A hybrid prediction model based on pattern sequence-based matching method and extreme gradient boosting for holiday load forecasting

摘要

In short-term load forecast (STLF), forecasting holiday load is one of the most challenging problems. Aimed at this problem, a hybrid prediction model based on pattern sequence-based matching method and extreme gradient boosting (XGBoost) is presented. It divides holiday STLF problem into the predictions for proportional curve and daily extremum of electricity demand, which are relatively independent and relate to different factors. It is benefit for holiday STLF by task decomposing. Based on the shape similarity measured by Euclidean distance, the proportional curve is predicted by pattern sequence-based matching method. Daily extremum of electricity demand is predicted by XGBoost considering holiday classification. Finally, the predicted holiday load profile is synthesized from the above two prediction results with segment correction. The proposed methodology can analyze holiday load characteristics more effectively and get a higher prediction accuracy independent of sufficient data and expert experience. We evaluate our methodology with many algorithms on a real data set of one provincial capital city in eastern China. The results of case studies show that the proposed methodology gives much better forecasting accuracy with an average error 2.98% in holidays.