为保障高比例清洁能源接入场景下电网的安全稳定运行,提出一种基于高频增强扩散模型的短期净负荷概率预测方法。首先,基于极限梯度提升（XGBoost）算法和最大信息系数（MIC）提取关键气象特征,形成多元输入条件特征集。其次,构建基于交叉注意力机制的条件扩散模型架构,并针对编码-解码环节,分别设计融合小波注意力的CNN-LSTM网络和多通道卷积神经网络,以增强对高频特征的捕捉能力。然后,基于初始预测值和核密度估计方法,利用改进的损失函数对预训练模型进行微调,以进一步提高生成区间的质量。最后,基于华东某地区电网实际负荷数据开展算例分析,并与多种模型进行对比,结果表明所提方法具有更高的预测精度。
Abstract
In order to ensure the safe and stable operation of the power grid under the scenario of high proportion of clean energy access, a short-term net load probability prediction method based on high frequency enhanced diffusion model is proposed. Firstly, key meteorological features are extracted based on the eXtreme Gradient Boosting（XGBoost） algorithm and maximum information coefficient（MIC）to form a multivariate input condition feature set. Secondly, a conditional diffusion model architecture based on cross-attention mechanism is constructed. For the encoding-decoding link, a CNN-LSTM network incorporating wavelet attention and a multi-channel convolutional neural network are designed respectively to enhance the ability to capture high-frequency features. Then, based on the initial predicted values and kernel density estimation methods, the pre-trained model is fine-tuned using an improved loss function to improve the quality of the generated intervals further. Finally, an example analysis is carried out based on the actual load data of a power grid in East China, and the results show that the proposed method has a higher prediction accuracy when compared with various models.
关键词
配电网 /
负荷预测 /
特征提取 /
条件扩散模型 /
小波注意力 /
模型微调
Key words
distribution networks /
load forecasting /
feature extraction /
conditional diffusion model /
wavelet attention /
model fine-tune
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] FAUSTINE A, NUNES N J, PEREIRA L.Efficiency through simplicity: MLP-based approach for net-load forecasting with uncertainty estimates in low-voltage distribution networks[J]. IEEE transactions on power systems, 2025, 40(1): 46-56.
[2] 黎静华, 骆怡辰, 杨舒惠, 等. 可再生能源电力不确定性预测方法综述[J]. 高电压技术, 2021, 47(4): 1144-1157.
LI J H, LUO Y C, YANG S H, et al.Review of uncertainty forecasting methods for renewable energy power[J]. High voltage engineering, 2021, 47(4): 1144-1157.
[3] 杨茂, 杨春霖, 董骏城. 基于预测误差分布优化模型的风电功率超短期概率区间预测研究[J]. 太阳能学报, 2019, 40(10): 2967-2978.
YANG M, YANG C L, DONG J C.Ultra-short term probabilistic intervals forecasting of wind power based on optimization model of forecasting error distribution[J]. Acta energiae solaris sinica, 2019, 40(10): 2967-2978.
[4] 赵长飞. 新能源电力系统概率预测-决策一体化研究[D]. 杭州: 浙江大学, 2022.
ZHAO C F.Integrated probabilistic forecasting and decision for power system with renewables[D]. Hangzhou: Zhejiang University, 2022.
[5] 杨国清, 李建基, 王德意, 等. 基于信息熵变权区间组合和边界逼近的短期光伏功率区间预测[J]. 太阳能学报, 2023, 44(2): 381-390.
YANG G Q, LI J J, WANG D Y, et al.Short-term photovoltaic power interval prediction based on information entropy variable weight interval combination and boundary approximation[J]. Acta energiae solaris sinica, 2023, 44(2): 381-390.
[6] 李丹, 张远航, 杨保华, 等. 基于约束并行LSTM分位数回归的短期电力负荷概率预测方法[J]. 电网技术, 2021, 45(4): 1356-1364.
LI D, ZHANG Y H, YANG B H, et al.Short time power load probabilistic forecasting based on constrained parallel-LSTM neural network quantile regression mode[J]. Power system technology, 2021, 45(4): 1356-1364.
[7] XIA W J, HUANG H Y, DUQUE E M S, et al. Comparative assessment of generative models for transformer-and consumer-level load profiles generation[J]. Sustainable energy, grids and networks, 2024, 38: 101338.
[8] SUN M Y, ZHANG T Q, WANG Y, et al.Using Bayesian deep learning to capture uncertainty for residential net load forecasting[J]. IEEE transactions on power systems, 2020, 35(1): 188-201.
[9] GAL Y, GHAHRAMANI Z.Dropout as a Bayesian approximation: representing model uncertainty in deep learning[C]//Proceedings of the 33rd International Conference on International Conference on Machine Learning-Volume 48. New York, NY, USA, 2016: 1050-1059.
[10] 孙浩, 万灿, 曹照静, 等. 基于条件生成对抗网络曲线生成的短期负荷概率预测[J]. 电力系统自动化, 2023, 47(23): 189-199.
SUN H, WAN C, CAO Z J, et al.Short-term load probabilistic forecasting based on conditional generative adversarial network curve generation[J]. Automation of electric power systems, 2023, 47(23): 189-199.
[11] DHARIWAL P, NICHOL A.Diffusion models beat gans on image synthesis[J]. Advances in neural information processing systems, 2021, 34: 8780-8794.
[12] LIANG R, DENG Y, WANG D.Probabilistic building load forecasting via conditional diffusion model[C]//Proceedings of the 15th ACM International Conference on Future and Sustainable Energy Systems. Singapore, 2024: 490-491.
[13] 刘锦翔, 张江丰, 董山玲, 等. 基于条件增强扩散模型的短期负荷概率预测[J]. 电力系统自动化, 2024, 48(23): 197-207.
LIU J X, ZHANG J F, DONG S L, et al.Short-term load probabilistic forecasting based on conditional enhanced diffusion model[J]. Automation of electric power systems, 2024, 48(23): 197-207.
[14] 李津, 史加荣, 张琰妮, 等. 基于最大信息系数的短期太阳辐射协同估计[J]. 太阳能学报, 2023, 44(9): 286-294.
LI J, SHI J R, ZHANG Y N, et al.Short-term solar radiation synergy estimation based on maximum information coefficient[J]. Acta energiae solaris sinica, 2023, 44(9): 286-294.
[15] YANG X Y, ZHOU D Q, FENG J S, et al.Diffusion probabilistic model made slim[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Vancouver, BC, Canada, 2023: 22552-22562.
[16] YANG Y Y, JIN M, WEN H M, et al. A survey on diffusion models for time series and spatio-temporal data[EB/OL].2024: arXiv: 2404.18886. https://arxiv.org/abs/2404.18886.
[17] LI S Y, XIONG H, CHEN Y Z.DiffPLF: a conditional diffusion model for probabilistic forecasting of EV charging load[J]. Electric power systems research, 2024, 235: 110723.
[18] 王东风, 刘婧, 黄宇, 等. 结合太阳辐射量计算与CNN-LSTM组合的光伏功率预测方法研究[J]. 太阳能学报, 2024, 45(2): 443-450.
WANG D F, LIU J, HUANG Y, et al.Photovoltaic power prediction method combinating solar radiation calculation and CNN-LSTM[J]. Acta energiae solaris sinica, 2024, 45(2): 443-450.
[19] TRIPURA T, CHAKRABORTY S.Wavelet neural operator for solving parametric partial differential equations in computational mechanics problems[J]. Computer methods in applied mechanics and engineering, 2023, 404: 115783.
[20] SZEGEDY C, LIU W, JIA Y Q, et al.Going deeper with convolutions[C]//2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Boston, MA, USA, 2015: 1-9.
[21] Solcast. Solar API and solar weather forecasting tool[EB/OL].[2024-11-20]. https://solcast.com.au.
[22] WANG Z X, WEN Q S, ZHANG C L, et al.DiffLoad: uncertainty quantification in electrical load forecasting with the diffusion model[J]. IEEE transactions on power systems, 2025, 40(2): 1777-1789.
[23] TRAN D, DUSENBERRY M, VAN DER WILK M, et al. Bayesian layers: a module for neural network uncertainty[C]//Proceedings of the 33rd International Conference on Neural Information Processing Systems. Vancouver BC, Canada, 2019: 14660-14672.
[24] LAKSHMINARAYANAN B, PRITZEL A, BLUNDELL C.Simple and scalable predictive uncertainty estimation using deep ensembles[C]//Neural Information Processing Systems. 2016.
[25] TAO X F, LU Y, YANG X L.Short-term power load probability density forecasting based on a double-layer LSTM-attention quantile regression[C]//2021 IEEE 21st International Conference on Communication Technology (ICCT). Tianjin, China, 2021: 1046-1051.
基金
江苏省碳达峰碳中和科技创新专项（产业前瞻与关键核心技术攻关）（BE2023093-2）; 江苏省研究生科研与实践创新计划（SJCX23_0131）
PDF(2602 KB)
