Algorithm 1 Training Process of the Ensemble Network.

1:Establish the ensemble network through the single networks (multi-layer LSTM network) 2:Initialize the weights of each single network in ensemble network 3:for iterations do 4:  for $z=1$ to $N_e$ do 5:    Make a batch set $b_z$ by random sampling in training data 6:    Only the variables in single network z are assigned as trainable variables 7:    Minimize the loss $L_z$ of the batch set $b_z$ through the Adam optimizer 8:  end for 9: end for 10:Verify the performance with test data