Missing-data-tolerant Diffusion-based Wind Power Scenario Forecasting Method

doi:10.16182/j.issn1004731x.joss.26-0213

Abstract

Abstract:

To address the issue of error accumulation in traditional "imputation-then-forecasting" approaches, a missing data tolerant diffusion framework (MDTDF) is proposed. An XGBoost regression model is employed to map numerical weather prediction data into deterministic power forecasts. The encoder in the denoising network extracts temporal features, which are fused with the deterministic forecasts and fed into the decoder through a cross-attention mechanism to guide the denoising process. A historical constraint mechanism is introduced to directly utilize incomplete historical data and dynamically correct the denoising result at each step through sample gradient updates and noise injection guided by historical information. The simulation results show that the proposed MDTDF achieves significantly better forecasting performance than existing state-of-the-art methods when data are missing and demonstrates good stability and robustness across different missing rates.

Key words: scenario forecasting, conditional diffusion model, missing data, adaptive, wind power output

CLC Number:

TP391.1

Shi Yingying, Dong Xiaochong, Fu Guobin, Ma Miaomiao, Li Yanhe, Wang Xuebin. Missing-data-tolerant Diffusion-based Wind Power Scenario Forecasting Method[J]. Journal of System Simulation, 2026, 38(8): 2167-2178.

Figures/Tables 15

Fig. 1

Fig. 2

Fig. 3

Fig. 4

Table 1

Hyperparameter settings for MDTDF

定义	取值
训练轮次E	500
批大小B	128
学习率r_l	10^–3
噪声强度参数 $β 1, β K$	(0, 0.999)
网络层{L₁, L₂}	{3, 3}
场景数R	200
扩散步K	100

Table 1

Fig. 5

Fig. 6

Fig. 7

Fig. 8

Fig. 9

Table 2

Table 3

Fig. 10

Table 4

Fig. 11

References 21

[1]	Global Wind Energy Council. Global Wind Report 2024[EB/OL]. [2025-08-12]. .
[2]	焦业峰, 王艳, 纪志成. 基于BLP-ALO-SVM的风电功率短期预测方法[J]. 系统仿真学报, 2022, 34(12): 2535-2545.
	Jiao Yefeng, Wang Yan, Ji Zhicheng. Short-term Prediction Method of Wind Power Based on BLP-ALO-SVM[J]. Journal of System Simulation, 2022, 34(12): 2535-2545.
[3]	唐新姿, 顾能伟, 黄轩晴, 等. 风电功率短期预测技术研究进展[J]. 机械工程学报, 2022, 58(12): 213-236.
	Tang Xinzi, Gu Nengwei, Huang Xuanqing, et al. Progress on Short Term Wind Power Forecasting Technology[J]. Journal of Mechanical Engineering, 2022, 58(12): 213-236.
[4]	董骁翀, 张姝, 李烨, 等. 电力系统中时序场景生成和约简方法研究综述[J]. 电网技术, 2023, 47(2): 709-721.
	Dong Xiaochong, Zhang Shu, Li Ye, et al. Review of Power System Temporal Scenario Generation and Reduction Methods[J]. Power System Technology, 2023, 47(2): 709-721.
[5]	Tu Qingyu, Miao Shihong, Yao Fuxing, et al. Forecasting Scenario Generation for Multiple Wind Farms Considering Time-series Characteristics and Spatial-temporal Correlation[J]. Journal of Modern Power Systems and Clean Energy, 2021, 9(4): 837-848.
[6]	Wen Honglin, Ma Jinghuan, Gu Jie, et al. Sparse Variational Gaussian Process Based Day-ahead Probabilistic Wind Power Forecasting[J]. IEEE Transactions on Sustainable Energy, 2022, 13(2): 957-970.
[7]	杨辉明, 雷勇. 基于改进马尔科夫链的风电日前不确定性建模方法[J]. 南方电网技术, 2021, 15(7): 54-60.
	Yang Huiming, Lei Yong. Day-ahead Uncertainty Simulation Method of Wind Power Based on Improved Markov Chain[J]. Southern Power System Technology, 2021, 15(7): 54-60.
[8]	陈静, 宋明哲, 袁康, 等. 基于Copula函数的风电场出力场景生成方法[J]. 电工技术, 2025(8): 39-42.
	Chen Jing, Song Mingzhe, Yuan Kang, et al. Wind Power Output Scenario Generation Based on Copula Functions[J]. Electric Engineering, 2025(8): 39-42.
[9]	董骁翀, 孙英云, 蒲天骄. 基于条件生成对抗网络的可再生能源日前场景生成方法[J]. 中国电机工程学报, 2020, 40(17): 5527-5535.
	Dong Xiaochong, Sun Yingyun, Pu Tianjiao. Day-ahead Scenario Generation of Renewable Energy Based on Conditional GAN[J]. Proceedings of the CSEE, 2020, 40(17): 5527-5535.
[10]	黄越辉, 孙亚南, 李驰, 等. 基于条件生成对抗网络的多区域风电短期出力场景生成方法[J]. 电网技术, 2023, 47(1): 63-72.
	Huang Yuehui, Sun Yanan, Li Chi, et al. Constructing Method of Short-term Output Scenarios for Multi-regional Wind Power Based on Conditional Generative Adversarial Network[J]. Power System Technology, 2023, 47(1): 63-72.
[11]	Zheng Zhong, Wang Long, Yang Luoxiao, et al. Generative Probabilistic Wind Speed Forecasting: A Variational Recurrent Autoencoder Based Method[J]. IEEE Transactions on Power Systems, 2022, 37(2): 1386-1398.
[12]	Dumas Jonathan, Wehenkel Antoine, Lanaspeze Damien, et al. A Deep Generative Model for Probabilistic Energy Forecasting in Power Systems: Normalizing Flows[J]. Applied Energy, 2022, 305: 117871.
[13]	张志洪, 胡旭光, 吴恩凯, 等. 基于自适应多任务扩散模型的风光荷场景生成方法[J]. 综合智慧能源, 2025, 47(11): 14-23.
	Zhang Zhihong, Hu Xuguang, Wu Enkai, et al. Wind-solar-load Scenario Generation Method Based on Adaptive Multi-task Diffusion Model[J]. Integrated Intelligent Energy, 2025, 47(11): 14-23.
[14]	Saxena Divya, Cao Jiannong. Generative Adversarial Networks (GANs): Challenges, Solutions, and Future Directions[J]. ACM Computing Surveys, 2021, 54(3): 63.
[15]	Dong Xiaochong, Mao Zhihang, Sun Yingyun, et al. Short-term Wind Power Scenario Generation Based on Conditional Latent Diffusion Models[J]. IEEE Transactions on Sustainable Energy, 2024, 15(2): 1074-1085.
[16]	王长刚, 刘伟, 曹宇, 等. 基于改进条件生成扩散模型的新能源日前场景生成方法[J]. 电网技术, 2025, 49(4): 1358-1368.
	Wang Changgang, Liu Wei, Cao Yu, et al. The Day-ahead Scenario Generation Method for New Energy Based on an Improved Conditional Generative Diffusion Model[J]. Power System Technology, 2025, 49(4): 1358-1368.
[17]	Sareen Karan, Bijaya Ketan Panigrahi, Shikhola Tushar, et al. A Robust De-Noising Autoencoder Imputation and VMD Algorithm Based Deep Learning Technique for Short-term Wind Speed Prediction Ensuring Cyber Resilience[J]. Energy, 2023, 283: 129080.
[18]	Xu Xuefang, Hu Shiting, Shao Huaishuang, et al. A Spatio-temporal Forecasting Model Using Optimally Weighted Graph Convolutional Network and Gated Recurrent Unit for Wind Speed of Different Sites Distributed in an Offshore Wind Farm[J]. Energy, 2023, 284: 128565.
[19]	Li Qiaoqiao, Xu Yan, Benjamin Si Hao Chew, et al. An Integrated Missing-data Tolerant Model for Probabilistic PV Power Generation Forecasting[J]. IEEE Transactions on Power Systems, 2022, 37(6): 4447-4459.
[20]	Meng Zichao, Guo Ye, Zhao Chenhao. Probabilistic Wind Power Forecasting with Missing Data Tolerance: An End-to-end Nonparametric Approach[J]. IEEE Transactions on Sustainable Energy, 2026, 17(2): 1202-1213.
[21]	Ho J, Jain A, Abbeel P. Denoising Diffusion Probabilistic Models[C]//Proceedings of the 34th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2020: 6840-6851.

季节	MAE	RMSE	PS	CRPS
平均	0.027 6	0.046 4	0.010 9	0.020 9
春季	0.024 2	0.042 4	0.009 5	0.018 3
夏季	0.029 3	0.044 7	0.011 5	0.022 0
秋季	0.026 0	0.045 3	0.010 3	0.019 8
冬季	0.031 0	0.052 7	0.012 3	0.023 5

缺失率/%	MAE	RMSE	PS	CRPS
0	0.027 5	0.046 3	0.010 9	0.020 8
5	0.027 6	0.046 4	0.010 9	0.020 9
15	0.027 8	0.046 7	0.011 0	0.021 0
25	0.028 0	0.046 9	0.011 0	0.021 2
35	0.028 3	0.047 3	0.011 2	0.021 4
45	0.028 5	0.047 5	0.011 2	0.021 5

模型	MAE	RMSE	PS	CRPS
MDTDF	0.015 4	0.027 6	0.005 9	0.011 3
LI+VAE	0.021 5	0.038 4	0.009 0	0.017 3
LI+WGAN-GP	0.053 0	0.084 4	0.019 8	0.038 0
文献[20]+DDPM	0.016 9	0.028 8	0.006 5	0.012 5