Prediction of wharf subsidence deformation degree based on deep learning technology

This paper presents an algorithm that combines a convolutional neural network (CNN) with a gated recurrent unit (GRU) to predict the wharf subsidence deformation. First, the digital elevation model (DEM) image features of the wharf area were extracted using the CNN, and then the patterns of change in wharf settlement were captured using the GRU. Moreover, the wharf in the Longtan Port area of Nanjing Port, located in Jiangsu Province, was analyzed. When the CNN comprised three convolutional layers and the activation function was set to sigmoid, the prediction performance of the proposed algorithm was the best. In both short-term and long-term scenarios, the CNN+GRU algorithm had better prediction performance than long short-term memory and GRU models.