Development and Implementation of short-term wind speed forecasting for wind power plants in Costa Rica, using Artificial Intelligence methods
The recording and presentation of the Technical Webinar of this project could be found here.
About the project
The efficient operation of wind power plants (WPP) requires, among other things, a precise weather forecast. Accurate wind speed forecasts are essential not only for optimizing the performance of WPPs, but also for strategic recourse planning to meet electricity demand over the next few days. Numerical models such as WRF, GFS, ECMWF provide grid point data of wind speed and direction for several hours and days ahead. To achieve a higher-resolution forecast for specific locations, particularly in regions with complex terrain, it is essential to consider and adjust the discrepancies between observed wind data and numerical forecasts. One effective method for addressing this issue is through the utilization of statistical models, or recently through artificial intelligence regression and classification methods, including neural networks.
This project was therefore defined to develop and implement a high-resolution wind speed forecast for several wind power plants in Costa Rica to provide short-term wind forecasts from a few hours to two weeks ahead. The tool is developed in collaboration with the WMO and Costa Rican Electricity Institute (ICE) with support from the Costa Rican Meteorological service. Specific objectives of the project are listed below. To provide with some background information, Costa Rica is located in one of the narrowest regions in Central America, ranging in width from approximately 150 to 250km. The wind farms location close to the mountain top and near the ocean presents a particular challenge for numerical models. The limitations in describing the contrasts between land and sea, as well as the roughness and the terrain – especially in global models – are widely known.
Objective
- Improve wind speed forecasts for specific locations of wind power plants, using neural networks for downscaling Numerical Weather Prediction results
- Develop a scalable wind speed forecast model for different time scales, from hourly forecasts (1-, 3-, 6-, 12-, and 24-hours ahead) to daily forecasts up to 14 days ahead
- Integrate the developed model into the ICE platform, designing the required visualization functions for effective use
