Wind is one of the fastest growing renewable energy sources of electricity in Canada. However, due to the natural variability of wind, the integration of wind energy into electrical power systems is challenging. Variable and fluctuating wind power causes numerous problems in power quality, system stability, and energy dispatch. These problems become more severe as the penetration level of wind energy increases. Therefore, it is desirable that accurate wind power forecasts are developed and implemented to reduce the risks of uncertainty in wind generation, improve stability of power systems and optimize economic dispatch. However, most wind forecasting methods in recent research and industrial applications require either a detailed physical description or considerable historical data of the investigated wind farm, which poses a challenge to a new wind farm and wind farms with upgrades.

The objective of this project is to develop wind forecasting methods based on little to no historical data for short-term wind power production forecasts.

Three datasets will be provided as: Dataset #1 will include one-year hourly produced wind speed and wind direction forecast data of Wind Farm A, as well as the measurement information of wind speed and wind farm power production which is with a time-resolution of five minutes.

Dataset #2 will include one-month hourly produced wind speed and wind direction forecast data of Wind Farm B, as well as the measurement information of wind speed and wind farm power production which is with a time-resolution of five minutes.

Dataset #3 will be used as a test dataset, providing another three-month hourly wind speed and wind direction forecast data of Wind Farm B, as well as the measurement information of wind speed and wind farm power production which is with a time-resolution of five minutes.

The expected results will include

According to Dataset #1, teams should develop their forecasting algorithms to provide a short-term wind power production forecast every hour up to 6.5 hours ahead with a time resolution of five minutes. Some common metrics such as Mean Average Error (MAE), Mean absolute percentage error (MAPE), or Root Mean Square Error (RMSE), could be used for performance evaluation. Other self-developed metrics could be also recommended to employ as well.

According to the knowledge learnt from (1), teams need to investigate how to provide accurate the wind forecast service as (1) based on only one-month (even less) data provided by Dataset #2.

The last step will be that using the Dataset #3 to test the forecasting accuracy of (2).

Built With

  • wind
Share this project:

Updates