Hello,
I was wondering if someone could tell me "maximum how many days ahead can the existing wind (power)-forecasting tools can predict the wind speed/power with an hourly time resolution and with the least degree of accuracy?".
How many days or hours ahead? With how much acceptable error?
What about Solar and PV forecasting tools?
Best regards,
Morteza
http://mortezash.blogfa.com/
Morteza: This question's answer depends completely on the location involved. In Denmark and Germany, for wind power, they seem to agree that ARFIMA models provide the best estimates in the time window extending 4-7 hours into the future, and further on it is physical models (i.e. atmospheric simulation) that become better. The statistics in the US are also very different depending on the location; persistence (and therefore prediction) is much better in the West and Central US than in the Northeast, due to the fact that the boundary between Northern cool air and Southern warm air is chaotic as the wind blows towards the Eastern seaboard. The eddies present at that boundary make prediction in the Northeast much harder There are weather prediction services, such as IBM's Deep Thunder, which has been some 20 years in the making, which could make a big difference for power system operations, but they're expensive https://www-03.ibm.com/press/us/en/pressrelease/49954.wss . So, to recapitulate, you need to specify the location and then identify studies that are specific to it.
Thed answer to this question depends on the model fitted. For example, if the model is autoregressive of order p, the maximum h-step ahead forecast is for h=p. If it is a moving average model of order q, fore instance, since a finite order moving average is equivalent to an autoregression of infinite order, theoritically, the maximum h is infinite.
Have a look at the Fraunhofer IWES project:
https://www.fraunhofer.de/en/press/research-news/2016/June/better-forecasting-for-solar-and-wind-power-generation.html
Best Regards, Arnulf
In some areas of the world, such as the trade wind zones (especially the middle of the trade wind zones which don't come under the influence of the equatorial doldrums or the horse latitudes as the wind belts follow the Sun with the seasons), the winds are so reliable that climatological wind data may be sufficient --- it may not even be necessary to use a weather-forecasting model.
The best forecasts for wind-farm power output are made by (1) numerical weather-prediction models (based on the laws of physics, not statistics), followed by (2) interpolation to the site of the wind farm/turbine, and (3) statistical post-processing to remove systematic biases, based on "history" or "experience" (that can be learned). The conversion of "windspeed" to "power output" is based on the power-curves for the particular turbines on each farm - that is entirely local to each site, and should really be the last step.
After all that, depending on your site characteristics, I would expect a mean absolute error (MAE) of 1.0 to 1.5 m/s in wind-speed for a 24-hour forecast, rising to perhaps 1.5 - 2.0 m/s for a 48-hour forecast. Note those are just round, ball-park numbers. Any location or system that can get MAE to 1.0 m/s or less for 24-hr forecasts is doing very well - i.e., has crossed the threshold to provide real commercial value. Forecast accuracy over flat terrain is better than over mountains or hills.
See descriptions of such a system (and some result of its performance) at http://www.sciencedirect.com/science/article/pii/S1876610215016793 (esp. fig. 1 in there) and http://www.sciencedirect.com/science/article/pii/S1876610213016068 . You could use a different NWP model, or a different method for statistical post-processing bias removal, but I can't see how to improve on that basic concept.
I think you could expect hourly forecasts from national weather services out to about 2 days, followed by 3-hourly forecasts for a further 2 or 3 days, followed by 6-hourly forecasts after that. Your particular weather service may offer more or less. But I wouldn't put much faith in a highly-localized forecast beyond about 3 days. Forecast skill is typically better at mid-high latitudes rather than the tropics, and is typically better in winter than in summer.
The same principles could be applied to solar/PV power generation as are used for wind - but I'm not aware of any such system in existence. Models are probably better at forecasting wind than radiation reaching the surface (since clouds are parameterized in models, not explicitly resolved - and parameterized quite crudely at that), so I would expect PV power forecasts to be less accurate than wind-power forecasts.
Best wishes,
Enda
http://www.sciencedirect.com/science/article/pii/S1876610213016068
http://www.sciencedirect.com/science/article/pii/S1876610215016793
Hi Morteza
In general, forecasting accuracy is decreasing with increases of forecasting time horizons. Statistical models are usually perform well for short-term (30min to 6h ahead ) forecasting. And physical models doing well for long term (more than 1 day ahead) forecasting.
- Badges
- Science topic
- Similar topics
- Agriculture
- Soil Science