An Ensemble Broad Learning Scheme for Short-Term Load Forecasting

The management of electricity systems benefits considerably from short-term load forecasting. Although various superior short-term prediction algorithms have been investigated, some problems are still encountered in practical applications: (1) When dealing with data containing noise, there are unpredictable effects on short-term load forecasting. (2) Current deep learning models are time-consuming. In this study, we suggest an ensemble kernel broad learning system based on residuals to resolve the aforementioned issues. Specifically, we merge the broad learning system with kernel learning, which helps obtain stable features to improve the performance and avoid getting into the non-optimal problem. In addition, to further improve the prediction accuracy, we fuse multiple base learners to obtain an optimal prediction model from the residuals of the loss function. The experimental findings exhibit the efficacy and efficiency of our suggested technique.