An Extensible Framework for Short-Term Holiday Load Forecasting Combining Dynamic Time Warping and LSTM Network

Due to the extreme change of human behavior, load consumption in public holidays fluctuates more significantly compared general weekdays resulting difficulty hourly holiday forecasting. The forecasting is even challenging because forecast practically predicted on nearest workday which might be than one days prior holiday. In this paper, we propose a Joint Dynamic time warping and LSTM (JDL) framework, predict at least day before incoming proposed JDL hybrid short-term framework combines dynamic (DTW) long-short term memory (LSTM) network. DTW predicts any preceding compensatory holiday(s), if any, based similar occurrence pattern. highly unpredictable target by univariate multivariate models. Current results show outperforms others.