Short-term load forecasting by a hybrid attention scheme: Multi-feature attention and context awareness

Fang Su

doi:10.1016/j.ijepes.2025.111065

Short-term load forecasting by a hybrid attention scheme: Multi-feature attention and context awareness

10.1016/j.ijepes.2025.111065

2025-09-16

0

PRE

AI

Save

Share

Original

0Request

Assist

Abstract

En 中文

• We propose a multi-feature attention (MFA) mechanism that adaptively adjusts feature weights across different input steps based on varying input–output conditions, thereby enhancing both the model’s trainability and flexibility. • We employ a context awareness mechanism that dynamically weights contextual information by evaluating its relevance to encoder output, thus improving the influence of critical information and improving the precision of model forecasting. • We develop a novel load forecasting architecture, termed MFACA, that integrates the MFA and context awareness layer (CA) within an encoder–decoder architecture. This architecture is jointly trained using backpropagation, significantly improving the robustness of the model. • The effectiveness of the developed model was evaluated using various datasets from real-world power load and consistently demonstrates superior forecast performance compared to other models under various forecast conditions.

Keywords:

Context awareness

Encoder–decoder architecture

Multi-feature attention

Short-term load forecasting

Short-term load forecasting by a hybrid attention scheme: Multi-feature attention and context awareness

Abstract

AI Summary

Journal

Researchers

Organization

Cited Papers

Citing Papers