Residential consumer behavior-aware short-term load forecasting using EM-MCGPBOASP methodology

Short-term load forecasting (STLF) is of critical importance for ensuring the secure and reliable operation of power systems. In this paper, a unifying STLF framework based on the EM-MCGPBOASP method that incorporates residential consumption behavior is proposed, which systematically integrate four synergistic stages with differentiated functional role: the user personas stage, the training stage, the exploring stage, and the ensemble stage. In the initial stage, user personas are systematically conducted to delineate distinct electricity consumption patterns across different groups, which facilitates the precise quantification of inter-user electricity behavior variability, thereby significantly enhancing the adaptability of the forecasting model to fine-grained application scenarios. The proposal of the training and exploring stage enables the acquisition of a population of high-precision base learners with diverse architectures and parameters. In the ensemble stage, the final predictor is constructed by nonlinear combination, which not only effectively mitigates the interference caused by abnormal predictions of individual base learners but also fully integrates the collective advantages of all learners, significantly improving the generalization and accuracy of predictions. Experimental results demonstrate that the constructed forecasting technology outperforms other benchmark methods in both daily and weekly load forecasting for different regions, exhibiting superior accuracy and generalization performance.​