基于物理神经网络的输电网运行方式智能生成方法

doi:10.19595/j.cnki.1000-6753.tces.260671

摘要
图/表
参考文献
相关文章 (15)

全文: PDF (1277 KB)
输出: BibTeX | EndNote (RIS)

摘要

摘要随着新能源大规模接入和输电网结构日益复杂,运行在危险区域的待研究样本愈发稀缺,传统输电网运行方式编制需根据负荷反复迭代试凑机组出力,计算极其耗时; 而现有数据驱动生成模型多为无条件盲目抽样,且极易因缺乏物理约束产生大量潮流不收敛的“废数据”。针对上述样本稀缺与生成失真的问题,本文提出一种基于物理约束与条件映射生成对抗网络（PINN-cWGAN）的输电网运行方式智能生成方法。将负荷、机组运行边界与系统节点导纳矩阵作为表征客观运行边界的前置条件,并引入迁移学习联合优化策略,将潮流方程作为软约束嵌入物理信息神经网络（PINN）,引导生成器在物理可行域内探索,不仅做到了有条件的针对性样本生成,还有效克服了高维空间下的模式坍塌难题。以IEEE118节点和欧洲PEGASE1354节点系统为例,在正常运行、新能源波动及极端工况多场景下进行了仿真验证,本文方法生成的大规模样本在覆盖全局多样性的同时,在常规潮流计算软件校验的物理合规率近乎100%,极危工况下的潮流收敛率保持在90%以上,为输电网运行方式样本构建及安全分析提供有效支撑。

	服务

	把本文推荐给朋友
	加入我的书架
	加入引用管理器
	E-mail Alert
	RSS
	作者相关文章
	李滨
	甘明浩
	叶进宇
	陈碧云
	李佩杰

关键词 ：输电网运行方式, 条件映射, 生成对抗网络, 迁移, PINN-cWGAN

Abstract：

With the large-scale integration of renewable energy and the increasing complexity of modern transmission network structures, critical operating samples located in high-risk regions have become increasingly scarce. Traditional scenario construction relies on iterative trial-and-error adjustments of generator outputs based on load variations, resulting in extremely high computational costs and low efficiency. Meanwhile, existing data-driven generative models predominantly adopt unconditional blind sampling. Due to the absence of physical constraints, they are highly susceptible to producing a massive amount of invalid samples with non-convergent power flow solutions.To address these issues of sample scarcity and generation distortion, a targeted generation method for transmission network operating scenarios based on a physics-informed conditional Wasserstein generative adversarial network (PINN-cWGAN) is proposed.
A feature space decoupling framework based on conditional mapping is constructed. The system nodal admittance matrix, load demand, and generator operational boundaries are formulated as prior condition features, while the generator outputs and bus voltages are defined as target features, which breaks the limitations of traditional unconditional blind sampling. Secondly, based on the Wasserstein distance and gradient penalty (WGAN-GP), a conditional generative adversarial architecture is designed to map the high-dimensional statistical distribution from specific objective operational boundaries to secure dispatch responses. Thirdly, to significantly improve the physical validity of the generated samples, AC power flow balance equations and generator output limits are embedded into the physical-informed neural network (PINN) as soft constraints. This mechanism utilizes physical residuals to provide gradient feedback, guiding the generator toward the physically feasible domain. Finally, a two-stage transfer learning strategy based on the global loading margin (LM) is proposed. The model is initially pre-trained on abundant high-margin source domain data, and then its parameters are transferred and fine-tuned on scarce low-margin target domain samples, which improves distribution learning under high-dimensional and small-sample conditions.
Simulation results on the IEEE 118-bus and PEGASE 1354-bus systems demonstrate the effectiveness of the proposed PINN-cWGAN model under multiple operating scenarios, including normal operating conditions, extreme load fluctuations, high renewable energy penetration, and N-k cascading line outages. Compared to traditional Monte Carlo trial-and-error algorithms, the proposed method reduces the generation time of 1000 valid samples from several hours to merely a few seconds. Furthermore, compared with the Generative Adversarial Network (GAN), Gaussian Mixture Model (GMM), Denoising Diffusion Probabilistic Model (DDPM), and Variational Autoencoder (VAE), the proposed method achieves improved distribution fidelity, as indicated by a lower Fréchet Inception Distance (FID). Notably, the physical compliance rate of the generated large-scale samples reaches nearly 100%, and the power flow convergence rate remains above 90% even under critically hazardous conditions (e.g., severe topological mutations), outperforming the compared data-driven methods.
The following conclusions can be drawn from the simulation analysis: (1) The conditional mapping mechanism empowers the deep generative model with conditional response capabilities, allowing for the targeted generation of massive dispatch scenarios under specified load fluctuation ranges and varied topologies. (2) Integrating the AC power flow physical residuals and hard operational boundaries into the PINN loss function effectively alleviates the physical inconsistency commonly observed in purely data-driven models, enhancing the physical consistency and practical value of the generated data. (3) The proposed LM-based transfer learning strategy successfully addresses the natural scarcity of critical low-margin samples, providing an effective data augmentation method for transmission network security assessment and risk analysis.

Key words： Transmission network operating scenarios Conditional mapping Generative adversarial network Transfer learning PINN-cWGAN

收稿日期: 2026-05-06

PACS:

TM732

基金资助:

国家自然科学基金（52267006）和南方电网广西电网公司科技（GXKJXM20240027）资助项目

通讯作者: 甘明浩男,2001年生,硕士研究生,研究方向为电力系统最优化。E-mail：187780556615@163.com

作者简介: 李滨女,1975年生,博士,副教授,研究方向为电力系统最优化。E-mail：lizhen@gxu.edu.cn

引用本文:

李滨, 甘明浩, 叶进宇, 陈碧云, 李佩杰. 基于物理神经网络的输电网运行方式智能生成方法[J]. 电工技术学报, 0, (): 260671-. Li Bin, Gan Minghao, Ye Jinyu, Chen Biyun, Li Peijie. Intelligent Generation Method of Transmission Network Operating Scenarios Based on Physics-Informed Constraints. Transactions of China Electrotechnical Society, 0, (): 260671-.

链接本文:

https://dgjsxb.ces-transaction.com/CN/10.19595/j.cnki.1000-6753.tces.260671 https://dgjsxb.ces-transaction.com/CN/Y0/V/I/260671

[1] 康重庆,姚良忠.高比例可再生能源电力系统的关键科学问题与理论研究框架[J].电力系统自动化,2017,41(09):2-11.
Kang Chongqing, Yao Liangzhong.Key scientific issues and theoretical research framework for power systems with high proportion of renewable energy[J]. Automation of Electric Power Systems, 2017, 41(9): 2-11.
[2] 韩富佳,王晓辉,乔骥,等.基于人工智能技术的新型电力系统负荷预测研究综述[J].中国电机工程学报,2023,43(22):8569-8592.DOI:10.13334/j.0258-8013.pcsee.221560.
Han Fujia, Wang Xiaohui, Qiao Ji, et al.Review on artificial intelligence based load forecasting research for the new-type power system[J]. Proceedings of the CSEE, 2023, 43(22): 8569-8592. DOI: 10.13334/j.0258-8013.pcsee.221560.
[3] 段师琪,余娟,杨知方,等.基于N/N-1潮流内嵌图卷积神经网络的电网运行方式智能调整[J].电工技术学报,2025,40(19):6115-6130.
Duan Shiqi, Yu Juan, Yang Zhifang, et al.Intelligent adjustment method of power grid operation mode based on N/N-1 power flow embedded graph convolutional neural network[J]. Transactions of China Electrotechnical Society, 2025, 40(19): 6115-6130.
[4] Tan B, Zhao J, Chen Y.Scalable risk assessment of rare events in power systems with uncertain wind generation and loads[J]. IEEE Transactions on Power Systems, 2024.
[5] 邵振国,张承圣,陈飞雄,等.生成对抗网络及其在电力系统中的应用综述[J].中国电机工程学报,2023,43(03):987-1004.DOI:10.13334/j.0258-8013.pcsee.212647.
Shao Zhenguo, Zhang Chengsheng, Chen Feixiong, et al.A review on generative adversarial networks for power system applications[J]. Proceedings of the CSEE, 2023, 43(3): 987-1004. DOI: 10.13334/j.0258-8013.pcsee.212647.
[6] 冉晴月,林伟,杨知方,等.基于可信深度神经网络的最优潮流计算方法[J].电工技术学报,2024,39(21):6687-6699.DOI:10.19595/j.cnki.1000-6753.tces.231295.
Ran Qingyue, Lin Wei, Yang Zhifang, et al.Optimal power flow calculation based on a trustworthy deep neural network[J]. Transactions of China Electrotechnical Society, 2024, 39(21): 6687-6699. DOI: 10.19595/j.cnki.1000-6753.tces.231295.
[7] Arjovsky M, Chintala S, Bottou L.Wasserstein generative adversarial networks[C]//International conference on machine learning. PMLR, 2017: 214-223.
[8] 李虹,韩雨萌.计及新能源不确定性的配电网-多微电网协同优化调度[J].电工技术学报,2025,40(17):5571-5588.DOI:10.19595/j.cnki.1000-6753.tces.241227.
Li Hong, Han Yumeng.Coordinated Optimization Scheduling of Distribution Networks and Multiple Microgrids Considering the Uncertainty of New Energy[J]. Transactions of China Electrotechnical Society, 2025, 40(17): 5571-5588. DOI: 10.19595/j.cnki.1000-6753.tces.241227.
[9] Yize C,Yishen W,Daniel K, et al.Model-Free Renewable Scenario Generation Using Generative Adversarial Networks[J].IEEE Transactions on Power Systems,2018,33(3):3265-3275.DOI:10.1109/tpwrs.2018.2794541.
[10] 肖泽文,王怀远.基于物理信息生成对抗网络的暂态稳定评估临界样本增强方法[J/OL].中国电机工程学报,1-12[2026-06-11].https://link.cnki.net/urlid/11.2107.TM.20260304.1410.004.
Xiao Zewen, Wang Huaiyuan.A Critical Sample Enhancement Method for Transient Stability Assessment Based on Physical Information Generative Adversarial Networks[J/OL]. Proceedings of the CSEE, 1-12[2026-06-11].
[11] 廖一帆,武志刚.基于迁移学习与Wasserstein生成对抗网络的静态电压稳定临界样本生成方法[J].电网技术,2021,45(9):3722-3728.
Liao Yifan, Wu Zhigang.Critical Sample Generation Method for Static Voltage Stability Based on Transfer Learning and Wasserstein Generative Adversarial Network[J]. Power System Technology, 2021, 45(9): 3722-3728.
[12] 宋云超,王丹,何伟,等.基于场景构建技术的含多种清洁能源微能源网多目标随机规划研究[J].电力系统保护与控制,2021,49(03):20-31.
Song Yunchao, Wang Dan, He Wei, et al.Research on multi-objective stochastic planning of a micro energy grid with multiple clean energy sources based on scenario construction technology[J]. Power System Protection and Control, 2021, 49(3): 20-31.
[13] 王铮澄,周艳真,郭庆来,等.考虑电力系统拓扑变化的消息传递图神经网络暂态稳定评估[J].中国电机工程学报,2021,41(07):2341-2350.
Wang Zhengcheng, Zhou Yanzhen, Guo Qinglai, et al.Transient stability assessment of power system considering topological change: a message passing neural network-based approach[J]. Proceedings of the CSEE, 2021, 41(7): 2341-2350.
[14] Bompard, Ettore, Di Wu, and Fei Xue. "Structural vulnerability of power systems: A topological approach." Electric Power Systems Research 81.7(2011): 1334-1340.
[15] 郭红霞,陈凌轩,张启,等.电力电量平衡视角下新型电力系统极端场景研究及应对综述[J].电网技术,2024,48(10):3975-3994.DOI:10.13335/j.1000-3673.pst.2023.2268.
Guo Hongxia, Chen Lingxuan, Zhang Qi, et al.Research and Response to Extreme Scenarios in New Power System: A Review From Perspective of Electricity and Power Balance[J]. Power System Technology, 2024, 48(10): 3975-3994. DOI: 10.13335/j.1000-3673.pst.2023.2268.
[16] De Souza, S. R., et al. (2020). Voltage Stability Margin Index Estimation Using a Hybrid Kernel Extreme Learning Machine Approach. Energies, 13(4), 857.
[17] Cui B, Wang Z.Voltage stability assessment based on improved coupled single‐port method[J]. IET Generation, Transmission & Distribution, 2017, 11(10): 2703-2711.
[18] Ajjarapu, V., & Christy, C. (1992). The continuation power flow: a tool for steady state voltage stability analysis. IEEE Transactions on Power Systems, 7(1), 416-423.
[19] 赵凯琳,靳小龙,王元卓.小样本学习研究综述[J].软件学报,2021,32(02):349-369.
Zhao Kailin, Jin Xiaolong, Wang Yuanzhuo.Survey on Few-shot Learning[J]. Journal of Software, 2021, 32(2): 349-369.
[20] Gulrajani I, Ahmed F, Arjovsky M, et al.Improved training of wasserstein gans[J]. Advances in neural information processing systems, 2017, 30.
[21] Raissi, Maziar, Paris Perdikaris, and George E. Karniadakis."Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations." Journal of Computational physics 378 (2019): 686-707.
[22] 向超群,尹雪瑶,伍珣,等.基于物理信息神经网络的牵引变流器直流支撑电容参数辨识方法[J].电工技术学报,2024,39(15):4654-4667.
Xiang Chaoqun, Yin Xueyao, Wu Xun, et al.Parameter Identification of DC-Link Capacitor in Traction Converter Based on Physical Information Neural Network[J]. Transactions of China Electrotechnical Society, 2024, 39(15): 4654-4667.
[23] Zhou T, Droguett E L, Mosleh A.Physics-informed deep learning: A promising technique for system reliability assessment[J]. Applied Soft Computing, 2022, 126: 109217.
[24] 邬永,王冰,陈玉全,等.融合精细化气象因素与物理约束的深度学习模型在短期风电功率预测中的应用[J].电网技术,2024,48(04):1455-1468.
Wu Yong, Wang Bing, Chen Yuquan, et al.Application of Deep Learning Model Integrating Refined Meteorological Factors and Physical Constraints in Short-term Wind Power Prediction[J]. Power System Technology, 2024, 48(4): 1455-1468.
[25] Huang, Bin,Jianhui Wang. "Applications of physics-informed neural networks in power systems-a review." IEEE Transactions on Power Systems 38.1(2022): 572-588.
[26] 王涛,杨远,申冰洁,等.面向运行场景变化的方差引导式域适应暂态稳定评估[J].电工技术学报,2025,40(21):6970-6983.DOI:10.19595/j.cnki.1000-6753.tces.241774.
Wang Tao, Yang Yuan, Shen Bingjie, et al.Variance-Guided Domain-Adaptive Transient Stability Assessment Framework[J]. Transactions of China Electrotechnical Society, 2025, 40(21): 6970-6983. DOI: 10.19595/j.cnki.1000-6753.tces.241774.
[27] 左建辉,石访,宋雪萌,等.主动学习与自训练相结合的电网暂态电压稳定评估方法[J].电力系统自动化,2026,50(2):177-189.
Zuo Jianhui, Shi Fang, Song Xuemeng, et al.Assessment Method for Transient Voltage Stability of Power Grids Based on Combined Active Learning and Self-training[J]. Automation of Electric Power Systems, 2026, 50(2): 177-189.
[28] 马燕峰,骆泽榕,赵书强,等.基于改进蒙特卡洛混合抽样的含风光电力系统风险评估[J].电力系统保护与控制,2022,50(09):75-83.
Ma Yanfeng, Luo Zerong, Zhao Shuqiang, et al.Risk assessment of a power system containing wind power and photovoltaic based on improved Monte Carlo mixed sampling[J]. Power System Protection and Control, 2022, 50(9): 75-83.
[29] 兰健,郭庆来,周艳真,等.基于生成对抗网络和模型迁移的电力系统典型运行方式样本生成[J].中国电机工程学报,2022,42(08):2889-2900.
Lan Jian, Guo Qinglai, Zhou Yanzhen, et al.Generation of Power System Typical Operation Mode Samples: A Generation Adversarial Network and Model-based Transfer Learning Approach[J]. Proceedings of the CSEE, 2022, 42(8): 2889-2900.