Daily wind speed forecasting through hybrid AR-ANN and AR-KF models

The nonlinearity and the chaotic fluctuations in the wind speed pattern are the reasons of inaccurate wind speed forecasting results using a linear autoregressive integrated moving average (ARIMA) model. The inaccurate forecasting of ARIMA model is a problem that reflects the uncertainty of modelling process. This study aims to improve the accuracy of wind speed forecasting by suggesting more appropriate approaches. An artificial neural network (ANN) and Kalman filter (KF) will be used to handle nonlinearity and uncertainty problems. Once ARIMA model was used only for determining the inputs structures of KF and ANN approaches, using an autoregressive (AR) Instead of ARIMA may be resulted in more simplicity and more accurate forecasting. ANN and KF based on the AR model are called hybrid AR-ANN model and hybrid AR-KF model, respectively. In this study, hybrid AR-ANN and hybrid AR-KF models are proposed to improve the wind speed forecasting. The performance of ARIMA, hybrid AR-ANN, and hybrid AR-KF models will be compared to determine which had the most accurate forecasts. A case study will be carried out that used daily wind speed data from Iraq and Malaysia. Hybrid AR-ANN and AR-KF models performed better than ARIMA model while the hybrid AR-KF model was the most adequate and provided the most accurate forecasts. In conclusion, the hybrid AR-KF model will result in better wind speed forecasting accuracy than other approaches, while the performances of both hybrid models will be provided acceptable forecasts compared to ARIMA model that will provide ineffectual wind speed forecasts.