上海交通大学学报

上海交通大学学报

• •    

基于物理信息嵌入的非固定长度电力系统暂态稳定快速评估（网络首发）

  

  1. 1. 武汉大学电气与自动化学院；2. 南方电网有限责任公司
  • 基金资助:
    南方电网有限责任公司科技项目（0000002022030101XT00031）

Rapid Non-Fixed Length Transient Stability Assessment of Power System Based on Physics-Informed Neural Networks

  1. （1. School of Electrical and Automation Engineering, Wuhan University, Wuhan 430072, China;2. Southern Power Grid Co., Ltd., Guangzhou 510663, China）

PDF (PC)

725

摘要： 在双碳目标下，构建以新能源为主体的新型电力系统是实现电力工业转型升级的主要方向和关键途径，新型电力系统背景下快速准确的暂态功角稳定评估研究具有重要意义。为此，文章基于物理信息嵌入序列到序列神经网络(PI-seq2seq)与级联卷积神经网络模型提出一种含构网型新能源的新型电力系统暂态功角稳定评估方法。首先，采用PI-seq2seq网络结构预测未来功角轨迹，通过构造含物理损失项的损失函数引导模型训练过程，避免了时域仿真耗时过长对快速暂态评估带来的影响。其次，级联卷积神经网络以预测的功角轨迹作为输入评估暂态稳定情况及其置信度，并配置了评估置信度阈值判断机制实现非固定评估长度的暂态稳定判断，克服了固定功角曲线长度对评估结果带来的影响。最后，在Kundur系统中进行了验证，仿真结果表明所提方法在功角曲线预测与稳定评估方法均获得了令人满意的结果。

关键词: 构网型新能源, 物理信息嵌入序列到序列神经网络, 功角轨迹预测, 级联卷积神经网络, 暂态功角稳定评估

Abstract: Under the double-carbon goal, the construction of a new power system with new energy as the main body is the main direction and key way to realize the transformation and upgrading of the power industry. The research on fast and accurate transient power angle stability evaluation under the background of new power systems is of great significance. To this end, the paper proposes a new transient power angle stability evaluation method for power systems with grid-forming new energy based on the physics-informed sequence-to-sequence neural networks (PI-seq2seq) and cascaded convolutional neural networks model. First, the PI-seq2seq network 

structure is used to predict the future power angle trajectory, and the loss function with physical
loss terms is constructed to guide the model training process, which avoids the impact of
time-domain simulation taking too long on fast transient evaluation. Secondly, the cascade
convolutional neural networks use the predicted power angle trajectory as input to evaluate the
transient stability and its confidence level, and configure the evaluation confidence level threshold
judgment mechanism to realize the transient stability judgment of the non-fixed evaluation length,
which overcomes the influence of the fixed power angle curve length on the evaluation results.
Finally, it is verified in the Kundur system, and the simulation results show that the method
proposed has obtained satisfactory results in both the power angle curve prediction and the
stability evaluation method.

Key words: grid-forming new energy, physics-informed sequence-to-sequence neural networks (PI-seq2seq), power angle trajectory prediction, cascade convolutional neural networks, transient power angle stability assessment

中图分类号: