Bridge Pier Displacement Prediction and Control in Subway Tunnel Construction

Abstract

As the scale of underground rail transit construction in urban areas continues to expand, the tunnel construction environment has become progressively more complex. In recent years, an emerging artificial intelligence (AI) method in the civil engineering field, called the Random Forest (RF) method, has been widely used. In the construction of Zhengzhou Metro Line 7, the RF method was used to predict and control the vertical displacement of the bridge pier pile foundation. Such displacement can indicate the deformation of the structure, particularly under long-term utilization and strenuous circumstances that could sink or lift the pier body. Moreover, the vertical displacement of the bridge pier can affect the stiffness and bearing capacity of the bridge, thus impacting driving safety and the bridge's service life. Therefore, the vertical displacement of bridge piers has become the main prediction and control indicator for research. In the Zhengzhou Metro Line 7 tunnel, the tunnel continuously passes through 78 bridge pier foundations, among which the pile foundations of 4 key bridge piers are less than 0.5 times the tunnel diameter with a clear distance planned for the tunnel. However, limitations such as surface traffic and environmental conditions prevent the reinforcement of the bridge pier foundation in advance. Therefore, determining and setting sensible shield construction parameters is crucial to effectively controlling the vertical displacement of these essential bridge piers. This project can serve as a model for future endeavors. The study combines Random Forest with Particle Swarm Optimization Algorithm (PSO) to upgrade the technology of shield tunneling through Pier 2, introduces the Bayesian principle for statistical analysis, and optimizes various main construction variables. Random Forest is an ensemble learning method based on decision trees, which has high flexibility and predictive performance. It can automatically filter out important features from a large number of input features, thereby establishing an effective prediction model. The primary research objective is to enhance tunnel construction by accurately predicting and controlling the vertical displacement of pier foundations. To achieve this objective, the study utilizes the PSO to optimize the parameters and structure of the RF model. By doing this, the model's ability to predict the pier's vertical displacement accuracy can be improved. By combining these two methods, the accuracy of the prediction model and the optimization effect of construction parameters can be improved. In addition, the reliability of the model is further improved by using the Bayesian principle for statistical analysis. The paper compares and evaluates the engineering data objectively, presenting the evaluation index and feature selection method. This approach is innovative and purposeful, aiming to enhance the predictive ability, construction efficiency, and quality. This method can provide support for decision-making and optimization of engineering projects and promote sustainable development of the project. After the construction was completed, the model was established, and the results were predicted. The actual engineering measurement data of Pier Two was taken for comparison with it. Two parameters, Root-Mean-Square Error (RMSE) and Linear Curvature (R2), were introduced to evaluate the prediction results, and the results were subjected to Correlation-based Feature Selection (CFS). The test sets for the downstream and the upstream tunnel were extracted, in which R2 for the three extracted comparisons of the downstream were 0.83, 0.82, and 0.89, respectively, while R2 for the upstream was 0.88, 0.86, and 0.86, respectively. From this, it can be seen that the optimized model has good predictive performance. In the construction process of other projects, the model can be used to predict the vertical displacement of bridge piers, which has real-time performance in preventing accidents.

Keywords

• particle swarm optimization - random forest (pso-rf) model
• tunnel
• bayesian
• pier
• displacement
• tunnel boring machine (tbm)