Response analysis of subway station and optimization of seismic intensity measures based on fully connected neural network

In order to reduce the calculation cost of random seismic response analysis, the artificial neural network method is used to build a probabilistic seismic demand model (PSDM) to predict the seismic response of subway station structures, and the seismic intensity measure (IM) suitable for the prediction of subway station structural response is optimized. First, 200 measured ground motions are selected, IM is calculated, and the typical three-story and three-span subway station structure is modeled by the finite element method. Then, the IM and the maximum layer drift are used as the input and output to train the fully connected neural network (FCNN), and the prediction model for the maximum layer drift is obtained. Finally, the IM is optimized based on the weight matrix from the FCNN input layer to the hidden layer after training and the traditional methods, and the IM that has the greatest impact on the maximum layer drift is obtained. The results show that the FCNN after training can predict the maximum layer drift of subway station with an accuracy of 0.95, and the calculation efficiency is 18000 times higher than that of numerical simulation. For the maximum layer drift, the impact of velocity type and velocity response spectrum type indices is significantly higher than other types of indices, among which the velocity spectrum intensity (VSI) has the largest impact on the maximum layer drift.