A Moving-Horizon Dynamic Mode Decomposition Method for Daily Residential Electric Consumption Prediction

In this paper, a data-driven method is proposed for the prediction of daily residential electric consumption. The proposed method is based on the dynamic mode decomposition (DMD) approach and the moving-horizon principle, resulting in the so-called moving-horizon DMD (MH-DMD) method. The essence of this MH-DMD method is the constant update of the historical data used for the DMD model training, after each prediction period. This is different from the conventional DMD approach where a set of historical input data is chosen and fixed for the derivation of a model which is then utilized to predict all the future system states. Numerical simulations with real-world data, under different sampling times, show the superiority of the proposed MH-DMD method over the classical DMD approach.