| 引用本文: | 胡力涛,王怀远,党然,童浩轩,张旸.基于梯度范数的暂态稳定评估模型的不平衡修正方法[J].电力自动化设备,2024,44(4):158-163,177 |
| HU Litao,WANG Huaiyuan,DANG Ran,TONG Haoxuan,ZHANG Yang.Imbalance correction method of transient stability assessment model based on gradient norm[J].Electric Power Automation Equipment,2024,44(4):158-163,177 |
|
| 摘要: |
| 为了解决电力系统中样本数量和质量不平衡造成的暂态稳定评估偏差问题，从评估模型的训练过程出发，通过预训练模型获得样本对模型参数修正的梯度范数，引入梯度范数均值比量化样本的不平衡程度，相较于先验信息，梯度范数均值比综合考虑了样本数量与样本质量的不平衡，并提出基于代价敏感法的不平衡修正方法，利用该方法改善模型的评估倾向性，以实现较好的修正效果。IEEE 39节点系统和华东电网系统的仿真结果验证了所提方法的有效性。 |
| 关键词: 深度学习 暂态稳定评估 代价敏感 梯度范数 堆叠稀疏自编码器 不平衡样本 |
| DOI：10.16081/j.epae.202308022 |
| 分类号:TM712 |
| 基金项目:福建省自然科学基金资助项目(2022J01113) |
|
| Imbalance correction method of transient stability assessment model based on gradient norm |
|
HU Litao1, WANG Huaiyuan1, DANG Ran2, TONG Haoxuan3, ZHANG Yang1
|
|
1.Fujian Key Laboratory of New Energy Generation and Power Conversion, College of Electrical Engineering andAutomation, Fuzhou University, Fuzhou 350116, China;2.Shaanxi Aircraft Industry Limited Liability Company, Hanzhong 723000, China;3.Taining County Power Supply Branch of State Grid Fujian Electric Power Co.,Ltd.,Sanming 354400, China
|
| Abstract: |
| In order to solve the problem of transient stability assessment deviation caused by the imbalance of power system sample quantity and quanlity, starting from the training process of assessment model, the gradient norm of samples to model parameters is obtained by the pre-training model, and the mean ratio of gradient norm is introduced to quantify the imbalance of samples. Compared with the prior information, the mean ratio of gradient norm comprehensively considers the imbalance between the sample quantity and quality. An imbalanced correction method based on the cost-sensitive method is proposed, which is used to improve the assessment preference of the model and realize a preferable correction effect. The simulative results of IEEE 39-bus system and East China Power System verify the effectiveness of the proposed method. |
| Key words: deep learning transient stability assessment cost-sensitive gradient norm stacked sparse auto-encoder imbalanced sample |
