An improved neural network-based approach for short-term wind speed and power forecast

Abstract

Accurate forecasts of wind speed and wind power generation are essential for the effective operation of a wind farm. This paper presents an improved radial basis function neural network-based model with an error feedback scheme (IRBFNN-EF) for forecasting short-term wind speed and power of a wind farm, where an additional shape factor is included in the classic Gaussian basis function associated with each neuron in the hidden layer and a simple parameter initialization method is proposed to effectively find initial values of two key parameters of the basis function when performing neural network training. A wind farm near central Taiwan area connected to Taipower system is served as the measurement target. Provided with 24 h of input data at 10-min resolution (i.e. 6 × 24 input time steps) for training the proposed neural network, a look-ahead time up to 72 h (i.e. 6 × 72 forecasted output time steps) have been performed. Test cases for different months over 2014 are reported. Results obtained by the proposed model are compared with those obtained by four other artificial neural network-based forecasting methods. It shows that the proposed model leads to better accuracy for forecasting wind speed and wind power while the computational efficiency is maintained.