In recent years there has been a significant growth in the proportion of electricity provided by renewable technologies. For example, in 2016 wind and solar power contributed approximately 15% of the UK's required electricity. While this growth has led to a large reduction in the carbon intensity of electricity generation, variability in renewable generation presents a challenge to National Grid with regards to balancing supply and demand of electricity. To mitigate the uncertainty associated with wind and solar power forecast errors, National Grid procure a level of reserve to maintain security of supply, resulting in an additional cost to the consumers. The aim of this fellowship is to reduce the error in National Grid's renewable power forecasts with better use of state-of-the-art meteorological data.
National Grid produce forecasts of wind and solar power for a range of lead times. At short forecast lead times (minutes to hours), wind and solar power forecasts are produced using statistical techniques based on observations of meteorological variables and metered power output. Beyond this time horizon (hours to days) there is a reliance on the meteorological output of numerical weather prediction (NWP) models. Well established transfer functions are then used to convert the meteorological variables to power.
Analysis of the performance of the forecasts has predominantly focused on determining the level of skill in estimating the instantaneous wind or solar power output for various lead times. The results are typically presented in terms of statistical metrics such as mean absolute error (MAE) or cumulative rank probability score (CRPS) which summarise the performance over a long period of time. However, they do not quantify how the forecast errors vary for different meteorological conditions or determine the meteorological cause of the largest forecast errors. Furthermore, wind and solar power forecasts are assessed independently and therefore there is little understanding of the correlation between errors.
This fellowship provides a unique opportunity to use techniques and skills I have developed at the University of Reading, to determine the relationship between errors in wind and solar power forecasts in the UK, investigate the meteorological conditions which lead to the largest renewable generation errors and determine how the errors could be reduced with better use of meteorological data.
The project will be divided into two work packages:
Work Package 1 aims to quantify how errors in wind and solar power forecast vary with large scale meteorological conditions. The historical performance of the National Grid forecast of wind, solar and total renewable power will be assessed for two lead times which are of particular relevance to the decision making processes within National Grid (day-ahead and in-day). From the analysis, the periods with the largest forecast errors for wind, solar and total renewables will be identified. The large scale meteorological conditions for each of these periods will be described using state-of-the-art historical datasets. Cluster analysis techniques will be used to determine particular weather types which lead to large forecast errors.
The derived weather types will form the basis of a tool which will be used by the energy forecasting team in National Grid to provide an early warning of a potentially large forecast error day. This will work by comparing the atmospheric conditions in the long term forecast (3-10 days) with the large forecast error weather types.
Work Package 2 aims to investigate why large forecast errors occur and whether they could be reduced with better use of meteorological data. For approximately 10 case studies, the cause of the errors will be determined by comparing meteorological observations with the forecast. The final part of the project will determine whether in-day errors could be reduced using real time meteorological observations at strategic locations.