Ensemble Nonlinear Autoregressive Exogenous Artificial Neural Networks for Short-Term Wind Speed and Power...

Area of Science:

• Renewable Energy Systems
• Artificial Intelligence in Meteorology
• Computational Fluid Dynamics

Background:

• Accurate short-term wind speed and power forecasting are crucial for grid integration of wind energy.
• Existing methods often lack robust uncertainty quantification.
• Numerical weather prediction and computational fluid dynamics provide valuable but complex wind field data.

Purpose of the Study:

• To develop and validate an ensemble artificial neural network (ANN) methodology for 72-hour wind speed and power forecasting.
• To incorporate numerical weather prediction (NWP) and computational fluid dynamics (CFD) data as exogenous inputs.
• To quantify forecast uncertainties arising from ANN structure and weight initialization.

Main Methods:

• Utilizing a nonlinear autoregressive exogenous (NARX) artificial neural network (ANN) architecture.
• Employing an ensemble approach to combine predictions from multiple ANNs.
• Training ANN members using particle swarm optimization (PSO).
• Integrating NWP or high-resolution CFD wind field data as exogenous inputs.

Main Results:

• The ensemble ANN approach demonstrated improved predictive skills for wind speed and power compared to single ANN models.
• The methodology successfully quantified uncertainties associated with the forecasts.
• Validation using data from an operational wind farm in Northern China confirmed the effectiveness of the proposed method.

Conclusions:

• The ensemble ANN methodology offers enhanced accuracy and crucial uncertainty information for short-term wind power forecasting.
• This approach provides a more reliable forecasting tool for wind farm operations and grid management.
• The integration of NWP/CFD data and ensemble techniques represents a significant advancement in wind energy forecasting.