电力系统的静态电压稳定性，对于保证系统正常运行起着关键作用。传统的静态电压稳定裕度评估方法难以满足 在线实时预测的要求，常用的离线监督预测方法则需要大量 的训练样本，且容易出现过拟合，会对预测精度造成影响。 研究能克服这些缺点的方法，具有重要意义。该文将神经网 络、半监督训练、集成学习等技术应用于电力系统静态电压 稳定裕度的预测分析中，提出基于 Tri-Training-LASSO-BP 网络的在线预测方法，由三体训练法(Tri-Training)、最小绝 对值收缩选择(least absolute shrinkage and select operator， LASSO)方法和误差反向传播(back propagation，BP)神经网 络组成。在 IEEE 39 节点和 IEEE 300 节点算例上的结果和 对其进行的非参数检验表明，该方法能够降低对训练集数据 量的要求，发挥电力系统日常运行过程中采集的海量数据的 优势，提高网络的预测精度，减少人工干预。

The static voltage stability of power system plays a key role in ensuring operation of the system. Traditional static voltage stability margin assessment methods are difficult to meet the requirements of online real-time monitoring. Common offline surveillance prediction methods require a large number of training samples, and are prone to over-fitting, leading to impact the accuracy of prediction. Therefore, it is of great significance to study ways overcoming these weak points. In this paper, neural network, semi-supervised training, integrated learning and other techniques were applied to the prediction and analysis of static voltage stability margin of power systems. An online prediction method based on Tri-Training-LASSO-BP network was proposed. The network consists of Tri-Training, the least absolute shrinkage and select operator (LASSO) algorithm and the back propagation (BP) neural network. The results on the IEEE 39-node example and IEEE 300-node example and non-parametric tests performed on the results show that the proposed method can reduce the requirement of the data volume of training set, take advantage of the massive data collected during the daily operation of the power system, improve the prediction accuracy of the network and reduce manual intervention.