.png){kind=icon}
Return
Short-term multi-regional load forecasting considering dynamic spatial features with missing data
.png){kind=icon}
PRE
AI
DOI：10.1016/j.enbuild.2025.116450.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Accurate multi-region load forecasting is the core requirement for ensuring the stable operation of the power grid and improving the economic dispatching efficiency of the power system. However, achieving accurate multi-region load forecasts in practice faces two major challenges. First, data is frequently missing due to device failures, network congestion, and operational errors. Second, modeling the complex dynamic spatial dependencies among regional loads is challenging. To address these challenges, a short-term multi-regional load forecasting method based on causal self-attention and multi-head graph attention network (GAT) is proposed in this paper. Specifically, the missing data reconstruction module is constructed based on the causal self-attention mechanism (CSAM), which reconstructs the missing data and ensures feature distribution consistent through time-series causal inference. Subsequently, discrete wavelet transform (DWT) converts multi-regional load data into the time–frequency domain, enhancing transient feature representation through multi-scale decomposition. The dynamic spatial feature extraction module then employs multi-head graph attention networks, which capture dynamic spatial dependencies by adaptively aggregating neighbor information. Finally, experiments using real multi-regional load data from the New York Independent System Operator (NYISO) validate the method. The experimental results show that the MAE, RMSE and MAPE decrease by 22.96%, 23.02% and 20.66% on average compared with the prediction results of several mainstream models. It demonstrates the superiority of the model proposed in this paper in forecasting multi-regional loads.
Keywords:
Multi-Region Load Forecasting
Missing Data
Multi-Scale Features
Dynamic spatial dependencies
Journal
IF:
7.1
Papers:
1.6W
Citations:
6.8W
Organization
No organization information available
Cited Papers
Short-Term Electrical Load Forecasting Using an Enhanced Extreme Learning Machine Based on the Improved Dwarf Mongoose Optimization Algorithm.png)
Symmetry
OAAI
SymmetryIF0
OAAIShareSaveShareSaveData Augmentation Technique Based on Improved Time-Series Generative Adversarial Networks for Power Load Forecasting in Recirculating Aquaculture SystemsSUSTAINABILITYOAAI
SUSTAINABILITYIF3.3
OAAIShareSaveShareSaveAdvances in Deep Learning Techniques for Short-term Energy Load Forecasting Applications: A Review
PREAI
ShareSave
View more