基于RKELM的中压电缆早期故障识别与定位方法研究

Research on incipient fault identification and localization method for medium voltage cables based on RKELM

实际运行经验表明,部分电缆在发生永久性故障前会经历早期故障,表现为瞬时性、重复性的对地电弧击穿。及时检测和识别早期故障扰动对于实现故障预警具有重要的理论价值和工程意义。由于早期故障通常表现为缆芯对屏蔽层的瞬时击穿,因此将在屏蔽层接地电流中产生明显的扰动。为此,提出了一种基于屏蔽层接地电流的早期故障识别和定位方法。该方法基于故障电流积分、故障波动长度特征值、双端接地电流方向差异度和双端初相位差4个波形特征进行识别和定位,并将这些提取的特征与快速简化核极限学习机(fast reduced kernel extreme learning machine,RKELM)模型相结合,用于故障波形的快速、准确识别和区段定位。利用大量仿真数据对RKELM模型进行训练,使其能够快速、准确地识别和定位故障波形。仿真测试的结果表明,即使在20 dB噪声影响下,该方法的分类准确率仍能保持在96.7%以上。

Operational experience indicates that some cables undergo incipient faults before progressing to permanent failures,characterized by transient and repetitive ground arc breakdowns.Timely detection and identification of incipient fault disturbances are of significant theoretical value and engineering importance for achieving fault warning.Since incipient faults typically manifest as instantaneous breakdowns between the cable core and the shielding layer,they generate noticeable disturbances in the shield layer grounding current.To address this,this paper proposes an incipient fault identification and localization method based on shield layer grounding current.The method identifies and locates faults based on four waveform features of fault current integration,fault fluctuation length eigenvalue,double-end current direction difference and double-end initial phase difference for identification and localization.These extracted features are combined with an RKELM model for rapid and accurate fault waveform recognition and segment localization.Extensive simulation data are used to train the RKELM model,enabling it to quickly and accurately identify and locate fault waveforms.The test results through simulation testing indicate that even under the influence of 20 dB noise,the classification accuracy of this method remains above 96.7%.