基于TCN-CCRELMS的电力系统暂态稳定评估

Transient Stability Assessment of Power System Based on TCN-CCRELMS

为进一步提高电力系统暂态稳定评估模型时序特征的提取能力及减少模型对失稳样本的漏判,本文提出一种将时间卷积网络与特定类别成本调节极限学习机相融合的暂态稳定评估方法。首先,利用时间卷积网络挖掘蕴藏在电力系统运行数据中的时序变化特性;然后,改进模型损失函数提升模型对失稳样本的感知能力,并采用集成学习策略提高模型的泛化能力;最后,通过算例分析证明了所提方法具有很好的预测性能和泛化能力,对噪声数据及数据缺失也具备较强的鲁棒性。

To further enhance the time-series feature extraction capability of a power system transient stability assess⁃ment(TSA)model and mitigate its misjudgment for unstable samples,a TSA method which combines temporal convo⁃lutional network(TCN)and class-specific cost regulation extreme learning machine(CCRELM)is proposed in this pa⁃per.First,TCN is used to mine the time-series variation characteristics embedded in the operation data of power sys⁃tem.Then,the loss function of the model is improved to enhance its capability to perceive unstable samples,and an en⁃semble learning strategy is used to improve its generalization capability.Finally,the analysis of examples demonstrates that the proposed method has a good prediction performance and a good generalization capability,and it is also robust to noisy data as well as missing data.