Short-Term Load Forecasting Using Regularized Greedy Forest-Based Ensemble Model

Binnie Wai-Keung Yiu

doi:10.1109/ACCESS.2024.3441642

Return

Short-Term Load Forecasting Using Regularized Greedy Forest-Based Ensemble Model

2024-01-01

Binnie Wai-Keung Yiu *

Tong Zhang

Cheuk Wing Lee

DOI：10.1109/ACCESS.2024.3441642

Original

Original request for help

Save

Abstract

En 中文

One of the key components of achieving sustainability and energy efficiency in power utility networks is accurate short-term load forecasting (STLF). STLF is essential for effective operational planning. Overestimation leads to unnecessary energy consumption and costs, while underestimation leads to energy shortages and potential blackouts, severely impacting the community and the economy. This study highlights the potential of regularized greedy forest (RGF) algorithm for STLF, which integrates the underlying tree structure with regularization to learn a decision forest. In particular, we propose the RGF model combined with two gradient-boosting frameworks, namely eXtreme gradient boosting and light gradient boosting machine models, to create a more robust ensemble model using Bayesian optimization techniques. The proposed ensemble model is evaluated in the real-world case study. It performs better than the existing models in terms of mean absolute percentage error (MAPE) and mean absolute error (MAE) in various scenarios. The experimental results show that the MAPE of the proposed model on average is 1.60% in the daily operation scenario, which provides accurate day-ahead (24 h) load forecasting with a half-day gap. The study demonstrates that the proposed model is suitable and practical for the STLF problem, and thus it contributes to the overall sustainability and resilience of power utility networks.

Keywords:

Ensemble method

machine learning

regularized greedy forest

short-term load forecasting

Ensemble method

machine learning

regularized greedy forest

short-term load forecasting