基于RNN的故障预测算法及在GIS上的应用

Fault Prediction Algorithm Based on RNN and Its Application of GIS

摘要:

随着工业化程度的提高，设备的故障预测的重要性日趋提高。提出了一种基于循环神经网络（RNN）的故障预测算法，通过数据训练，充分发掘了RNN对时间序列数据的拟合能力。RNN故障预测模型由数据处理模块和神经网络识别模块组成。在数据处理模块中，采用数学函数分配的方法建立了RNN 模型的训练样本和测试样本。在神经网络识别模块中，针对当前故障预测技术中异常点难以确定的问题，应用了一种逐步逼近的神经网络训练方法。最后利用气体绝缘开关(GIS)故障数据对该算法进行了验证，结果表明，该方法可以在故障发生前检测到故障发生趋势，进而实现故障预测，并且能在逐步训练中确定异常点的位置。

Abstract:

With the development of industrialization, the importance of equipment fault prediction is increasing day by day. A fault prediction algorithm based on Recurrent Neural Network (RNN) is proposed. The fitting ability of RNN to time series data is fully explored through data training. The fault prediction model of the RNN is composed of a data processing module and a neural network recognition module. In the data processing module, the method of mathematical function assignment is used to build the training and test samples of the RNN model. In the Neural network recognition module, due to the problem that the change point in current fault prediction technology is difficult to determine, a neural network training method based on successive approximation is used. At last，the RNN fault prediction algorithm is verified by Gas insulated switchgear（GIS） fault data. The results show that the trend of fault can be detected by this method before it occurs, the fault prediction can be realized, and the change point can be located in the gradual approximation training.