Abstract: To solve the problem of low prediction accuracy caused by lack of subgrade settlement data,a prediction model of settlement interval of high-speed railway subgrade combined with PS-InSAR technology,generative adversarial network(GAN) data expansion and GRNN-BP composite neural network is proposed.Firstly,PS-InSAR technology is used to obtain the settlement value of the subgrade,analyze the correlation between environmental factors and the settlement of the subgrade,and build the original sample set.Secondly,the network connection layer is set up to connect the two kinds of networks,and the advantages of GRNN and BP neural network are used to form a composite neural network.Finally,GAN is used to expand the data set,and the expanded data set is input into GRNN-BP to predict the settlement of high-speed railway subgrade.The experimental results show that the prediction accuracy of the model can be improved effectively by inputting the expanded data samples into the composite neural network and training it.GRNN-BP can not only provide high-precision point prediction results,but also construct clear and reliable prediction intervals.Compared with the other four models,GRNN-BP has more reliable prediction results for the settlement of high-speed railway subgrade.

Key words: PS-InSAR, generative adversarial network, composite neural network, subgrade settlement, interval prediction