变压器油色谱在线监测中BP神经网络算法分析

Application of Back Propagation Neural Network in Dissolved Gas Analysis Based on Transformer Oil Chromatographic On-line Monitoring

针对现今变压器故障诊断方法存在的编码不齐全、准确率不够高等问题,提出了一种基于BP神经网络的变压器油色谱在线监测综合智能诊断方法。该方法结合国标阈值诊断以及改良三比值法,运用BP神经网络理论诊断变压器综合运行状态。运用Matlab建立基于特征气体的BP神经网络变压器故障诊断模型,发现BP神经网络具有良好的特征提取功能,但是通过不断训练发现,只运用BP神经网络对变压器进行诊断得到的变压器运行状态并不是十分准确。最后,结合常用的比值法,通过仿真对实例进行综合诊断,得出此方法运用到变压器故障诊断中具有更高的准确性。

In the light of the exiting problem of incomplete code, low accuracy in transformer fault diagnosis method, the paper puts forward a synthetically intelligent diagnosis method based on the back propagation artificial neural network and dissolved gas analysis of transformer oil chromatographic on-line monitoring. The BP neural network theory is used in diagnosing transformer synthetical operation state with the combination of the methods of national standard threshold diagnosis and improved three-ratio method. ~ Matlab is used in establishing the transformer diagnosis model which is based on characteristics gas and BP neural network. Besides, the BP neural network has good feature extraction function. However, only using BP neural network diagnosing transformer operation state is not very accurate through continual training. Finally, it is shown that the method which is used to transformer synthetical fault diagnosis combining with the common ratio method has higher precision through the simulation of practical case.