变压器油中故障气体的复合预测方法  

Compound Approach of Predicting Fault Gases Dissolved in Transformer Oil

作　　者：周利军[1] 吴广宁[1] 宿冲[1] 王洪亮[1] 

机构地区：[1]西南交通大学电气工程学院,四川成都610031

出　　处：《西南交通大学学报》2006年第2期150-153,共4页Journal of Southwest Jiaotong University

基　　金：由铁道部科技开发项目(2002J036);电力设备电气绝缘国家重点实验室开放课题基金项目

摘　　要：为了提高对变压器故障的预测能力,提出了灰色粗预测、自学习神经网络在线修正的复合预测法.此法是利用GM(1,1)模型初步预测油中溶解气体的浓度及变化趋势,通过分析故障气体组分之间的影响及气体浓度时间序列之间的关系确定修正参数,将初步预测结果与修正参数作为自学习BP网络的输入,从而完成预测结果的在线修正.该方法已用于实际变压器油中溶解气体的预测,结果验证了其有效性.To improve the prediction result for transformer faults, a compound approach combining GM（ 1,1 ） model with self-learning BP-neural networks was proposed to predict fault gases dissolved in transformer oil. In this approach, the concentration and development trend of gases dissolved in transformer oil are predicted primarily using GM （1,1） model, and then the predicted results are calibrated by self-learning BP-neural networks with calibrated parameters obtained by analyzing the interaction of different types of gases and the relationship between the time sequences of gas concentrations. The proposed approach has been used in the practice of transformer fault prediction to show its validity.

关 键 词：变压器油 故障气体预测 GM(1 1) 自学习BP网络 时间序列 

分 类 号：TM762[电气工程—电力系统及自动化]