Load forecasting impacts directly financial returns and information in electrical systems planning. A promising approach to load forecasting is the Echo State Network (ESN), a recurrent neural network for the processing of temporal dependencies. The low computational cost and powerful performance of ESN make it widely used in a range of applications including forecasting tasks and nonlinear modeling. This paper presents a Bayesian optimization algorithm (BOA) of ESN hyperparameters in load forecasting with its main contributions including helping the selection of optimization algorithms for tuning ESN to solve real-world forecasting problems, as well as the evaluation of the performance of Bayesian optimization with different acquisition function settings. For this purpose, the ESN hyperparameters were set as variables to be optimized. Then, the adopted BOA employs a probabilist model using Gaussian process to find the best set of ESN hyperparameters using three different options of acquisition function and a surrogate utility function. Finally, the optimized hyperparameters are used by the ESN for predictions. Two datasets have been used to test the effectiveness of the proposed forecasting ESN model using BOA approaches, one from Poland and another from Brazil. The results of optimization statistics, convergence curves, execution time profile, and the hyperparameters’ best solution frequencies indicate that each problem requires a different setting for the BOA. Simulation results are promising in terms of short-term load forecasting quality and low error predictions may be achieved, given the correct options settings are used. Furthermore, since there is not an optimal global optimization solution known for real-world problems, correlations among certain values of hyperparameters are useful to guide the selection of such a solution.
Time series forecasting plays a key role in many areas of science, finance and engineering, mainly for the estimation of trend or seasonality of a variable under observation, aiming to serve as basis for future purchase decisions, choice of design parameters or maintenance schedule. Artificial Neural Networks (ANNs) have proven to be suitable in linear or nonlinear functions mapping. However, the ANNs, implemented in its most simplistic form, tend to have a loss in overall performance. This work aims to obtain a prediction model for a short-term load problem through the usage of wavenets ensemble, which is an ANN approach capable in combining the best characteristics of each ensemble component, in order to achieve a higher overall performance. We adopted the usage of bootstrapping, cross-validation and the inputs decimation approaches for the ensemble construction. For the components selection, `constructive' and `no selection' methods were applied. Finally, the combination is held though simple average, mode or stacked generalization. The results show that it is possible to improve the generalization ability through effective committees depending on the methods used to construct the ensemble. The total relative improvement achieved in respect to the naive model, was over 95%, regardless the number of sub wavenets, and for the best component, the relative improvement was 93.91% using five wavenets. We conclude that the most frequent and effective set, but not always with the lower MSE (Mean Squared Error), was using constructive bagging with simple average.