Short-term load prediction with a special emphasis on weather compensation using a novel committee of wavelet recurrent neural networks and regression methods

In this paper, a novel committee of wavelet and recurrent neural networks to predict the next hour, 24 hour and one-week-ahead load is addressed. Using wavelet multiresolution analysis, the load series are decomposed to different sub-series, which show the different frequency characteristics of the load. Different recurrent neural networks are optimally designed and developed to predict each sub-series according to its characteristics, finally the best recurrent neural network on each sub-series is chosen based on the performance measures such as mean square error, correlation coefficient and mean absolute percentage error on prediction dataset. Feasibility of Daubechies wavelet at different scales with suitable number of decomposition levels is investigated to choose the best mother wavelet for different seasonal load series. The estimated models are evaluated over different weather parameters in order to judge the impact on accurate load prediction. The reliability and consistency in prediction by the adopted technique is proved even in the presence of controlled Gaussian noise to the predicted temperature series. The traditional regression models are developed for the same data as a benchmark. The results are compared with traditional statistical techniques and offered a high prediction precision.