基于RF和自注意力ResNet的变压器PDPR

Transformer Partial Discharge Pattern Recognition Based on RF and ResNet with Self-Attentional

变压器局放是影响电网高效稳定运行的重要因素,也是变压器状态检测的重要内容。准确检测局放和识别局放类型是及时发现故障、排除隐患的前提。针对现有识别方法存在准确率不高的问题,提出一种基于RF和自注意力残差网络的变压器局放模式识别方法。方法构建了基于RF和自注意力残差模型,针对残差网络分类性能较低的问题,引入随机森林RF作为分类器,提高识别率的同时避免网络退化;并在特征提取器中融入自注意力机制,优化局放特征,加快模型训练。实验结果表明,所提方法可以较为准确地识别变压器局放的模式,准确率达98.96%,通过交叉实验验证,实际场景下,能够对不同环境下多种电压等级的变压器局放模式有效识别,具有较好的通用性。

Partial discharge of transformer is an important factor affecting the efficient and stable operation of power grid,and it is also an important content of transformer state detection.Accurate detection of partial discharge and identification of partial discharge pattern are the preconditions for timely fault detection and elimination of haz⁃ards.Aiming at the problems of low recognition rate of existing recognition methods,the paper presented a transformer partial discharge pattern recognition based on Random Forest(RF)and Residual Network(ResNet)with self-atten⁃tional.This method built a RF and self-attention ResNet model.Aiming at the problem of low classification perform⁃ance of residual network,this paper introduced RF as a classifier,which can improve the recognition rate and avoid network degradation.And it integrated the self-attention mechanism to optimize partial discharge features and accel⁃erate the model training.The experiment results show that the method can more accurately identify transformer partial discharge pattern,with 98.96%accuracy.And through the crossover experiments,in practical situations,the method can effectively identify transformer partial discharge pattern of various voltage levels in different environments.And it has good versatility.