基于深度特征提取与物理可解释回归的新型电力系统节点惯量估计方法

doi:10.19595/j.cnki.1000-6753.tces.260237

摘要
图/表
参考文献
相关文章 (15)

全文: PDF (1580 KB)
输出: BibTeX | EndNote (RIS)

摘要

在高比例可再生能源接入的新型电力系统中,准确估计系统节点惯量对于保障频率稳定具有重要意义。针对现有深度学习方法在特征提取过程中易造成信息丢失、且估计结果缺乏物理可解释性的问题,本文提出了一种基于深度特征提取与物理可解释回归的新型电力系统节点惯量估计方法。首先,采用残差卷积神经网络（ResNet）提取频率变化率序列中的多尺度动态特征,通过残差结构有效保留关键初始信息;然后,引入傅里叶柯尔莫哥洛夫-阿诺德网络（FKAN）作为回归器,以显式函数形式刻画特征与惯量之间的非线性关系,从而提升模型的可解释性。最后,在改进的IEEE 39节点系统及IEEE 118节点系统上开展消融实验、抗噪声评估、对比实验、泛化性验证和可解释性分析。结果表明,所提方法具有较高的估计精度、良好的鲁棒性、泛化能力及可解释性。

	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
	杨苓
	李毅博
	许钊洋
	龚忠亮
	许方园

关键词 ：新型电力系统, 节点惯量估计, 特征提取, 可解释性, 残差卷积网络, 傅里叶柯尔莫哥洛夫-阿诺德网络

Abstract：

Accurate estimation of node-level inertia is essential for maintaining frequency stability in power systems with high penetration of renewable energy sources. The increasing replacement of synchronous generators by power electronic converters leads to reduced system inertia and more pronounced spatial variability, which increases the difficulty of inertia estimation and requires models with both high accuracy and physical interpretability. To address these challenges, a node-level inertia estimation framework combining deep feature extraction and interpretable regression is developed.
The proposed framework consists of a feature extraction module and an interpretable regression module. First, frequency disturbance responses are represented by Rate of Change of Frequency (RoCoF) sequences, which serve as the input to the estimation model. A Residual Convolutional Neural Network (ResNet) is employed to extract multi-scale dynamic features from the RoCoF sequences. Residual connections are introduced to alleviate feature degradation during deep network propagation, thereby preserving critical transient information and improving feature representation capability. Then, the extracted feature vectors are fed into a Fourier Kolmogorov-Arnold Network (FKAN). Compared with conventional multilayer perceptrons, the FKAN constructs explicit nonlinear functional mappings between input features and inertia values. The regression process is expressed through interpretable basis functions while retaining strong nonlinear approximation capability. Finally, the integration of deep feature extraction and interpretable regression enables accurate estimation of node-level inertia while maintaining model transparency.
The effectiveness of the proposed method is verified using the modified IEEE 39-bus system and the IEEE 118-bus system. Ablation studies, noise robustness tests, comparative experiments, and interpretability analyses are conducted on the modified IEEE 39-bus system to evaluate the contribution of each model component and the estimation performance under different operating conditions. The IEEE 118-bus system is used to assess the generalization capability of the trained model in a larger and more complex network.
The results show that the proposed method achieves accurate node-level inertia estimation over a wide range of operating conditions. The residual feature extraction strategy effectively preserves transient dynamic information, while the FKAN establishes explicit nonlinear relationships between features and inertia. Stable estimation performance is maintained under noisy measurement conditions, and consistent results are obtained across different system scales.
The proposed framework combines the representation capability of deep neural networks with the interpretability of explicit functional regression, providing a data-driven approach for node-level inertia estimation in renewable-dominated power systems. The results support its applicability in frequency stability assessment and inertia monitoring.

Key words： novel power system node-level inertia estimation feature extraction interpretability Residual Convolutional Neural Network (ResNet) Fourier Kolmogorov-Arnold Network (FKAN)

收稿日期: 2026-02-08

PACS:

TM712

基金资助:

国家自然科学基金（52107185）、广东省基础与应用基础研究基金（2026A1515010066）和电能高效高质转化全国重点实验室开放基金（2024KF008）资助项目

通讯作者: 许方园男,1984年生,博士,副教授,硕士生导师,研究方向为人工智能在能源系统领域的应用。E-mail：datuan12345@hotmail.com

作者简介: 杨苓女,1992年生,博士,副教授,硕士生导师,研究方向为分布式储能系统的建模与稳定性分析。E-mail：1650148795@qq.com

引用本文:

杨苓, 李毅博, 许钊洋, 龚忠亮, 许方园. 基于深度特征提取与物理可解释回归的新型电力系统节点惯量估计方法[J]. 电工技术学报, 0, (): 260237-. Yang Ling, Li Yibo, Xu Zhaoyang, Gong Zhongliang, Xu Fangyuan. Node-Level Inertia Estimation for Novel Power Systems Based on Deep Feature Extraction and Physically Interpretable Regression. Transactions of China Electrotechnical Society, 0, (): 260237-.

链接本文:

https://dgjsxb.ces-transaction.com/CN/10.19595/j.cnki.1000-6753.tces.260237 https://dgjsxb.ces-transaction.com/CN/Y0/V/I/260237

[1] 赵冬梅,宋晨铭,冯向阳,等.100%新能源场景下考虑频率稳定约束的源网荷储一体化系统储能优化配置[J].电工技术学报,2025,40(07):2146-2161.
Zhao Dongmei, Song Chenming, Feng Xiangyang, et al.The optimal configuration of energy storage in the source-grid-load-storage integrated system considering frequency stability constraints in 100% new energy scenarios[J]. Transactions of China Electrotechnical Society, 2025, 40(07): 2146-2161.
[2] 孙秋野,于潇寒,王靖傲.“双高”配电系统的挑战与应对措施探讨[J].中国电机工程学报,2024,44(18):7115-7136.
Sun Qiuye, Yu Xiaohan, Wang Jingao.Discussion on challenges and countermeasures of “Double High” power distribution system[J]. Proceedings of the CSEE, 2024, 44(18): 7115-7136.
[3] 王翔宇,张庚午,陈武晖,等.模型-数据可信集成的低惯量电力系统频率预测及主动减载方法[J/OL].电工技术学报,1-17[2026-01-27].
Wang Xiangyu, Zhang Gengwu, Chen Wuhui, et al.Frequency prediction and active load shedding of low inertia power system based on confidence-driven model-data integration[J]. Transactions of China Electrotechnical Society, 1-17[2026-01-27].
[4] 李明,熊凌飞,邵冰冰,等.电能路由器储能与并网端口双惯量控制及惯量匹配研究[J/OL].电工技术学报,1-16[2026-05-13].
Li Ming, Xiong Lingfei, Shao BingBing, et al. Research on Dual-Inertia Control and Inertia Matching for Energy Storage and Grid Connection Port in Electric Energy Router[J/OL].Transactions of China Electrotechnical Society, 1-17[2026-05-13].
[5] 刘雷,王广华,梅军,等.计及功率振荡的新能源电力系统节点惯量薄弱区域补偿方法[J].电力系统自动化,2026,50(02):156-166.
Liu Lei, Wang Guanghua, Mei Jun, et al.Compensation Method for Weak-inertia Areas of Bus in Power System with Renewable EnergyConsidering Power Oscillations[J]. Automation of Electric Power Systems, 2026, 50(02): 156-166.
[6] 李卫东,丁奇力.大扰动下电力系统惯性时空分布特性的表征形式初探[J].中国电机工程学报,2024,44(12):4601-4615.
Li Weidong, Ding Qili.A preliminary research on the representation form of spatial-temporal distribution characteristics of power system inertia under large disturbances[J]. Proceedings of the CSEE, 2024, 44(12): 4601-4615.
[7] 鲁宗相,姜继恒,乔颖,等.新型电力系统广义惯量分析与优化研究综述[J].中国电机工程学报,2023,43(05):1754-1776.
Lu Zongxiang, Jiang Jiheng, Qiao Ying, et al.A review on generalized inertia analysis and optimization of new power systems[J]. Proceedings of the CSEE, 2023, 43(05): 1754-1776.
[8] 刘方蕾,胥国毅,王凡,等.基于差值计算法的系统分区惯量评估方法[J].电力系统自动化,2020,44(20):46-53.
Liu Fanglei, Xu Guoyi, Wang Fan, et al.Assessment method of system partition inertia based on differential calculation method[J]. Automation of Electric Power Systems, 2020, 44(20): 46-53.
[9] 马燕峰,李金媛,王子建,等.基于量测数据的新能源电力系统区域等效惯量评估方法[J].电工技术学报,2024,39(17):5406-5421.
Ma Yangfeng, Li Jinyuan, Wang Zijian, et al.Assessment method of regional equivalent inertia of new energy power system based on measured data[J]. Transactions of China Electrotechnical Society, 2024, 39(17): 5406-5421.
[10] Li Hongyu, You Shutang, Jiang Zhihao, et al.Ambient-frequency-data based system-level inertia estimation using physical equation and its practice on hawaii islands[J]. IEEE Transactions on Power Systems, 2024, 39(06): 6948-6959.
[11] Liu Muyang, Chen Junrn, Milano Federico, et al.On-line inertia estimation for synchronous and non-synchronous devices[J]. IEEE Transactions on Power Systems, 2021, 36(03): 2693-2701.
[12] 曹永吉,李常刚.基于可变数据窗口的电力系统惯量估计方法[J].上海交通大学学报,2025,1-19.
Cao Yongji, Li Changgang.Estimation approach for inertia of power system based on variable data window[J]. Journal of Shanghai Jiaotong University, 2025, 1-19.
[13] Schiffer J, Aristidou P and Ortega R. Online estimation of power system inertia using dynamic regressor extension and mixing[J]. IEEE Transactions on Power Systems, 2019, 34(06): 4993-5001.
[14] 李元臣,文云峰,叶希,等.基于多新息辨识的电力系统节点惯量估计方法[J].电力自动化设备,2022,42(08):89-95.
Li Yuanchen, Wen Yunfeng, Ye Xi, et al.Estimation method of power system nodal inertia based on multi-innovation identification[J]. Electric Power Automation Equipment, 2022, 42(08): 89-95.
[15] 曹斌,原帅,刘家豪,等.考虑频率动态响应实时分区的电力系统惯量在线评估方法[J].电网技术,2023,47(03):930-941.
Cao Bin, Yuan Shuai, Liu Jiahao, et al.Power system online inertia estimation considering real-time clustering of frequency dynamic response[J]. Power System Technology, 2023, 47(03): 930-941.
[16] Phurailatpam C, Rather Z H, Bahrani B, et al.Estimation of non-synchronous inertia in AC microgrids[J]. IEEE Transactions on Sustainable Energy, 2021, 12(04): 1903-1914.
[17] Liu Kai, Xu Yijun, Gu Wei, et al.A bayesian approach for online inertia estimation of synchronous and nonsynchronous generators[J]. IEEE Transactions on Instrumentation and Measurement, 2024, 73: 1-12.
[18] 黄海东,徐云清,张琦兵,等.基于贝叶斯推理的电力系统转动惯量和新能源虚拟惯量估计方法[J].可再生能源,2024,42(11):1546-1553.
Huang Haidong, Xu Yunqing, Zhang Qibing, et al.Estimation method of rotational inertia of power system and virtual inertia of new energy based on Bayesian inference[J]. Renewable Energy Resources, 2024, 42(11): 1546-1553.
[19] Linaro D, Bizzarri F, Giudice D, et al.Continuous estimation of power system inertia using convolutional neural networks[J]. Nature Communications, 2023, 14(01): 4440-4453.
[20] Tuo Mingjian and Li Xingpeng. Machine learning assisted inertia estimation using ambient measurements[J]. IEEE Transactions on Industry Applications, 2023, 59(04): 4893-4903.
[21] Lecun Y, Bengio Y and Hinton G. Deep learning[J]. Nature, 2015, 521(7553): 436-444.
[22] Fan Feng-Lei, Xiong Jinjun and Li Mengzhou. On interpretability of artificial neural networks: A survey[J]. IEEE Transactions on Radiation and Plasma Medical Sciences, 2021, 5(06): 741-760.
[23] Liu Ziming, Wang Yixuan, Vaidya S, et al.Kan: Kolmogorov-arnold networks[C]. 2025 International Conference on Learning Representations (ICLR), Expo Drive, Singapore, 2025.
[24] Cai Yaojie, Wang Xiaozhe, Joós G, et al.An online data-driven method to locate forced oscillation sources from power plants based on sparse identification of nonlinear dynamics (SINDy)[J]. IEEE Transactions on Power Systems, 2023, 38(03): 2085-2099.
[25] Poudyal A, Fourney R, Tonkoski R, et al.Convolutional neural network-based inertia estimation using local frequency measurements[C]. 2020 52nd North American Power Symposium (NAPS), Tempe, AZ, USA, 2021.
[26] Farrar D E and Glauber R R. Multicollinearity in regression analysis: The problem revisited[J]. The Review of Economics and Statistics, 1967, 49(01): 92-107.