船用三相变压器运行状态监测与故障预警诊断技术研究

Marine Three-phase Transformer Running State Monitoring and Fault Warning Diagnosis Technology Research

采用电气量信息样本作为变压器状态监测与故障预警诊断的判据,比用变压器油中溶解气体成分作为判据更具可操作性,样本便于获得,适用于船用变压器的动态监测。采用概率神经网络以变压器电气量信息波形系数作为故障鉴别的"阈值",具有准确、快速判别故障原因的特性,可对变压器故障进行先期预测,为主动维修争取时间。实验证明,变压器故障样本空间越大,样本数据越准确,概率神经网络的训练效果越好,预测效果越佳。

This paper take the electric samples of information as a criterion of the transformer condition monitoring and fault warning diagnosis. It is more maneuverability than taking the transformer oil dissolved gas composition as criterion.Because of the sample is easy to obtain,so meeting marine transformer dynamic monitoring. Probabilistic neural network was adopted to information waveform coefficient transformer electric parameters as the 'threshold' of fault identification.It is accurate and rapid identifying the cause of the problem characteristics,and be advanced for transformer fault forecast and get proactive maintenance time. The experiment proved that if the transformer fault sample space is bigger and the sample data is more accurate,and the training of probabilistic neural network is more effect,then the prediction effect is better.