Using ensembles to predict and hydro to balance wind power forecast errors

Principal Investigator

Corinna Moehrlen
University College Cork
Dept of Civil & Environm Engineering
University College Cork
Ireland

c.moehrlen@ucc.ie

Other investigators: Dr Jess U. Joergensen and Dr Markus Pahlow

Project description

Objectives

Large amounts of wind power have been installed in Denmark and Germany over the past 20 years. This wind power development has been ongoing over long time frames, such that single wind turbines, smaller groups of wind turbines and wind farms have been dispersed over the countries. In Spain, which is the third largest market, and other European countries, the development started later, and hence more larger wind farms, rather than single turbines or smaller groups were built.

The new energy source, that had to been integrated into the electrical grid with its intermittent character requires forecasting. The quality of the wind power forecasting is not considered sufficient by most transmission system operators, because the value of wind power is dependent on the forecasting accuracy. Any forecast error leads to balance costs, which imply reduced value of the energy and increased production price per energy unit.

The analysis of the wind power forecast errors carried out in the special projects "Verification of Ensemble Prediction Systems for a new market: Wind Energy" (2003-present) and "HONEYMOON" (2003-2005) indicate that the forecast error is unlikely to disappear.

It is unrealistic to believe that any forecasting system can manage to keep phase errors consistently on a timescale of only 15 minutes to avoid balance costs of energy. Considering the accuracy of the SYNOP and TEMP observations, the relatively long forecast horizon of 24-48 hours in combination with the fact that meso-scale weather activity is important for wind power forecasting, it is obvious that forecasts have much longer time scales than 15 minutes. Nevertheless, this is what European TSO's standard requirement is at present.

For a single wind farm, such requirements are in fact impossible to achieve, because a considerable fraction of the wind power production is influenced by either meso-scale weather or planetary boundary layer processes with a short time scale. Therefore, we can postulate that from a theoretical perspective the wind power prediction error will remain to a certain extend.

The special project report of the project "Verification of Ensemble Prediction Systems for a new market: Wind Energy" from June 2006 will publish the benefit of predicting spinning reserve requirements by using an ensemble of wind power forecasts based on weather forecasts on synoptic scale. This is done via a stability dependent parametrisation using the bias corrected ensemble spread.

In this project we will investigate, if further improvement of the prediction of the reserve may be achieved by using the native bias corrected ensemble spread and by using enough spatial resolution to explicitly resolve meso-scale phenomena.

The next step will then be to combine the uncertainty of the forecast with pumped hydro power. A pumped hydro station can store un-forecasted power from wind and deliver energy with short notice, if the wind power was over predicted. However, the combined wind hydro system is also going to link to other forecasting quantities with large local uncertainty: precipitation and runoff.

In this way we combine renewable energy sources, reduce the side effects of wind power (intermittency) and ensure that the need for fossil fuel based energy generation will decrease.

Wind power is at the moment considered the cheapest energy source, provided that it is possible to limit the cost of the integration. That means that the remaining fossil fuel generation can be scheduled efficiently according to how the wind is blowing. In light of the increasing energy prices, there is reason to accelerate research leading to optimised wind energy integration.

This project will put focus on the wind power prediction in some of the most windy areas in Europe, namely the Irish counties Donnegal, Sligo and Mayo. The results from these areas will be of relevance to many other areas in Europe. Especially rural areas with similar high wind resources such as in the northern part of Norway and Scotland, all of which have hydro energy that can be used to level out the intermittent nature of the wind power.

Description of Work

An initial study for a pumped 1GW hydro plant in Germany (Langhans 2006) using wind power forecasts, has given promising results with respect to combining wind and hydro energy, even though the timescale of the hydro plant is less than 12 hours.

However, the dispersed wind power in Germany has a high level of inertia as it is driven by large scale weather. The high level of inertia simplifies the prediction complexity considerable in comparison to wind power generation in for example Ireland. The higher concentration of wind power in wind farms of up to 72MW and the higher wind speeds in Ireland cause make it more difficult to maintain balance between generation and demand. This is due to the high variance of the wind power production from large and frequent changes in the wind speeds within short times.

Wind power is installed at locations, where the wind resource is the highest. This is in many countries in rural areas with considerable connection cost. Thus, only the construction of large farms is feasible, and the drawback is that the wind power generation changes fast and frequent due to the intense mesoscale weather present at any time of the year. This is most pronounced in the north west of Ireland.

It has been demonstrated by Lang et al. (2006) that the forecast error in Ireland for a single site is much higher than in Germany. Recent results have confirmed this for a large fraction of wind power in Ireland.

To balance the wind power forecast error in Ireland with storage techniques such a pumped hydro is therefore a considerable more demanding task than what was already demonstrated in Germany. The inclusion of the forecasted uncertainty will thereby be crucial.

An optimised scheduling of the two renewable energy resources, wind and hydro, also need consideration of up to 10 days ahead, as the management of the water reservoirs need to take the catchment inflow into consideration. For the purpose of illustration, the precipitation of short term forecasts and medium range Ensemble forecasts will be used to obtain realistic forecasts and forecast errors of the hydro inflow on virtual hydro stations.

The project will use the following data sources:

• One year wind power measurements made available from ESB National Grid.
• ECMWF medium range Ensemble forecasts of wind and precipitation
• Short Range Ensemble Forecasts of wind power from the special project
• "Verification of wind power for a new market: Wind Energy"
• Synop observations to calibrate the short term area integrated precipitation forecasts.

The use of an ensemble of weather forecasts is a way to estimate the uncertainty of weather. This has been demonstrated on synoptic scale at medium range. We want to extend this knowledge to the meso-scale, short-range, if realistic physical parametrisation perturbations are added in the planetary boundary layer. Thus, high resolution ensembles are likely to natively model the

uncertainty of the wind power better than a stability dependent parameterisation from an ensemble in lower resolution. The two methods will be tuned and verified.

The result of this analysis is important for the future design of high resolution short range ensemble forecasting systems as well as the further development of energy supply.

References

Langhans, Leif, Untersuchung zum Potential einer optimierten Kopplung von Wind- und Wasserkraft. Dilpomarbeit Institut für Hydrologie, Wasserwirtschaft und Umwelttechnik, Ruhr-Universiaet Bochum, 2006.

Lang S., Möhrlen C., Jørgensen J., Ó Gallachóir B. P. & McKeogh E. J. Aggregate forecasting of wind generation on the Irish grid using a Multi-Scheme Ensemble Prediction System, Proc Conference on Renewable Energy in Maritime Climates, April, Dublin, 2006.

Lang, S., Möhrlen, C., Jørgensen, J., Ó Gallachóir B.P. & McKeogh, E.J. Application of a Multi-Scheme Ensemble Prediction System for wind power forecasting in Ireland and comparison with validation results from Denmark and Germany, Scientific Proceedings, European Wind Energy Conference, March, Athens, Greece, 2006.

For more details, please refer to the latest progress report.

Additional information

Project started in 2003.