Training method and device for intelligent prediction model of shield tunneling speed of composite stratum

The invention provides a training method and device for a composite stratum shield tunneling speed intelligent prediction model. The overall generalization ability and operation efficiency of the model can be improved. The method comprises the steps that a shield tunneling parameter time series data set is obtained, wherein the shield tunneling parameter time series data set comprises composite stratum characteristic parameters, tunneling machine operation parameters and shield tunneling mode parameters; preprocessing the shield tunneling parameter time sequence data set to obtain a model training data set; dividing the model training data set into a training data subset and a verification data subset based on each type of shield tunneling parameter time sequence data in the shield tunneling parameter time sequence data set; constructing an initial tunneling speed prediction model based on the LSTM, wherein hyper-parameters of the initial tunneling speed prediction model comprise the number of neurons of a hi