3100. A Novel Forecasting Framework for Monthly Energy Consumption: AN Explainable Machine Learning Approach
Invited abstract in session WA-44: Forecasting methods and electricity markets, stream Energy Economics & Management.
Wednesday, 8:30-10:00Room: Newlyn 1.01
Authors (first author is the speaker)
| 1. | Hamid Eskandari
|
Abstract
This study presents an innovative framework that integrates feature selection (FS) methods with machine learning (ML) models to forecast national energy consumption (EC) across multiple energy sources in the United Kingdom. The proposed framework combines three FS approaches—ensemble filter, wrapper, and a hybrid filter-wrapper—with five interpretable ML models, employing Shapley Additive Explanations (SHAP) to enhance both the accuracy and transparency of EC predictions. By incorporating diverse features such as meteorological data, socioeconomic parameters, and historical consumption patterns, the framework optimizes the selection of relevant features, resulting in a more robust and interpretable forecasting model. Experimental results demonstrate the effectiveness of the ensemble FS approach, which refines the feature set and improves forecasting performance. The findings highlight that different FS methods yield varying feature subsets, which contribute to more precise predictions. This study not only advances the field of energy consumption forecasting by integrating advanced FS techniques with ML but also provides a valuable tool for policymakers and energy analysts, offering both accuracy and interpretability in EC predictions. The framework promises to guide future decision-making in energy resource allocation and policy planning.
Keywords
- Energy Policy and Planning
- Forecasting
- Machine Learning
Status: accepted
Back to the list of papers