Transient Stability Assessment Based on Two-Step Improved One-Dimensional Convolutional Neural Networks with Long and Short Time-Series Input

The traditional power system transient stability assessment (TSA) is based on time domain simulation calculation, which has high time complexity and cannot meet the requirement of rapid TSA for online emergency control. To solve this problem, this paper proposes a fast TSA method based on improved one-dimensional convolutional neural networks (ID-CNN) to support quick and accurate online stability assessment. The type and characteristics of input data is utilized to design the convolutional layers to facilitate feature extraction by avoid information confusion. Furthermore, 1D-CNNs with different lengths time-series input are designed to balance assess time and credibility, which is useful for the following control decision. The effectiveness of the proposed method is verified by the TSA on the IEEE39 system with three-phase short-circuit faults under various operation patterns and different fault last time.