Recursive wind speed forecasting based on Hammerstein Auto-Regressive model

• Developed a new recursive WSF model for 1–24 h horizon based on Hammerstein model.
• Nonlinear HAR model successfully captured chaotic dynamics of wind speed time series.
• Recursive WSF intrinsic error accumulation corrected by applying rotation.
• Model verified for real wind speed data from two sites with different characteristics.
• HAR model outperformed both ARIMA and ANN models in terms of accuracy of prediction.

A new Wind Speed Forecasting (WSF) model, suitable for a short term 1–24 h forecast horizon, is developed by adapting Hammerstein model to an Autoregressive approach. The model is applied to real data collected for a period of three years (2004–2006) from two different sites. The performance of HAR model is evaluated by comparing its prediction with the classical Autoregressive Integrated Moving Average (ARIMA) model and a multi-layer perceptron Artificial Neural Network (ANN). Results show that the HAR model outperforms both the ARIMA model and ANN model in terms of root mean square error (RMSE), mean absolute error (MAE), and Mean Absolute Percentage Error (MAPE). When compared to the conventional models, the new HAR model can better capture various wind speed characteristics, including asymmetric (non-gaussian) wind speed distribution, non-stationary time series profile, and the chaotic dynamics. The new model is beneficial for various applications in the renewable energy area, particularly for power scheduling.

Figure optionsDownload as PowerPoint slide