Short-term wind speed forecasting based on spectral clustering and optimised echo state networks

Predicting the wind speed at multiple time points over a time span between two and 4 h typically requires a multi-input/multi-output model. This study investigates a wind speed forecasting method based on spectral clustering (SC) and echo state networks (ESNs). A wavelet transformation was used to decompose the wind speed into multiple series to eliminate irregular fluctuation. The decomposed series were modelled separately. For every decomposed wind speed series, principal component analysis was used to reduce the number of variables and thus the redundant information among the input variables. SC was used to select similar samples from the historical data to form training and validation sets. An ESN was used to simultaneously predict multiple outputs, and a genetic algorithm was employed to optimise the ESN parameters and ensure the forecast accuracy and the generalisation of the model. The forecasts of the decomposed series were summed to get the wind speed. Tests based on actual data show that the proposed model can simultaneously forecast wind speeds at multiple time points with high efficiency, and the accuracy of the proposed model is significantly higher than that of the traditional models.