An Integrated Graph Convolutional Network and Finite Element Analysis Model for Urban Ground Settlement Prediction

In the engineering problem of urban ground settlement prediction, the shortcoming of traditional methods is that a single model has limited predictive ability for ground settlement problems, especially in balancing precision, efficiency and global correlation. This paper aims to establish a prediction model with physical rationality and data-driven capabilities by combining the advantages of graph convolutional networks (GCN) and finite element analysis (FEA). The research’s core lies in integrating physical simulation and graph structure learning to solve complex problems in urban ground settlement prediction and provide effective solutions for practical engineering. The model’s construction in this paper first involves the graph model construction, in which the node definition is derived from the grid unit in the FEA model. Each node corresponds to a discrete foundation area unit and contains physical properties such as soil type, bearing capacity and compression modulus. Edge weight calculation considers geographical proximity, building load, and groundwater level differences. Feature representation combines physical simulation data from FEA with other engineering features. Then, the integration of FEA and GCN is realized through three key steps: data preprocessing, feature mapping and spatial dependency modeling. Finally, the constructed model is trained, and hyperparameters are tuned for the best performance. The model built in this paper is tested. The experimental results show that the mean squared error (MSE) of the GCN[Formula: see text]FEA model in the city center, suburbs, and industrial areas is 0.0123, 0.0154 and 0.0112, respectively. The coefficient of determination (R-squared, R 2 ) reaches 0.95, 0.92 and 0.96, respectively, which are significantly better than those of traditional machine learning models such as support vector machine (SVM), random forest (RF) and gradient boosting decision tree (GBDT). This demonstrates the GCN[Formula: see text]FEA model’s advantage in prediction precision and adaptability in different geographical areas.