基于动态模式识别的交直流电力系统换相失败早期征兆捕获与分类研究
Research on Early Detection and Classification of Commutation Failure in AC-DC Power Systems Based on Dynamic Pattern Recognition

换相失败是交直流电力系统运行中的常见故障，尤其在高压直流输电(HVDC)系统中，其发生会导致直流电压下降、交流电网电压畸变，甚至引发系统振荡和大规模停电。换相失败通常由交流系统电压扰动、直流系统功率波动及控制策略不足等因素引起，准确捕获其早期征兆并进行分类对于提升电网稳定性具有重要意义。现有方法主要依赖物理模型计算或基于传统特征提取的机器学习方法，难以满足复杂电网环境下的实时检测需求。动态模式识别技术能够从时序数据中提取关键特征，并结合深度学习方法提高故障检测的准确性。针对换相失败的时序特征，基于动态模式识别的故障检测方法能够在换相失败发生前识别关键信号，提高系统的预警能力，并实现对不同类型换相失败的分类，有助于电网安全策略的优化。
Commutation failure is a common fault in the operation of AC-DC hybrid grids, particularly in high-voltage direct current (HVDC) systems. Its occurrence can result in a reduction of DC voltage, distortion of AC grid voltage, and may even lead to system oscillations and large-scale power outages. Commutation failure is typically caused by factors such as voltage disturbances in the AC system, fluctuations in the DC system’s power, and inadequate control strategies. Accurately capturing its early signs and classifying them is critical for enhancing grid stability. Current methods primarily rely on physical model calculations or machine learning techniques based on traditional feature extraction, which are inadequate for meeting the real-time detection requirements in complex grid environments. Dynamic pattern recognition methods, when combined with deep learning, can extract key features from time-series data, improving fault detection accuracy. For commutation failure’s time-series characteristics, the dynamic pattern recognition-based fault detection method can identify key signals before failure occurs, enhancing early warning capabilities and enabling classification of different types of commutation failures. This helps optimize grid safety strategies.