BP神经网络在MOA故障诊断中的应用

New Application of BP Neural-network to MOA Fault Diagnosis

介绍了BP神经网络在MOA故障诊断中的应用情况。运用Matlab神经网络工具箱对避雷器故障进行训练,比较可知,采用改进的自适应修改学习率算法,具有较好的网络收敛性能和稳定性,并且能够有效解决局部最小问题,但是隐含层神经元数目只能通过试凑法得到。实例分析表明,运用神经网络方法能较为准确诊断避雷器故障,能够有效地克服单纯隶属函数对正常运行工况的误诊断。

A new application of BP neural-network to MOA fault diagnosis is introduced in this paper. The MOA fault is trained by Matlab neural-network tool-box. In the contrast, the improved algorithm with self-adaption modified learning rate have better net-work convergence property and stability, further more, it can solve the problem of local minimum effectively. But, we can only gain the nerve unit number of implication layer by method of trial and error. The analysis of example demonstrates that the proposed method can diagnose the fault of MOA accurately and can avoid the incorrect diagnosis of regular condition by pure membership function effectively.