부산대학교 도서관

This paper proposes a real-time electrical load forecasting framework that supports two prediction modes: one-step-ahead and one-day-ahead. The one-step-ahead predictor relies on a feedback mechanism to reduce the impact of random electrical load activity on prediction results. A prediction evaluator assesses previous prediction outcomes to automatically determine the most suitable of the two forecasting modes and consistently ensure prediction accuracy. A training data generator is used to ensure the high quality of training data and decrease forecasting runtime. The proposed framework is evaluated empirically using a real-world power consumption dataset from the UC Irvine campus. Our results show that compared with traditional machine learning and deep learning approaches, it achieves consistently high prediction accuracy under a wide variety of evaluation metrics while relying solely on raw meter data, without any other input sources (e.g., weather data) or preprocessing steps. It therefore represents a promising approach in practice for accurate real-time electrical load forecasting in smart grids.