大规模新能源并网、远距离大容量输电和高比例电力电子化等提高了电力系统发生大功率缺额事故和连锁故障的风险，而复杂电力系统的高维时变非线性特性使得实现快速准确地在线安全稳定分析与控制，存在理论制约和技术瓶颈。本项目针对电力系统暂态态势预测问题，基于物理分析方法与数据驱动方法相结合的思想，研究了暂态态势预测模型架构的构建、暂态态势关键因素的判定以及融合预测模型的自生长等三方面内容，主要包括：1）基于物理-数据融合的暂态态势预测模型架构及融合模式研究；2）提出暂态问题关键因素的物理-数据双层递进式判定技术，并基于池化技术研究暂态问题关键特征的降维与提取；3）提出基于继承思想的暂态态势预测模型的自生长方法，实现暂态运行场景数据在时间断面和空间层面上知识的深度和广度继承。最后，应用于暂态频率态势，提出了基于物理-数据融合的电力系统暂态频率态势预测模型，并基于多场景验证了所提理论与方法的合理性和有效

The risk of large power shortage and cascading failures in power systems has been increased with the emergence of large-scale renewable generation penetration, long-distance and large-capacity transmission and power electronics domination. Meanwhile, the accuracy and efficiency of online security and stability analysis are constrained both theoretically and technically due to modern complex power systems’ characteristics of high dimension, time variation and nonlinearity. This project aims at the prediction of transient situation awareness of power systems. In this project, three main issues will be addressed based on the idea of combination of physical-analysis and data-driven method, including the construction of transient situation awareness model frame, key factor selection and self-growing of the combined awareness model. The research contents are detailed as follows: 1) analyze the frame and fusion patterns of the prediction model based on combined physical and data-driven modeling; 2) propose a double-layer determination technique based on combined physical and data-driven idea for key factors of transient stability assessment and implement the dimensionality reduction and feature selection using pooling technology; 3) propose a self-growing method for transient situation awareness modeling based on the idea of knowledge inheritance in both time and space scale. Finally, the research result is applied to the prediction of transient frequency situation awareness in which an awareness model is established based on physical and data-driven modeling. The rationality and validity of the proposed theory and method is verified in multiple scenarios.