# Probabilistic Load Forecasting via Point Forecast Feature Integration

**Authors:** Qicheng Chang, Yishen Wang, Xiao Lu, Di Shi, Haifeng Li, Jiajun Duan, Zhiwei Wang  
**arXiv:** [1903.10684](https://arxiv.org/abs/1903.10684)  
**Date:** 2019-03-26  
**Summary page:** https://www.summarizepaper.com/en/paper/1903.10684v1/

## Key points

- Importance of short-term load forecasting in power systems energy management
- Growing interest in probabilistic load forecasting (PLF) due to its ability to provide uncertainty information
- Two-stage framework proposed for improving PLF by integrating point forecast as a key feature
- First stage: training a point forecast model and determining feature importance
- Second stage: developing a forecasting model that considers both point forecast features and selected feature subsets
- Final probabilistic load forecast results used for obtaining both point forecasting and probabilistic forecasting outcomes
- Effectiveness demonstrated through numerical results obtained from ISO New England demand data
- Proposed framework employs gradient boosting regression for point forecasting and quantile regression neural networks for probabilistic forecasting
- Comprehensive framework for probabilistic load forecasting by integrating point forecast features
- Potential to improve hour-ahead load forecasting accuracy while providing valuable uncertainty information for power system operators

## Summary

The paper titled "Probabilistic Load Forecasting via Point Forecast Feature Integration" addresses the importance of short-term load forecasting in power systems energy management. The authors highlight the growing interest in probabilistic load forecasting (PLF) due to its ability to provide uncertainty information that enhances system operation reliability and economics. To improve PLF, the paper proposes a two-stage framework that integrates point forecast as a key probabilistic forecasting feature. In the first stage, all relevant features are used to train a point forecast model and determine feature importance. This step helps identify the most influential factors for load forecasting. In the second stage, the authors develop a forecasting model that considers both point forecast features and selected feature subsets. During testing, the final probabilistic load forecast results are utilized to obtain both point forecasting and probabilistic forecasting outcomes. The effectiveness of this approach is demonstrated through numerical results obtained from ISO New England demand data. The proposed framework employs gradient boosting regression for point forecasting and quantile regression neural networks for probabilistic forecasting. Overall, this paper presents a comprehensive framework for probabilistic load forecasting by integrating point forecast features which can help improve hour-ahead load forecasting accuracy while providing valuable uncertainty information for power system operators. The results indicate its potential in improving accuracy while providing useful uncertainty information for power system operators.