An ensemble of echo state networks for predicting the energy production of wind plants

The electricity grid relies on a mixture of conventional (e.g. oil and gas) and renewable (e.g. wind, solar, and geothermal) energy sources. Ensuring the reliability of electric power distribution becomes a fundamental and complex issue due to the stochasticity of the production from renewable sources and the fluctuating behaviour of energy market demand. The necessity to integrate and mitigate these two uncertainty sources requires to tackle the problem of the unit commitment. It is, therefore, fundamental the capability of predicting electrical power output from plants with intermittent energy sources. We propose an approach to predict wind energy production based on an ensemble of Echo State Networks (ESNs) trained with different sets of historical data. A novel Local Fusion (LF) strategy is employed to aggregate the predictions of the individual ESN models. The proposed approach is applied to the prediction of the energy production of a wind plant located in Italy. The obtained results show that the proposed ensemble provides more accurate predictions than a single ESN model and an ensemble approach of literature