基于聚类自适应主动学习的电力系统暂态稳定评估

Transient stability assessment of power system based on clustering adaptive active learning

为了解决电力系统暂态稳定评估中机器学习方法所需样本数多、仿真耗时长的问题,提出主动学习的方法。为了降低主动学习过程中选择样本的冗余度,提出一种聚类自适应主动学习选择策略。通过聚类选择初始样本,使初始样本具有代表性,加快了主动学习进程;将不确定性和代表性2种指标结合自适应地选择权重参数,使选择的样本冗余度低。CEPRI 36节点系统仿真结果表明,主动学习的方法可以在保证准确率的情况下显著降低所需样本数,节省了大量的仿真耗时,参数自适应的选择策略相比传统的不确定性策略大幅提高了训练效率。

In order to solve the problems that large number of samples and long simulation time are needed by machine learning method in transient stability assessment of power system,an active learning method is proposed.In order to reduce the redundancy of sample selection in active learning progress,a clustering adaptive active learning selection strategy is proposed.The initial samples are selected by clustering,which makes the initial samples representative and speeds up the active learning process.The two indicators of uncertainty and representativeness are combined to adaptively select the weight parameters,which makes the redundancy of the selected samples low.The simulative results of CEPRI 36-bus system show that the active learning method can significantly reduce the needed number of samples while ensuring the accuracy,which saves much simulation time,and the adaptive parameter selection strategy greatly improves the training efficiency compared with the traditional uncertainty strategy.