting. In particular, the increase in behind-the-meter solar generation with lower demand makes spring forecasting challenging.
Mid-day tends to be difficult to forecast in Great Britain (GB) due to behind-the-meter (BTM) solar generation increasing, large temperature swings from cool nights to warmer mid-days, and a wider range of potential weather scenarios that will affect demand. According to Solar Energy UK, there is an estimated 15 GW of solar capacity in the UK, with 5 GW of that being rooftop solar, significantly impacting GB’s accuracy metrics when forecasting mid-day.
Taking a closer look at the full month of April, GB recorded a MAPE (Mean Absolute Percentage Error) of 5.00% across HE10-18, as well as a high positive bias. This implies that Elexon may be over-predicting demand due to behind-the-meter solar reducing demand metrics.
Amperon, however, uses machine learning to capture these types of mid-day trends, so that things like behind-the-meter solar are baked into our forecast. This has led Amperon to a 3.06% MAPE across these hours, with 39% improvements over the TSO for the mid-day hours (HE10-18) in the month of April. On April 16, we saw a great example of our forecast achieving a 1.66% MAPE compared to GB’s 6.79% across HE10-HE18, delivering a forecast that is over 4 times more accurate than the TSO.
At the utility level, Amperon builds BTM solar directly into our net load forecasts, rather than treating it as a separate variable. Amperon's forecasting models are partially driven by weather forecast variables including, but not limited to, irradiance, cloud cover, and precipitation, which are heavily correlated to solar generation.
Our team recently made additional model improvements focused behind the meter solar generation. These enhancements were made with a focus on capturing mid-day solar impacts in areas with higher penetration. By modeling the relationship of 15+ weather variables and net load, we’re able to improve accuracy during volatile periods -especially those sunny spring days when BTM solar is working hardest to flatten the mid-day curve.
The residential solar market faced a challenging 2024, marked by a significant slowdown after years of impressive growth. According to McKinsey, from 2020 to 2023, the industry flourished due to factors such as geopolitical instability, favorable policies, and low interest rates. However, the market’s rapid expansion proved unsustainable, as rising interest rates, policy changes, and the expiration of incentives led to a sharp decline in installations.
Despite a 15% decrease in new installations, key markets like the U.S., France, and the U.K. continue to grow, though at a slower pace, and are expected to experience a steady recovery. Residential solar remains a competitive energy option in many regions, with unsubsidized solar costs often lower than grid electricity, and increasing demand for battery backup systems.
While growth has slowed, the long-term outlook for the industry remains positive, with a forecasted return to steady growth around 2026–2030. For load forecasting, the evolving dynamics in the residential solar market, especially the slower growth and increased adoption of residential battery systems, will continue to influence forecast accuracy and energy demand patterns, requiring models to continuously evolve to account for the changes in generation and storage capacity.
If you’ve had trouble forecasting demand during mid-day hours, contact us here to learn how we can help you make faster decisions with deeper insights.
.png)