Towards Automated Model Selection for Wind Speed and Solar Irradiance Forecasting.

Area of Science:

• Energy Science
• Electrical Engineering
• Environmental Science

Background:

• Growing electricity demand necessitates integrating renewable energy sources (RESs), primarily solar and wind power.
• Variability in solar irradiance and wind speed poses challenges for stable RES integration into power grids.
• Accurate forecasting is essential for efficient operation, cost-effectiveness, and grid reliability.

Purpose of the Study:

• To evaluate various models for medium-term (24-hour ahead) forecasting of wind speed and solar irradiance.
• To analyze the impact of preprocessing steps and different forecasting algorithms (classical and deep learning).
• To identify key features (time steps) influencing prediction accuracy.

Main Methods:

• Utilized real-time measurement data from Crete, Greece.
• Applied diverse preprocessing techniques to the time series data.
• Compared the performance of classical and deep learning forecasting models.
• Investigated the influence of specific time steps on forecast accuracy.

Main Results:

• Identified the most accurate forecasting model for the specific case study.
• Demonstrated that certain time steps significantly impact prediction accuracy.
• Highlighted the importance of feature engineering related to time steps.

Conclusions:

• Accurate medium-term forecasting of wind speed and solar irradiance is achievable with appropriate model selection and feature engineering.
• The choice of time steps is a critical factor in improving forecasting performance for renewable energy.
• Extensive model and feature searches are recommended for similar renewable energy forecasting studies.