基于双模型融合的电缆局部放电模式识别

Cable Partial Discharge Pattern Recognition Based on Dual Model Fusion

由于恶劣多变的海洋环境,电缆局部放电信号采集过程中存在很多未知而又复杂的因素,导致放电类型的识别准确率不高,易发生误识别。针对这一问题,提出一种基于双模型融合的电缆局部放电模式识别方法。分别构建基于残差网络和移动端网络的识别模型,引入D-S证据理论对单一模型的识别结果进行融合。针对证据冲突的情况,引入Hellinger距离改进D-S理论中的权重分配,提高识别准确率与稳定性。试验采用现场采集的局放信号所构成的PRPD图谱数据集进行验证。试验表明,融合后的模型平均识别率为97.22%,召回率为95.59%,双模型融合的各项性能均比单一模型有所提高,有效降低了误识别的发生,提升了识别结果的可信度,可更准确地实现对各类局部放电的模式识别,为海底电缆的稳定运行提供更好地支撑。

Due to the harsh and changeable ocean environment, there are many unknown and complex factors in the cable partial discharge signal acquisition process, which leads to the low accuracy of the discharge type recognition and prone to misrecognition. To address this problem, a method of cable partial discharge pattern recognition is proposed based on dual model fusion. The recognition models based on resnet and mobilenet are constructed respectively, and the D-S evidence theory is introduced to fuse the recognition results of a single model. In the case of conflicting evidence, Hellinger distance is introduced to improve the weight allocation in D-S theory to enhance the recognition accuracy and stability. The experiment is verified by the PRPD atlas dataset composed of partial discharge signals collected on site. The experiment shows that the average recognition rate of the fused model is 97.22%, and the recall rate is 95.59%. The performance of the fused model is better than that of the single model, which can effectively reduce the occurrence of misrecognition, enhance the credibility of recognition results, and realize the pattern recognition of various partial discharge more accurately which can provide good support for the stable operation of submarine cables.