Research on Forecast and Alert Model for Journal Temperature of EMU

Changing pattern of Journal temperature of EMU shows the performance of shaft. Recent advances in deep learning technologies make it possible to learning features of Journal temperature. In this paper, we designed a special convolution-recursive neural network for pattern learning of Journal temperature, forecasting short-term variation and early warning of potential failures. We design multiple low-level convolutional neural network (CNN) learning Journal temperature changing mode which are then given as inputs to some Recursive neural network (RNN) in order to forecast Journal temperature. The recursive network also have fault data as inputs to learning probability of potential failures. Our main result is that this convolution-recursive neural network obtains state of the art performance parameters. On this basis, the network can be effective on short-term forecasting and potential failure warning.1