A Method for Ultra-Short-Term Regional Load Forecasting Based on Contrastive Learning-Based Feature Extraction and Hierarchical Modeling of Spatiotemporal Correlations

Accurate ultra-short-term regional load forecasting provides critical support for power grid dispatch and electricity market transactions. Modeling the spatio-temporal correlations among load nodes within a region is an effective method for improving the accuracy of regional load forecasting. However, existing methods for ultra-short-term regional load forecasting suffer from issues such as limited and redundant load feature extraction, as well as insufficient modeling depth of graph structures, making it difficult to accurately capture the spatio-temporal correlations of loads within the region. To address this, this paper proposes a method for ultra-short-term regional load forecasting based on contrastive learning for feature extraction and hierarchical modeling of spatio-temporal correlations. First, conventional K-means clustering is applied directly to the historical load time series, using Euclidean distance to partition substations with similar electricity-consumption patterns into sub-regional clusters. Next, contrastive learning is independently employed to extract deep temporal representations from the load data, thereby reducing noise and redundant information and uncovering latent temporal features. Subsequently, a dynamic adjacency matrix is constructed to characterize the time-varying correlations among nodes, establishing a hierarchical graph structure both within and between subgraphs; finally, a dynamic hierarchical graph convolutional network is employed to perform regional load forecasting. Simulation results based on the Ausgrid dataset demonstrate that the proposed method achieves the lowest RMSE and MAE among the evaluated methods for ultra-short-term forecasts ranging from 1 to 16 steps, effectively adapting to the highly random and volatile characteristics of regional load.

Authors

Institutions

Publication Details

Journal
Electronics
Published
2026-11-17
DOI
https://doi.org/10.3390/electronics15225343
Primary Topic
Energy Load and Power Forecasting
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A Method for Ultra-Short-Term Regional Load Forecasting Based on Contrastive Learning-Based Feature Extraction and Hierarchical Modeling of Spatiotemporal Correlations

Ling Hao, Fei Xu, Zhaoyang Yan, Yuze Zhou et al.
2 links
Electronics
Energy Load and Power Forecasting
article

A Method for Ultra-Short-Term Regional Load Forecasting Based on Contrastive Learning-Based Feature Extraction and Hierarchical Modeling of Spatiotemporal Correlations

Ling Hao, Fei Xu, Zhaoyang Yan, Yuze Zhou, Yong Hu, Lizi Deng
article en

Abstract

Accurate ultra-short-term regional load forecasting provides critical support for power grid dispatch and electricity market transactions. Modeling the spatio-temporal correlations among load nodes within a region is an effective method for improving the accuracy of regional load forecasting. However, existing methods for ultra-short-term regional load forecasting suffer from issues such as limited and redundant load feature extraction, as well as insufficient modeling depth of graph structures, making it difficult to accurately capture the spatio-temporal correlations of loads within the region. To address this, this paper proposes a method for ultra-short-term regional load forecasting based on contrastive learning for feature extraction and hierarchical modeling of spatio-temporal correlations. First, conventional K-means clustering is applied directly to the historical load time series, using Euclidean distance to partition substations with similar electricity-consumption patterns into sub-regional clusters. Next, contrastive learning is independently employed to extract deep temporal representations from the load data, thereby reducing noise and redundant information and uncovering latent temporal features. Subsequently, a dynamic adjacency matrix is constructed to characterize the time-varying correlations among nodes, establishing a hierarchical graph structure both within and between subgraphs; finally, a dynamic hierarchical graph convolutional network is employed to perform regional load forecasting. Simulation results based on the Ausgrid dataset demonstrate that the proposed method achieves the lowest RMSE and MAE among the evaluated methods for ultra-short-term forecasts ranging from 1 to 16 steps, effectively adapting to the highly random and volatile characteristics of regional load.

ElectronicsVol. 15(22)
State Key Laboratory of Power System Operation and Control (CN), State Grid Jiangsu Electric Power (China) (CN), Tsinghua University (CN)
Affordable and clean energy, Industry, innovation and infrastructure, Climate action
Energy Load and Power Forecasting
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon