Hybrid HMM-MGAN approach for long -term renewable energy output scenario generation

Authors

• Fang Liu, Economic and Technological Research Institute of State Grid Sichuan Electric Power Company ， Chengdu 610041 ， China
• Yang Liu, Economic and Technological Research Institute of State Grid Sichuan Electric Power Company ， Chengdu 610041 ， China
• Ruichen Hao, Sichuan Energy Internet Research Institu te， Tsinghua University ， Chengdu 610213 ， China
• Xiang Wang, Sichuan Energy Internet Research Institu te， Tsinghua University ， Chengdu 610213 ， ChinaFollow
• Yuyan Song, Economic and Technological Research Institute of State Grid Sichuan Electric Power Company ， Chengdu 610041 ， China
• Yunche Su, Economic and Technological Research Institute of State Grid Sichuan Electric Power Company ， Chengdu 610041 ， China
• Haibo Li, Sichuan Energy Internet Research Institu te， Tsinghua University ， Chengdu 610213 ， China

Keywords

scenario generation, data reconstruction, multi-scale modeling, modified generative adversarial network, hidden Markov model, volatility constraint

Abstract

To address the limitations of existing numeric al models for long-term renewable energy scenario generation，such as oversimplified structures，poor output quality，a predominant focus on seasonal trends，and insufficient modeling of output details，this paper proposes a long-term renewable power scenario generation method based on a dual-layer weighted ensemble strategy.The proposed approach integrates multiple power reconstruction models to achieve high-fidelity reconstruction of historical renewable energy output.On this basis，a hidden Markov model （HMM ） is employed on a monthly basis to construct daily-scale state transition sequences of renewable energy output，while a modified generative adversarial network （MGAN ） is used to generate monthly renewable energy output curves.The monthly output curves are then concatenated month by month to generate annual renewable energy output scenarios covering 8 760 h.To further enhance the quality of the generated scenarios，the MGAN model is augmented with a volatility constraint and a self-attention mechanism.In addition，an information redundancy mechanism is designed to mitigate boundary effects introduced by convolutional operations.Finally，case studies are conducted using actual historical wind and solar power output data and multi-year meteorological data from a province in southwestern China.The results show that the proposed method offers significant advantages in terms of scenario generation accuracy and the ability to reconstruct output details.

DOI

10.19781/j.issn.1673-9140.2026.04.019

First Page

235

Last Page

246

Recommended Citation

Liu, Fang; Liu, Yang; Hao, Ruichen; Wang, Xiang; Song, Yuyan; Su, Yunche; and Li, Haibo (2026) "Hybrid HMM-MGAN approach for long -term renewable energy output scenario generation," Journal of Electric Power Science and Technology: Vol. 41: Iss. 4, Article 19.
DOI: 10.19781/j.issn.1673-9140.2026.04.019
Available at: https://jepst.researchcommons.org/journal/vol41/iss4/19