基于堆叠稀疏降噪自编码器的暂态稳定评估模型  

Transient stability assessment model based on stacked sparse denoising auto-encodern

作　　者：温涛 张敏 王怀远 WEN Tao;ZHANG Min;WANG Huaiyuan(Electric Power Dispatching Control Center of Guangdong Power Grid Co.,Ltd.,Guangzhou 510600,China;Guangdong Power Grid Co.,Ltd.Guangzhou Power Supply Bureau,Guangzhou 510620,China;College of Electrical Engineering and Automation,Fuzhou University,Fuzhou 350116,China)

机构地区：[1]广东电网有限责任公司电力调度控制中心,广东广州510600 [2]广东电网有限责任公司广州供电局,广东广州510620 [3]福州大学电气工程与自动化学院,福建福州350116

出　　处：《电力工程技术》2022年第1期207-212,共6页Electric Power Engineering Technology

基　　金：福建省自然科学基金资助项目(2018J01482);福建省科技厅引导性项目(2019H01010204)。

摘　　要：深度学习模型凭借其良好的性能被引入到电力系统的暂态稳定性评估中,但进行在线应用时,须关注模型的抗噪能力和泛化能力。文中提出一种基于堆叠稀疏降噪自编码器(SSDAE)的暂态稳定性评估模型,首先对原始输入数据加入噪声得到受损数据样本,然后对受损数据样本进行高阶特征提取,最后将提取的高阶特征重构成未受损的数据,这一训练过程大大提高了模型的抗噪能力。同时,在对输入特征进行重构的过程中,对隐藏层神经元权重和激活程度进行抑制,实现模型的稀疏化,以此提高模型的泛化能力。仿真结果表明,相对于其他机器学习算法,SSDAE模型具有良好的抗噪能力和泛化能力。With its good performance,the deep learning model is introduced into the transient stability assessment of power system.However,when applying online,the anti-noise ability and generalization ability of the models must be paid attention to.A deep learning model based on stacked sparse denoising auto-encoder(SSDAE)is proposed for real-time transient stability assessment.The corrupted data is generated by adding white Gaussian noise to the original input data.Then,the high-order features are extracted from the corrupted data.Finally,the original data is reconstructed by using the high-order features.The anti-noise ability is greatly improved after training.Besides,the weights and the degree of activation of the hidden layer neurons are suppressed to achieve the sparseness of the model,which can improve the generalization ability of the model.The simulation results show that the SSDAE based model has good anti-noise ability and generalization ability compared to other machine learning algorithms.

关 键 词：深度学习 堆叠稀疏降噪自编码器(SSDAE) 暂态稳定 抗噪声能力 泛化能力 机器学习 

分 类 号：TM744[电气工程—电力系统及自动化]