MTMC: Prediction of Solar Irradiance based on Multi-scale Attention Mamba and Channel Clustering Optimization

Haoming Cheng; Zhenglei Wang

doi:10.6919/ICJE.202601_12(1).0002

Authors

Haoming Cheng
Zhenglei Wang

DOI:

https://doi.org/10.6919/ICJE.202601_12(1).0002

Keywords:

Global Horizontal Irradiance (GHI); Multi-Scale Attention; TE/TF-Mamba; Channel Cluster Module.

Abstract

Accurate prediction of total horizontal solar irradiance (GHI) is vital for enhancing photovoltaic efficiency and ensuring grid stability. This study proposes a hybrid framework that integrates physical insights with deep learning. To improve the accuracy of GHI time series forecasting, this study first applies seasonal trend decomposition (STL) to divide the original series into trend, seasonal, and residual components. Considering the unique characteristics of each part, corresponding processing strategies are adopted. For local anomalies, daily variations, and seasonal patterns, a multi-scale attention module (MSAM) is developed, which integrates variable-scale convolution with dynamic QKV mechanisms to extract key features at different temporal scales. To capture long-term trends while suppressing local noise, the framework employs a TE-Mamba state space model combined with exponential smoothing. In addition, to address dynamic seasonal effects and high-frequency disturbances, a TF-Mamba module is designed, using sparse gating to adaptively identify and model periodic behaviors. Furthermore, a frequency-domain channel clustering approach based on Mahalanobis distance is introduced to remove redundant meteorological inputs, ensuring that the retained features are closely related to GHI prediction. Experiments on GHI datasets from Beijing and Xining show that the framework improves prediction accuracy by over 40% compared to traditional methods, demonstrating a physically grounded yet data-driven solution for complex weather time series forecasting.

Downloads

Download data is not yet available.

References

[1] Kumari, Pratima, and Durga Toshniwal. "Long short term memory–convolutional neural network based deep hybrid approach for solar irradiance forecasting." Applied Energy 295 (2021): 117061.

[2] Didugu, Ganesh, et al. "VWFTS-PSO: a novel method for time series forecasting using variational weighted fuzzy time series and particle swarm optimization." International Journal of General Systems (2024): 1-20.

[3] MAO Y H,SUN C C,XU L Y,et al. "A survey of time series forecasting methods based on deep learning[J]. " Microelectronics & Computer,2023,40(4): 8-17. DOI: 10.19304/J.ISSN1000-7180.2022.0

[4] Cao Danyang, Ma Jinfeng "Research on Time Series Prediction Algorithm Based on STL and EMD[J]. Electronic Components and Information Technology" 2021,5 (08): 76-78. DOI: 10.19772/j.cnki. 2096-4455.2021.8.033

[5] ZHAO Jiandong, ZHU Dan, LIU Jiaxin. "Metro Transfer Passenger Flow Prediction Based on STL-GRU[J] " Journal of South China University of Technology(Natural Science Edition), 2022, 50(5): 22-31.

[6] Jiao, Feng, et al. "An improved STL-LSTM model for daily bus passenger flow prediction during the COVID-19 pandemic." Sensors 21.17 (2021): 5950.

[7] Cai Yi, ZHANG Wei. "Multi-Load Forecasting of Integrated Energy Systems Based on STL-Crossformer[J]. " Journal of Northeast Electric Power University, 2024,44(1):34-41.

[8] Gu, Albert, and Tri Dao. "Mamba: Linear-time sequence modeling with selective state spaces." arXiv preprint arXiv:2312.00752 (2023).

[9] Xie, Xinyu, et al. "Fusionmamba: Dynamic feature enhancement for multimodal image fusion with mamba." Visual Intelligence 2.1 (2024): 37.

[10] Han, Dongchen, et al. "Demystify mamba in vision: A linear attention perspective." arXiv preprint arXiv:2405.16605 (2024).

[11] Cao, Bing, et al. "Predictive dynamic fusion." arXiv preprint arXiv:2406.04802 (2024).

[12] Qiu, Xiangfei, et al. "Duet: Dual clustering enhanced multivariate time series forecasting." arXiv preprint arXiv:2412.10859 (2024).

[13] Cleveland, Robert B., et al. "STL: A seasonal-trend decomposition." J. off. Stat 6.1 (1990): 3-73.

[14] Woo, Gerald, et al. "Etsformer: Exponential smoothing transformers for time-series forecasting." arXiv preprint arXiv:2202.01381 (2022).

[15] Hyndman, Rob, et al. Forecasting with exponential smoothing: the state space approach. Springer Science & Business Media, 2008.

[16] Wang, Xiao, et al. "State space model for new-generation network alternative to transformers: A survey." arXiv preprint arXiv:2404.09516 (2024).

[17] Wang, Peihao, et al. "Understanding and Mitigating Bottlenecks of State Space Models through the Lens of Recency and Over-smoothing." arXiv preprint arXiv:2501.00658 (2024).

[18] Zhou, Tian, et al. "Fedformer: Frequency enhanced decomposed transformer for long-term series forecasting." International conference on machine learning. PMLR, 2022.

[19] Wu, Haixu, et al. "Autoformer: Decomposition transformers with auto-correlation for long-term series forecasting." Advances in neural information processing systems 34 (2021): 22419-22430.

[20] Wu, Yuhan, et al. "Effective LSTMs with seasonal-trend decomposition and adaptive learning and niching-based backtracking search algorithm for time series forecasting." Expert Systems with Applications 236 (2024): 121202.

[21] Lai, Zhichen, et al. "E2usd: Efficient-yet-effective unsupervised state detection for multivariate time series." Proceedings of the ACM Web Conference 2024. 2024.

[22] Yi, Kun, et al. "Frequency-domain mlps are more effective learners in time series forecasting." Advances in Neural Information Processing Systems 36 (2023): 76656-76679.

[23] Wang, Yan, et al. "Multi-scale attention network for single image super-resolution." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2024.

[24] Ouyang, Daliang, et al. "Efficient multi-scale attention module with cross-spatial learning." ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2073.

[25] Boussif, Oussama, et al. "Improving day-ahead solar irradiance time series forecasting by leveraging spatio-temporal context." Advances in Neural Information Processing Systems 36 (2023): 2342-2367.

[26] Cao, J., Xu, T. T., Deng, L. H., et al. "Hemispheric prediction of solar cycles 25 and 26 from multivariate sunspot time-series data via Informer models. " Astronomical Techniques and Instruments, 2025,2(1): 16−26.

[27] Guo, Yan, et al. "LSTM time series NDVI prediction method incorporating climate elements: A case study of Yellow River Basin, China." Journal of Hydrology 629 (2024): 130518.

[28] Fu, En, and Yanyan Hu. "Frequency-Masked Embedding Inference: A Non-Contrastive Approach for Time Series Representation Learning." arXiv preprint arXiv:2412.20790 (2024).

[29] Yi, Kun, et al. "FourierGNN: Rethinking multivariate time series forecasting from a pure graph perspective." Advances in neural information processing systems 36 (2023): 69638-69660.

[30] Wang Jing, He Jianjun, Yi Shanxin, etc. "Remote sensing image dehazing method based on CSC Mamba model[J/OL]." Geophysical and geochemical exploration calculation technology: 1-12[2020-03-28]

[31] Zhao Yilin, Liu Wenfeng, Li Zheng, etc. "Analysis of the Evolution Law and Influencing Factors of Natural Runoff Based on STL Time Series Decomposition[J]. " Journal of Water Resources, 2025, 56 (02): 216-226+239.

[32] Wu Tiantian, Li Yankai, Liu Yang. "Multi stage Rain Removal Algorithm Based on Multi scale Frequency Attention[J]." Computer and Modernization, 2024, (02):50-55.

[33] Ahmad, Muhammad, et al. "Hybrid State-Space and GRU-based Graph Tokenization Mamba for Hyperspectral Image Classification." arXiv preprint arXiv:2502.06427 (2025).

[34] Theodosiou, Marina. "Forecasting monthly and quarterly time series using STL decomposition." International Journal of Forecasting 27.4 (2011): 1178-1195.

[35] Parnichkun, Rom N., et al. "State-free inference of state-space models: The transfer function approach." arXiv preprint arXiv:2405.06147 (2024).