Data-driven transient stability assessment models are usually trained offline using many samples from preset conditions

but the performance of the pre-trained models usually fails to meet the requirements when the operation conditions of power grids change greatly. To solve this problem

this paper proposes a transient stability assessment method of power systems based on domain adversarial neural networks (DANN) feature mapping calibration and improved active learning. Firstly

the feature extractor of DANN is used to map the data of the original condition and the new condition into a high-dimensional feature space to reduce the distributional difference between the two; secondly

a pair of stable and unstable samples in the new condition are labeled

and the distributional calibration is used to transfer the distributional information of the original condition to expand the sample set of the new condition

to train a high-performance base model of the new condition. Then

an improved active learning algorithm is utilized to select high-value and class-balanced samples to fine-tune the base model

to improve the model's performance quickly. Finally

the effectiveness and time efficiency of the proposed method are verified on the IEEE-39 node system and the 10

000-node test system.