|
|
|
| Variance-Guided Domain-Adaptive Transient Stability Assessment Framework |
| Wang Tao1, Yang Yuan1, Shen Bingjie1, Zhang Dan2 |
1. Hebei Key Laboratory of Physics and Energy Technology North China Electric Power University Baoding 071000 China;
2. Yunnan Power Grid Co. Ltd Kunming 650032 China |
|
|
|
Abstract The reliability of transient stability assessment methods based on deep learning typically depends on the assumption that data from the original (source) domain and the new (target) domain follow an independent and identical distribution. However, in renewable energy power systems, this assumption often fails due to the system's complex and dynamic operating conditions. To address this challenge, some studies have attempted to map source and target domain data into a shared latent feature space using maximum mean discrepancy (MMD), which evaluates the mean difference between the two domains. By adjusting model parameters to reduce the MMD value between the source domain and the target domain in the feature space, these methods aim to achieve consistent feature representation and mitigate distributional differences across domains. However, the Gaussian kernel function used in MMD lacks robustness against the long-tail distribution characteristics of transient stability data. Moreover, MMD's reliance on mean difference alone is insufficient to fully capture the volatility and dispersion of transient stability data, limiting its ability to achieve optimal domain alignment. To overcome this limitation, this paper proposes a variance-guided domain-adaptive transient stability assessment (TSA-VGA) framework designed to handle operational scenario changes.
Firstly, to address the issue of the Gaussian kernel function′s insufficient robustness in handling transient stability data, an enhanced kernel function was proposed. This function accounted for both the long-tail distribution characteristics and abrupt variations in transient stability data, optimizing the model's tolerance to extreme values through mathematical refinement, thereby improving its resilience in processing such data. Secondly, to address the insufficient inter-domain distribution alignment in MMD, a variance-guided domain distribution alignment mechanism was introduced. Built upon the improved kernel function, this mechanism constructed a basis set that captures the variance characteristics of distributions and established a new Hilbert space. In this space, the variance-guided domain distribution difference metric was employed to precisely quantify inter-domain distributional discrepancies. By iteratively minimizing these differences, the mechanism achieved refined inter-domain alignment, enhancing the model's adaptability. Furthermore, both biased and unbiased estimators of the variance-guided distribution difference metric were formulated. The mathematical formulations were also derived to establish the error bounds between these two statistical estimators and the true distribution difference, providing theoretical support for cross-domain knowledge transfer in evaluation tasks.
In the case analysis, the effectiveness and accuracy of the TSA-VGA framework were thoroughly validated using the New England 10-machine 39-bus system. To further assess its practical applicability, the framework was also tested on a larger-scale provincial power grid in Southwest China with greater complexity. The results demonstrated that when using the New England 10-machine 39-bus system as the source domain and the provincial power grid as the target domain, the model achieved a prediction accuracy of 97.18% with only 500 target domain samples. This performance effectively meets the requirements for real-time applications in large-scale power systems.
This study presents the following conclusions: (1) An enhanced kernel function is designed to improve the model's tolerance to extreme values, thereby effectively enhancing its robustness to transient stability data exhibiting long-tail characteristics. (2) A variance-guided domain distribution alignment mechanism is proposed. This mechanism constructs a set of basis functions that capture the variance characteristics of the distribution and defines a new Hilbert space. Within this space, the model can precisely quantify inter-domain distribution differences. By iteratively minimizing these differences, refined inter-domain alignment is achieved, thus enhancing the model's adaptability.
|
|
Received: 11 October 2024
|
|
|
|
|
|
[1] 杜东来, 韩松, 荣娜. 基于时空图卷积网络和自注意机制的频率稳定性预测[J]. 电工技术学报, 2024, 39(16): 4985-4995.
Du Donglai, Han Song, Rong Na.Frequency stability prediction method based on modified spatial temporal graph convolutional networks and self-attention[J]. Transactions of China Electrotechnical Society, 2024, 39(16): 4985-4995.
[2] 杨浩, 伍柏臻, 刘铖, 等. 基于暂态关键特征逻辑推理的复杂电网响应驱动暂态稳定性判别[J]. 电工技术学报, 2024, 39(13): 3943-3955.
Yang Hao, Wu Baizhen, Liu Cheng, et al.Response-driven transient stability assessment for complex power grids based on logical reasoning with transient key feature[J]. Transactions of China Electrotechnical Society, 2024, 39(13): 3943-3955.
[3] Zadkhast P, Jatskevich J, Vaahedi E.A multi-decomposition approach for accelerated time-domain simulation of transient stability problems[J]. IEEE Transactions on Power Systems, 2015, 30(5): 2301-2311.
[4] Vu T L, Turitsyn K.Lyapunov functions family approach to transient stability assessment[J]. IEEE Transactions on Power Systems, 2015, 31(2): 1269-1277.
[5] Zhou Xiaomei, Guan Xin, Sun Di, et al.Transient stability assessment based on gated graph neural network with imbalanced data in Internet of energy[J]. IEEE Internet of Things Journal, 2021, 9(12): 9320-9331.
[6] 陆旭, 张理寅, 李更丰, 等. 基于内嵌物理知识卷积神经网络的电力系统暂态稳定评估[J]. 电力系统自动化, 2024, 48(9): 107-119.
Lu Xu, Zhang Liyin, Li Gengfeng, et al.Transient stability assessment of power system based on physics informed convolution neural network[J]. Automation of Electric Power Systems, 2024, 48(9): 107-119.
[7] 甄永赞, 阮程. 基于代价敏感支持向量机和多变量决策树的分级自适应暂态电压稳定评估[J]. 电网技术, 2024, 48(2): 778-791.
Zhen Yongzan, Ruan Cheng.Hierarchical self-adaptation transient voltage stability assessment based on cost-sensitive SVM and multivariate decision tree[J]. Power System Technology, 2024, 48(2): 778-791.
[8] 陈宜尊, 朱容更, 邓晗应, 等. 基于支持向量机的机载自耦变压整流器故障诊断方法[J]. 电气技术, 2024, 25(8): 41-46.
Chen Yizun, Zhu Ronggeng, Deng Hanying, et al.Fault diagnosis method for airborne autotransformer rectifier units based on support vector machine[J]. Electrical Engineering, 2024, 25(8): 41-46.
[9] 叶圣永, 王晓茹, 刘志刚, 等. 基于受扰严重机组特征及机器学习方法的电力系统暂态稳定评估[J]. 中国电机工程学报, 2011, 31(1): 46-51.
Ye Shengyong, Wang Xiaoru, Liu Zhigang, et al.Power system transient stability assessment based on severely disturbed generator attributes and machine learning method[J]. Proceedings of the CSEE, 2011, 31(1): 46-51.
[10] 武宇翔, 韩肖清, 牛哲文, 等. 基于变权重随机森林的暂态稳定评估方法及其可解释性分析[J]. 电力系统自动化, 2023, 47(14): 93-104.
Wu Yuxiang, Han Xiaoqing, Niu Zhewen, et al.Transient stability assessment method based on variable weight random forest and its interpretability analysis[J]. Automation of Electric Power Systems, 2023, 47(14): 93-104.
[11] 任顺鑫, 王怀远, 李剑, 等. 主导模式引导的电力系统暂态稳定数据驱动评估方法[J]. 中国电机工程学报, 2025, 45(14): 5589-5601.
Ren Shunxin, Wang Huaiyuan, Li Jian, et al.Data-driven method for transient stability assessment of power system guided by dominant pattern[J]. Proceedings of the CSEE, 2025, 45(14): 5589-5601.
[12] Su Tong, Liu Youbo, Zhao Junbo, et al.Probabilistic stacked denoising autoencoder for power system transient stability prediction with wind farms[J]. IEEE Transactions on Power Systems, 2020, 36(4): 3786-3789.
[13] Cui Mingjian, Li Fangxing, Cui Hantao, et al.Data-driven joint voltage stability assessment considering load uncertainty: a variational Bayes inference integrated with multi-CNNs[J]. IEEE Transactions on Power Systems, 2022, 37(3): 1904-1915.
[14] 马阳阳, 李永建, 孙鹤, 等. 基于深度置信网络算法的面向铁磁材料旋转磁滞损耗的矢量磁滞模型[J]. 电工技术学报, 2023, 38(15): 4063-4075.
Ma Yangyang, Li Yongjian, Sun He, et al.Vector hysteresis model for rotational hysteresis loss of ferromagnetic materials based on deep belief network algorithm[J]. Transactions of China Electrotechnical Society, 2023, 38(15): 4063-4075.
[15] Chen Qifan, Lin Nan, Bu Siqi, et al.Interpretable time-adaptive transient stability assessment based on dual-stage attention mechanism[J]. IEEE Transactions on Power Systems, 2023, 38(3): 2776-2790.
[16] 宋向金, 孙文举, 刘国海, 等. 深度子领域自适应网络电机滚动轴承跨工况故障诊断[J]. 电工技术学报, 2024, 39(1): 182-193.
Song Xiangjin, Sun Wenju, Liu Guohai, et al.Across working conditions fault diagnosis for motor rolling bearing based on deep subdomain adaption network[J]. Transactions of China Electrotechnical Society, 2024, 39(1): 182-193.
[17] Xie Shushuai, Cheng Wei, Nie Zelin, et al.Multidimensional attention domain adaptive method incorporating degradation prior for machine remaining useful life prediction[J]. IEEE Transactions on Industrial Informatics, 2024, 20(5): 7345-7356.
[18] 王恒泓, 王激尧, 徐炜, 等. 基于领域对抗网络的永磁同步电机初始位置估计[J]. 电工技术学报, 2025, 40(2): 425-438.
Wang Henghong, Wang Jiyao, Xu Wei, et al.Initial position estimation of surface permanent magnet synchronous motor base on domain-adversarial neural networks[J]. Transactions of China Electrotechnical Society, 2025, 40(2): 425-438.
[19] 申锦鹏, 杨军, 李蕊, 等. 基于改进域对抗迁移学习的电力系统暂态稳定自适应评估[J]. 电力系统自动化, 2022, 46(23): 67-75.
Shen Jinpeng, Yang Jun, Li Rui, et al.Self-adaptive transient stability assessment of power system based on improved domain adversarial transfer learning[J]. Automation of Electric Power Systems, 2022, 46(23): 67-75.
[20] Li Yunkai, Meng Qinghao, Wang Yaxin, et al.MASS: a multisource domain adaptation network for cross-subject touch gesture recognition[J]. IEEE Transactions on Industrial Informatics, 2023, 19(3): 3099-3108.
[21] Xu Gaowei, Huang Chenxi, Santos da Silva D, et al. A compressed unsupervised deep domain adaptation model for efficient cross-domain fault diagnosis[J]. IEEE Transactions on Industrial Informatics, 2023, 19(5): 6741-6749.
[22] 李保罗, 孙华东, 张恒旭, 等. 基于两阶段迁移学习的电力系统暂态稳定评估框架[J]. 电力系统自动化, 2022, 46(17): 176-185.
Li Baoluo, Sun Huadong, Zhang Hengxu, et al.Transient stability assessment framework of power system based on two-stage transfer learning[J]. Automation of Electric Power Systems, 2022, 46(17): 176-185.
[23] Li Baoqin, Wu Junyong.Adaptive assessment of power system transient stability based on active transfer learning with deep belief network[J]. IEEE Transactions on Automation Science and Engineering, 2023, 20(2): 1047-1058.
[24] Yu Xiao, Wang Youjie, Liang Zhongting, et al.An adaptive domain adaptation method for rolling bearings' fault diagnosis fusing deep convolution and self-attention networks[J]. IEEE Transactions on Instrumentation and Measurement, 2023, 72: 3509814.
[25] Zhu Lipeng, Wen Weijia, Li Jiayong, et al.Integrated data-driven power system transient stability monitoring and enhancement[J]. IEEE Transactions on Power Systems, 2024, 39(1): 1797-1809.
[26] Hijazi M, Dehghanian P, Wang Shiyuan.Transfer learning for transient stability predictions in modern power systems under enduring topological changes[J]. IEEE Transactions on Automation Science and Engineering, 2024, 21(3): 3274-3288.
[27] Ugli O E M, Lee K H, Lee C H. Automatic optimization of one-dimensional CNN architecture for fault diagnosis of a hydraulic piston pump using genetic algorithm[J]. IEEE Access, 2023, 11: 68462-68472.
[28] Qian Quan, Pu Huayan, Tu Tianjia, et al.Variance discrepancy representation: a vibration characteristic-guided distribution alignment metric for fault transfer diagnosis[J]. Mechanical Systems and Signal Processing, 2024, 217: 111544.
[29] Qian Q, Wang Y, Zhang T, et al.Maximum mean square discrepancy: a new discrepancy representation metric for mechanical fault transfer diagnosis[J]. Knowledge-Based Systems, 2023, 276: 110748.
[30] Li Baoqin, Wu Junyong, Hao Liangliang, et al.Anti-jitter and refined power system transient stability assessment based on long-short term memory network[J]. IEEE Access, 2020, 8: 35231-35244. |
|
|
|
2025, Vol. 40 
