Multi-step ahead wind power forecasting based on N-BEATS model

For wind power systems to operate successfully and efficiently, accurate forecasting is critical. The integration of large wind farms with smart grid technology mainly depend on accurate models for wind power output prediction. This study introduces a neural basis expansion approach named N-BEATS (neural basis expansion analysis for interpretable time series forecasting) for time series multi-step wind power forecasting. N-BEATS is a deep neural network architecture composed of multiple fully connected layers, arranged with both forward and backward residual connections. The performance of N-BEATS was benchmarked using classical approaches including AutoRegressive Integrated Moving Average (ARIMA), Seasonal AutoRegressive Integrated Moving Average (SARIMA), Holt-Winters, and the recurrent neural network named Long Short-Term Memory (LSTM). All models hyperparameters were tuned by Bayesian optimization. The N-BEATS model demonstrated superior forecasting scores and outperformed traditional methods in metrics such as mean absolute error and root mean square error for different forecasting horizons.