Enhancing Short-Term Load Forecasting Through K-Shape Clustering and Deep Learning Integration

Abstract

Short-term load forecasting (STLF) is essential for the efficient operation and management of modern power grids, impacting dispatch and trading strategies in electricity markets. However, accurately forecasting short-term loads remains challenging due to the difficulty in categorizing diverse operational modes and the limited availability of exogenous variables such as temperature and economic indicators. To address these challenges, this study introduces K-NBEATSx, a novel model that integrates clustering and deep learning methodologies. The methodology begins by using K-Shape clustering to categorize electric load data based on shape similarity, effectively distinguishing different operational modes. Subsequently, the Neural Basis Expansion Analysis With Exogenous Variables (NBEATSx) method is applied by incorporating trend and seasonality modules to enhance forecasting accuracy. Case studies using load datasets from 3 different countries demonstrate that the proposed model outperforms traditional deep learning models across various operational scenarios. Additionally, the integration of clustering algorithms is validated as an effective strategy for improving prediction performance. This research offers an effective new methodology for deep-learning-based STLF, contributing to enhancing the reliability and efficiency of power system operation.

Keywords

• Short-term load forecasting (STLF)
• K-shape clustering
• deep learning
• NBEATSx
• power system operation