利用反向传播神经网络研究变压器油多关联参数

介绍用反向传播（back propagation,BP）神经网络对变压器油的重要参数——击穿电压建立预测模型,实现对变压器油性能监测的方法,阐述了网络层数、神经元个数、训练函数的设计过程,样本训练的实验结果证明该网络模型具有较好的预测能力;同时,基于BP神经网络的建模方法建立包括变压器油击穿电压、闪点、酸值、总烃、水分等参数之间关联的BP网络预测模型,将2种模型进行比较发现,网络预测模型的预测结果与实际结果的相对误差较小,从而证明该预测模型具有一定的实际意义。

Using BP Neural Network in Studying Multi-correlation Parameters of Transformer Oil

This paper dcscribes the method of using BP（back propagation）neural network to develop prediction model for breakdown voltage, an important parameter of transformer oil, so as to realizc the monitoring on transformer oil performance. The design process of the number of network expounded. The test results of sample training prove that the layers, the number of neurons and the training function is network model is of bcttcr predicting ability. Meanwhile, BP network prediction model reflecting the correlation among transformer oil parameters such as breakdown voltage, flash point, acid number, total hydrocarbon and water content is built based on the modeling method of BP neural network. Upon comparison, it is found that the network prediction model has smaller relative error; therefore it is of practical meaning.