An Online Probability Density Load Forecasting Against Concept Drift Under Anomalous Events
Chaojin Cao, Yaoyao He
Abstract
Concept drift (i.e., the data pattern to be learned can change over time) is becoming more common in power loads due to the volatile external environment, posing huge challenges on load forecasting. The phenomenon could be exacerbated by anomalous events, such as the unprecedented coronavirus disease 2019 (COVID-19) pandemic. This article focuses on handling the problem of concept drift in probabilistic load prediction under abnormal events. To address this challenge, we propose a novel online model for probability density load forecasting, which utilizes least absolute shrinkage and selection operator combined with quantile regression long short-term memory network as the base learner to capture time dependencies in abnormal events. Continuous ranked probability score integrated with kernel density estimation is developed to monitor the performance of probabilistic model. Two online modules, buffering and tuning module, will timely adjust the model parameters according to the performance to adapt to the new concepts in the data. Data from the COVID-19 period verify the effectiveness of the proposed model in dealing with concept drift.