为了解决运营线路站台门结构变形难以实时监控的问题，精准预测列车过站时高速铁路站台门的结构变形数据，采用基于人工智能的神经网络方法，选取210种不同车长、阻塞比、安装距离和车速的站台门结构变形数据作为网络模型训练样本，运用CNN（convolutional neural network）和基于K-Fold（K-fold cross-validation）的GRNN（general regression neural network）两种神经网络模型，建立了不同工况下的高速铁路站台门结构变形的预测模型，并与剩余样本数据进行对比验证。研究表明，两种模型均可有效预测铁路站台门结构运维数据，在预测精度上，基于K-Fold优化的GRNN模型优于CNN模型，基于K-Fold优化的GRNN模型的预测均方差能够维持在0.22之内，均方根误差维持在0.27之内，处于研究领域的领先水平。基于K-Fold优化的GRNN模型能够较好预测列车过站时的站台门结构形变量，为高速铁路站台门的设计与运维提供数据参考。

In order to solve the problem of real-time monitoring and accurate prediction of structural deformation of platform doors on high-speed railway lines, an artificial intelligence-based neural network method was used. Structural deformation data of platform doors, involving 210 different conditions of train length, blocking ratio, installation distance, and speed, were selected as training samples for the network model. Two neural network models, CNN(convolutional neural network) and K-Fold(K-Fold cross-validation) optimized GRNN(general regression neural network), were used to establish predictive models for platform door structural deformation under different working conditions of high-speed railways. These models were compared and verified with the remaining sample data. The research shows that both models effectively predict the operation and maintenance data of railway platform door structures. The K-Fold optimized GRNN model is superior to the CNN model in prediction accuracy. The Mean Square Error of the K-Fold optimized GRNN model is maintained within 0.22, and theRoot Mean Square Error is within 0.27, which is at the leading level in the field. The K-Fold optimized GRNN model better predicts the structural deformation of platform doors when trains pass, providing data references for the design and maintenance of high-speed railway platform doors.