6th International Conference Energy & Meteorology: Abstract Submission

Mapping and measuring geographical accuracy and uncertainty of UK PV forecasts using central and ensemble members of Numerical Weather Prediction models (777)

Dan Travers 1 , Alastair Buckley 1
  1. Physics, University of Sheffield, Sheffield, Yorkshire, United Kingdom

While the total annual solar irradiation available for PV generation is published for geographical regions across the UK, this information is not useful for informing short term decisions on balancing the grid, optimizing wholesale power purchasing and scheduling storage.  

It is useful for market participants to be able to quantify and rely on uncertainty estimates for renewable generation.  From a financial point of view, providing or owning flexibility can be considered a financial option like a derivative contract, and optionality has more value under high uncertainty.  For instance, providing “spinning reserve” to balance the electricity grid is more costly when more power must be set aside for fast-dispatch - usually within a 6 hour window - and battery storage is more valuable for owners in times of net demand (hence price) volatility - intra-day out to 1-2 days ahead.  Measuring and standardizing uncertainty will aid reliable financial modelling, decrease risk and ultimately increase penetration of renewable assets.

The poster helps to address the questions of how much value do current Numerical Weather Predictions (NWP) models give to participants in the energy and storage markets and in which geographic locations do the models perform best.  

We present the results of research on forecasting accuracy and uncertainty on the UK PV generation fleet.  Utilizing the data from thousands of PV installations across the UK and training a machine learning model to forecast the PV outturn based on weather variables, we are able to measure the changes in accuracy of the NWP in predicting PV across forecast time and location.  

Further, we test the hypothesis that the NWP ensemble represents the actual probability distribution of PV generation.  We utilize NWP ensemble forecasts to examine where the distribution of actual PV outturn values is predicted by the distribution generated the using ensemble.   Again, this is done on a geographic basis to produce a map of ensemble performance across the UK. It is hoped this analysis can prove useful in providing a metric against which to measure the performance of NWP ensembles for the use case of PV generation.

Results will be presented primarily graphically with supporting text to explain the methodology employed.  The research is supported by the open science organization Open Climate Fix.