Research and Application of Deformation Prediction Model for Deep Foundation Pit Based on LSTM

Algorithm 4

AdaGrad algorithm.

Require: set the initial learning rate to , and the value of of the initial constant maturity is 10^-7, the initial parameter is set to

Require: set the initial gradient accumulative value to 0

While stop condition not satisfied do

Randomly extract samples from the training set, and is the true value corresponding to .

Forecast point update:

Calculate the average gradient of samples:

Square of accumulated value of historical gradient:

Update gradient value: (element operation)

Parameter updating:

End while