Day-Ahead Load Forecasting Based on Conditional Linear Predictions with Smoothed Daily Profile

Abstract

In a smart grid, load forecasting has a crucial role in the efficient and sustainable operation of the power system by reducing uncertainties and enabling a pre-coordination to avoid any potential problems. This study proposes day-ahead load forecasting based on a statistical approach using the records of the previous days’ pre-processed through a smoothing approach. A prediction exploiting the latest available observations has the advantage of recency effect. However, the direct use of a full-day profile may cause a wide variation in the predicted outcomes owing to noisy and correlated observations. To avoid an undesired correlation effect, binning is employed to smooth the profile of the previous day, thereby enhancing the inter-day correlations. For a performance evaluation, the 3.5-year record of the 15-min sampled electric power of a campus was investigated for day-ahead load forecasting based on a correlation analysis between days. The results indicate that the inter-day correlation in a smoothed profile is improved compared to that in a raw data and that the day-ahead load forecasting yields smaller prediction errors on average with less variability through the proposed binning approach.