基于KNN改进的BP神经网络的油浸式配电变压器故障诊断

Fault diagnosis technology of oil-immersed distribution transformer based on KNN improved BP neural network

电力变压器一直是高压变电站系统和中小型发电厂使用的一项重要辅助电力设备,假如其工作一旦设备发生故障,系统将不能稳定的正常工作。同时,伴随着变压器电气等级的上升,变压器的价格就越贵重,如果因为故障受到损坏,对我国的经济社会的发展造成不可挽回的后果。由于科技的进步,传统的检测方法精度已经无法满足判别油浸式变压器故障的安全要求,因此需要一个安全性和可靠性高的便携的检测变压器故障的保护方法。从BP神经网络出发,构造了BP模型并对之训练仿真,得到识别率为60%。为了进一步提高故障识别率和精度,引入深度学习的概念,阐述了KNN网络的基本原理,以及其网络的结构和工作机制,据此又设计了一个KNN网络的模型,并使用DGA数据做好网络训练和结果剖析,识别率可达到90%。接着对两个模型做了对比,发现KNN优于BP神经网络。

Power transformer is the main power equipment of power plant and substation.If it fails,the system will not work stably and normally.With the development of transformer,the more expensive the cost of transformer will be,and the more serious the damage to the whole society will be.Due to the progress of science and technology,the accuracy of the existing traditional detection methods can not meet the safety requirements of judging transformer faults.Therefore,a safe,reliable and portable protection method for detecting transformer faults is needed.Starting from BP neural network,BP model is constructed and trained and simulated,and the recognition rate is 60%.In order to further improve the fault recognition rate and accuracy,the concept of deep learning is introduced,and the basic principle of KNN network,as well as its network structure and working mechanism are described.Based on this,a KNN network model is designed,and the network training and result analysis are done with DGA data.The recognition rate can reach 90%.Then the two models are compared,and it is found that KNN is better than BP neural network.