MAIN PUBLICATION :

• APPENDIX
• ABREVIATIONS
• GLOSSARY
• REFERENCES
• EVENTS
• Downloads

Examples of Time Series Power Prediction Results

An example of time series plots for two separate forecast horizons is shown in Figure I.2.14.  It shows how well the model predicted the power output of the wind farm, both 1 hour (T+1) and 12 hours (T+12) in advance.  Also shown within the figure are ‘P10’, ‘P25’, ‘P75’, and ‘P90’ error bars associated with the prediction: for example the P75 level is a power level with a 75 per cent probability of being exceeded.

 

Figure I.2.14: Time Series of Power Forecast, T+1 (Left) and T+12 (Right)

 

Source: Garrad Hassan

To an ‘engineering eye’ this prediction looks good.  It is clear that the forecast has captured the shape of the profile rather well.  To the ‘commercial eye,’ however, the situation is not so good.  For example, on 1 February, there was greater volatility in the power output than predicted.  For hourly trading purposes, such a prediction would then be poor.  Whether or not the forecast is ‘good’ or ‘bad’ therefore depends very strongly on its precise purpose.  For scheduling plant maintenance it is acceptable, whereas for hourly trading it is poor.  The purpose of the forecast needs to be very carefully defined.  This is a strategic as well as a technical question.  That said, forecasts of the accuracy of that defined in Figure I.2.14 are currently substantially increasing the value of wind energy in some markets and where suitable commercial trading strategies are employed, and in other markets such forecasts  help TSOs balance the system.