Probabilistic Load Forecasting via Quantile Regression Averaging on Sister Forecasts

Although the probabilistic load forecasting literature can be traced back to 1970s, the importance of the subject was not well recognized until recent years. There are several approaches to producing probabilistic load forecasts, such as generating weather scenarios to feed to point forecasting models, applying probabilistic modeling and forecasting techniques, and identifying the density function of residuals. This paper starts a whole new category for probabilistic load forecasting methods - combining point load forecasts.

The proposed methodology includes two parts:

1. Generating sister point forecasts. The concept of sister models/forecasts is a brand new in this field. The models were first developed from my big data paper. This paper further materializes the concept and applies it to probabilistic load forecasting. 
2. Combining forecasts with quantile regression. While quantile regression is not a rare technique, there has not been many studies that apply quantile regression. In this paper, we use quantile regression to combining the point forecasts. The combination is dominantly better than the probabilistic load forecasts from individual models. 

The paper was accepted by IEEE Transactions on Smart Grid this week. The working paper is available HERE. I will update the citation once the paper is on IEEE Xplore.

Citation

Bidong Liu, Jakub Nowotarski, Tao Hong and Rafal Weron, "Probabilistic Load Forecasting via Quantile Regression Averaging on Sister Forecasts", IEEE Transactions on Smart Grid, accepted, working paper available from http://www.drhongtao.com/articles.

Probabilistic Load Forecasting via Quantile Regression Averaging on Sister Forecasts

Bidong Liu, Jakub Nowotarski, Tao Hong and Rafał Weron

Abstract

Majority of the load forecasting literature has been on point forecasting, which provides the expected value for each step throughout the forecast horizon. In the smart grid era, the electricity demand is more active and less predictable than ever before. As a result, probabilistic load forecasting, which provides additional information on the variability and uncertainty of future load values, is becoming of great importance to power systems planning and operations. This paper proposes a practical methodology to generate probabilistic load forecasts by performing Quantile Regression Averaging (QRA) on a set of sister point forecasts. There are two major benefits of the proposed approach: 1) it can leverage the development in the point load forecasting literature over the past several decades; and 2) it does not rely so much on high quality expert forecasts, which are rarely achievable in load forecasting practice. To demonstrate the effectiveness of the proposed approach and make the results reproducible to the load forecasting community, we construct a case study using the publicly available data from the Global Energy Forecasting Competition 2014. Comparing with several benchmark methods, the proposed approach leads to dominantly better performance as measured by the pinball loss function and the Winkler score.