判别式
计算机科学
断层(地质)
生成语法
数据挖掘
基线(sea)
网格
特征(语言学)
能量(信号处理)
对抗制
人工智能
机器学习
数据建模
人工神经网络
序列(生物学)
故障检测与隔离
联合概率分布
模式识别(心理学)
数据驱动
稀缺
节点(物理)
生成模型
作者
Zaiyan Zhou,Chunxiu Liu,Yimin Liu,Longtan Li,Yuqi Zhang,Wenxin Wang,Xuan Liu,Ruixue Yan
标识
DOI:10.1088/2631-8695/ae6754
摘要
Abstract This paper proposes a fault data generation method for new energy distribution networks based on a grid time sequence generative adversarial network (GTSGAN), aiming to alleviate the scarcity of measured fault data and reduce reliance on scenario-specific simulation configurations. The proposed framework incorporates sliding-window temporal feature injection, dual-stream embedding, static–dynamic joint representation, and consistency-enhanced training objectives to support the synthesis of high-dimensional fault sequences. Experiments are conducted on operational measurements from a real distribution network, with Simulink-generated data included as a simulation-based reference for comparison. The proposed model attains a mean discriminative score of 0.1034 and a mean predictive score of 0.0153, which are lower than those of representative baseline generative models under the same settings. In addition, ablation studies are performed to examine the contributions of the main components. The results indicate that GTSGAN offers a practical data-driven approach for generating high-dimensional fault sequences to support subsequent analyses in new energy distribution networks.
科研通智能强力驱动
Strongly Powered by AbleSci AI