具有增量学习能力的智能孤岛检测方法

基于机器学习的智能孤岛检测方法能有效地提高防孤岛保护的性能,但现有方法皆采用离线学习方案,对配电网因运行条件变化而导致的概念漂移现象缺乏自适应性。提出了一种具有在线增量学习能力的孤岛检测方法。首先,提出利用保护自采数据以及数据采集与监视控制(SCADA)系统采集的开关状态构成原始样本,并基于增量聚类方法进行样本筛选,实现有效样本的在线积累;然后,以各子样本集对系统最新状况的分类性能作为竞争准则,提出了一种样本集的优选方法,并利用加权支持向量机完成了增量学习。仿真结果表明,所提方法能够自主探测概念漂移的发生并进行持续的学习,有效地提高了孤岛检测的准确性和自适应性。

Intelligent islanding detection method with incremental learning capability

The machine learning-based intelligent islanding detection method can improve the performance of anti-islanding protection effectively. However, the existing methods are all based on the offline learning scenario, which can not adapt to the concept drifting phenomenon caused by the changes of operating condition in the distribution network. An islanding detection method with online incremental learning capability is proposed. Firstly, the primary samples are composed of the field data collected by protection devices and the switch states collected by the SCADA(Supervisory Control And Data Acquisition) system, and selected based on the incremental clustering method to accumulate the effective samples online. Then, an optimal selection method of sample set is proposed by taking the classification performance of each subsample set to the latest situation of the system as the competition rule, and the support vector machine is applied to realize the incremental learning. Simulative results show that, the proposed method can effectively improve the accuracy and adaptability of islanding detection by detecting the concept drift automatically and performing learning continuously.