A Unifying Framework for Short-Term Load Forecasting via EM-CCGPBOATT Methodology

Short-term load forecasting (STLF) has long been a crucial research topic in energy management since it is significant to the secure and reliable operation of energy systems. The Ensemble Improved Fuzzy C-Means-Coordinated Grouping Polar Bear Optimization-Assisted TRUST-TECH (EM-CCGPBOATT), which has outstanding performance predicting capability and accuracy, is explored in this research as a unifying framework for STLF. This method first extracts similar user behavior with improved fuzzy C-means (FCM), which is defined as the clustering stage. Then, the optimal and diverse neural network candidates are generated based on the grouping polar bear optimization (GPBO) method in the training stage. In the exploring stage, multiple locally optimal solutions or even the globally optimal solutions of neural networks can be computed effectively. Finally, the final predictor is produced by the optimal ensemble of each diverse and excellent-quality neural network candidate, which is referred to the ensemble stage. Experiments on a variety of test circumstances demonstrate that the proposed EM-CCGPBOATT method shows a higher performance, which achieves better accuracy and generalization over other forecasting methods through three actual load datasets.