Wind prediction model of fan gearbox based on LSTM network

  1. LSTM principle

LSTM model is a special form of recurrent neural network (RNN), which can mine the relationship between the time of running data and solve the problem of gradient disappearance and explosion in the process of training. As shown in Figure 1, LSTM model controls the information input and output of memory unit through input gate, forgetting gate and output gate. The calculation process is shown in the formula.

Where: ft is the forgetting gate; it is the input gate; OT is the output gate; CT is the cell state; HT is the output value; W is the weight value; B is the bias; โŠ— is the multiplication point by point.

  1. Gearbox bearing temperature prediction model

Similar to the traditional neural network, the network structure of the prediction model is composed of input layer, hidden layer and output layer. The input of the model is the feature parameters screened by mutual information method, and the output of the model is the temperature of the gearbox bearing. By setting an appropriate time step, the feature information of the operation data at the previous time and the current time is extracted in depth. In order to avoid over fitting, dropout layer should be added after hidden layer, and output value of hidden layer should be given up according to a certain proportion. According to the evaluation criteria of the model, the number of neurons in the hidden layer, the number of iterations (epochs) and the number of samples per processing (batch size) of the model are determined through many experiments.