Two Machine Learning Approaches for Short-Term Wind Speed Time-Series Prediction.

Area of Science:

• Energy Systems Engineering
• Computational Intelligence
• Renewable Energy Forecasting

Background:

• Liberalizing European electricity markets and integrating intermittent renewables necessitate intelligent power grids.
• Challenges in energy consumption patterns, like electric mobility, demand flexible energy production and distribution.
• Intermittent renewable energy sources (wind, solar) pose engineering and economic challenges due to limited dispatchability.

Purpose of the Study:

• To compare two machine learning approaches for assessing prediction intervals (PIs) in time-series wind speed forecasting.
• To evaluate the effectiveness of multilayer perceptron neural networks and extreme learning machines for short-term wind speed prediction.
• To provide confidence information in wind speed predictions, which is more valuable than point estimates.

Main Methods:

• Utilized multilayer perceptron neural networks trained with a multiobjective genetic algorithm.
• Employed extreme learning machines combined with the nearest neighbors approach.
• Applied both methods to a real dataset of hourly wind speed measurements for short-term prediction in Regina, Saskatchewan, Canada.

Main Results:

• Both machine learning approaches demonstrated good prediction precision for short-term wind speed.
• The study successfully assessed prediction intervals, offering confidence levels for the forecasts.
• Each method presented complementary advantages based on different evaluation criteria.

Conclusions:

• The developed machine learning models are effective for short-term wind speed prediction.
• Prediction intervals provide crucial confidence information for managing intermittent renewable energy integration.
• The comparison offers insights into selecting appropriate methods for wind power forecasting in smart grids.