Bi-Mamba-Load: A bidirectional structured state space model with sequence relationship adaptive perception for electricity load forecasting
- 1Information and Communication Branch, Southern Power Grid Energy Storage Co, Guangzhou, Guangdong, China
Abstract
The accurate prediction of electricity load is a critical component for smart-grid management, energy allocation, and grid stability. Traditional statistical methods and Convolutional Neural Networks (CNNs) often struggle with nonlinear relationships or long-range temporal dependencies, whereas Transformer architectures incur quadratic complexity in the sequence length. The Structured State Space Model Mamba provides linear sequence modeling but uses a unidirectional scan and a fixed treatment of multivariate dependencies. To address these limitations, we propose Bi-Mamba-Load, which combines a bidirectional Mamba encoder, a custom forget gate, and a Sequence Relationship Adaptive Perception (SRAP) module. Both encoder branches operate exclusively on the observed historical window, so no observations from the forecasting horizon are used as inputs. SRAP computes Spearman correlation statistics from the training split and performs dataset-level selection between channel-independent and channel-mixing tokenization; the selected route remains fixed for all samples in that dataset. Experiments across eight benchmark datasets indicate competitive or superior forecasting accuracy against the evaluated baselines while retaining linear scaling with the input length, subject to the constant-factor overhead of bidirectional scanning.
Related
- Long-term electricity load forecasting model based on multi-period and multi-scale analysis
- Causal relationship between cardiac arrhythmias and sepsis: a two-sample bidirectional mendelian randomization study
- Green finance curbs direct carbon emissions but exacerbates lifecycle carbon emissions in China’s electricity sector
- Bidirectional interactions between REM sleep and stress responses through the lens of the resilience framework
- Force Induces a Collective Inactivity in Dyneins to Auto-Regulate Bidirectional Transport of Vesicles
BibTeX
@article{zhu2026bimambaload,
title = {{Bi-Mamba-Load: A bidirectional structured state space model with sequence relationship adaptive perception for electricity load forecasting}},
author = {Junwei Zhu and Lingzhi Xi and Chongyang Liao and Baoqiang Zhong and Jianxu Zhong and Lei Zhang},
journal = {PLOS One},
year = {2026},
month = nov,
doi = {10.1371/journal.pone.0355805},
url = {https://doi.org/10.1371/journal.pone.0355805},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Bi-Mamba-Load: A bidirectional structured state space model with sequence relationship adaptive perception for electricity load forecasting
AU - Junwei Zhu
AU - Lingzhi Xi
AU - Chongyang Liao
AU - Baoqiang Zhong
AU - Jianxu Zhong
AU - Lei Zhang
PY - 2026
DA - 2026/11/02
JO - PLOS One
DO - 10.1371/journal.pone.0355805
UR - https://doi.org/10.1371/journal.pone.0355805
AB - The accurate prediction of electricity load is a critical component for smart-grid management, energy allocation, and grid stability. Traditional statistical methods and Convolutional Neural Networks (CNNs) often struggle with nonlinear relationships or long-range temporal dependencies, whereas Transformer architectures incur quadratic complexity in the sequence length. The Structured State Space Model Mamba provides linear sequence modeling but uses a unidirectional scan and a fixed treatment of multivariate dependencies. To address these limitations, we propose Bi-Mamba-Load, which combines a bidirectional Mamba encoder, a custom forget gate, and a Sequence Relationship Adaptive Perception (SRAP) module. Both encoder branches operate exclusively on the observed historical window, so no observations from the forecasting horizon are used as inputs. SRAP computes Spearman correlation statistics from the training split and performs dataset-level selection between channel-independent and channel-mixing tokenization; the selected route remains fixed for all samples in that dataset. Experiments across eight benchmark datasets indicate competitive or superior forecasting accuracy against the evaluated baselines while retaining linear scaling with the input length, subject to the constant-factor overhead of bidirectional scanning.
SN - 1932-6203
ER -