摘要
基于变压器油性能参数之间的关联性,采用BP神经网络方法,在Matlab平台下研究预测多参数关联变压器油的性能,利用变压器油日常的监测数据,建立击穿电压与4个影响因素的关联模型。论文分别就常规BP算法和变学习速率、变动量因子的改进BP算法进行了比较研究,结果表明,改进BP算法模型的预测结果精度较高,预测值与实际值的相对误差在5%左右。本方法可以为变压器故障的早期诊断、预测防范和及时处理提供科学依据,具有重要的实际应用价值。
Based on the fact that performance parameters of transformer oil do correlate, two BP networks were developed respectively to simulate the correlation between performance parameters, under the development environment of Matlab. The first BP network was trained with BP algorithm, and the second BP network was trained with an improved BP algorithm with a variable momentum factor and variable learning rate. Both networks used the monitoring data of transformer oil as training samples to simulate the correlation between breakdown voltage and 4 relevant parameters. Breakdown voltage is one of the most important parameters of transformer oil that reflects its insulation performance. The results show the network trained with improved BP algorithm gives more accurate predicted values of breakdown voltage; the relative errors between predicted values and real values are around 5% , which proves to be of practical value. Prediction of transformer oil performance is of great significance in the early fault diagnosis of transformer, it also provides scientific basis for the prevention of fault in transformer.
出处
《计算机与应用化学》
CAS
CSCD
北大核心
2008年第6期737-740,共4页
Computers and Applied Chemistry
关键词
变压器油
性能预测
击穿电压
BP神经网络
多参数关联
transformer oil, prediction, breakdown voltage, Back-Propagation neural network, multi-parameter correlation
作者简介
李睿（1985-），女，硕士研究生，水质工程系应用化学专业；
导师：曹顺安（联系人）。
