Jun 2017
Electricity Distribution
Optimisation of Energy Forecasting - analysis of datasets of metered embedded wind and PV generation
Jun 2017
Feb 2018
National Grid Electricity System Operator
Jeremy Caplin (.box.SO.Innovation@nationalgrid.com)
Click here to send a question to the contact.
Network Innovation Allowance
ED - Network improvements and system operability
Asset Management
PV Generation

  1. Analyse the data sets available in order to determine additional factors that can be included in PV power curves (temperature, wind speed, geographical location etc.), and derive optimal power curves for the conversion of weather forecasts into PV generation forecasts.

  2. Investigate the need for seasonal power curves to take account of the difference in height of the sun, and its interaction with angle of inclination of the panels

  3. Investigate the need for regional power curves to take account of the differences in sun angle, or local pollution issues or any other effects

  4. Investigate the use satellite imagery of cloud cover to improve forecasting accuracy.

  5. Determine how accurately national installed PV capacity be estimated from analysis of demand forecast error combined with the available PV data. (Information on PV installed capacity lags by many months).

  6. Investigate the validity of our current methodology of extrapolating from around 80 weather forecasts for specific geographical locations to around 900,000 separate locations – effect of local variations in cloud etc, and suggest any ways of improving the methodology.

  7. Investigate how much response should we carry on the power system to cope with short term fluctuations in total PV generation

  8. Investigate how much reserve should be carried on the system to cope with potential sustained changes in total PV generation – in the form of a probability distribution for possible changes in total PV over say a 15 minute period.

Embedded Wind Power Generation

  1. Investigate whether a system that automatically updates wind power curves for metered data (such as an ARIMA based model) give a better fit to the data

  2. Investigate whether we can use anomaly detection on the time series components of the metered output to improve the power curves? (e.g. recognising a period where output is capped at say 50% of capacity as likely to be a partial outage on the wind farm and so adapting / ignoring this data in the derivation of the power curves – building on the work recognising unusual days from the recent ATI Study Group).

  3. Seek ways of improving our representation of the effects of high wind speed shutdown. (The amount of data for this task is relatively limited, which makes it more challenging).

  4. Investigate the possibility of developing better wind power curves for embedded wind generation modelled at Grid Supply Point level – is a single generic curve appropriate or can the data be classed into several types of wind farm with different curves?

  5. Determine how accurately embedded PV capacity can be estimated from analysis of the demand forecast error combined with wind speed measurements – either nationally or regionally.

The project plans to deliver:

  • Power curves for PV and embedded wind Generation, refined by location, season and/or any other factors found to be relevant.
  • Methodologies for estimating installed capacity of PV and embedded wind generation, including review of the feasibility of using satellite imagery to identify solar panels
  • Methodology for forecasting dispersed PV and embedded wind generation, including definition of appropriate levels of granularity for modelling, and making use of satellite imagery if this proves practicable.
  • Analysis of generation volatility that can be used as an input to response and reserve holding policy.
  • Anomaly detection to identify (for example) partial outages of wind farms in the dataset.
  • Improved models for wind speed cutout.
  • An analysis of the reduction in demand forecasting error that can be achieved by the implementation of the new models
  • A plan for implementation of the new models, and a methodology for demonstrating the ongoing reduction in demand forecast error as a consequence of this project 
Delivery of objectives listed above, leading to a reduction in mean demand forecast error in the three months following completion of the project compared to the previous three months of at least 20 MW. The benchmarking analysis will be based on the 365 day average error. It is expected that a 20 MW step change would show as a 5 MW reduction after 3 months on a rolling 365 day average.