Project start date: subject to availability
Application deadline: open until filled
Description:
The integration of solar energy into the power system faces significant challenges due to its natural intermittency and uncertainty. Solar power generation forecasting is considered as a cost-effective method to mitigate its negative impact and therefore to ensure the grid stability.
With the advancement in numerical weather prediction (NWP) models, this master thesis project aims to develop an NWP-based model for regional solar power generation forecasting using both data-driven and physical methods. The data-driven method is to correct the bias of NWP forecasts (e.g., irradiance, air temperature, and wind speed), and the physical method is to convert these refined NWP forecasts to photovoltaic (PV) power production as shown in Fig.1. Therefore, the following two tasks will be accomplished within this project:
1. Develop machine learning models for post-processing or downscaling NWP forecasts (e.g., ECMWF or ICON from MeteoSwiss) to improve their spatial resolution and accuracy.
2. Convert these improved NWP forecasts to PV power generation for the region of interest (e.g., Switzerland) using the physical method.
Your responsibilities:
1. Collect and process data of NWP forecasts and ground measurements.
2. Develop machine/deep learning methods for downscaling and post-processing NWP forecasts.
3. Use the physical method to convert these NWP forecasts to PV power production.
4. Analyse and visualize the results for locations with different climate features.
Your qualifications:
1. Good data processing and visualization skills with Python or Matlab. Experience on large scale datasets would be a plus.
2. Experience with machine/deep learning, especially on image processing and computer vison.
3. Demonstrable interest in weather forecasting and solar energy.
This project will provide hands-on experience on dealing with NWP datasets, developing machine learning models for real applications, and data analysis and visualizations. Depending on the progress, probabilistic PV power forecasting may also be explored to better quantify the uncertainty. For more information or to express your interest in this project, please contact:
Prof. Fernando Porté-Agel (Email: [email protected])
Dr. Shanlin Chen (Email: [email protected])
Wind Engineering and Renewable Energy Laboratory