Time Series Prediction on Settlement of Metro Tunnels Adjacent to Deep Foundation Pit by Clustering Monitoring Data

Abstract

High requirements are put forward for the settlement control of metro tunnel to ensure the normal and safe operation of adjacent metro line during the process of deep foundation pit construction. Monitoring and predicting could constantly monitor the settlement of the tunnel and make safety early-warning, and massive data to be processed is collected by sensors in this process. In the study, an improved clustering method based on Gaussian mixture model (GMM) is proposed to deal with a large amount of monitoring data. Four initial eigenvalues are defined and the initial core points of clustering are selected by grouping monitoring sensors based on the characteristics of the project site and sensors. An improved method is utilized to the metro tunnel of Metro Line 9 near Xujiahui station. Compared with the traditional clustering method, the improved method has more reliable results, and reduces the operation time by 57.9%. Representative monitoring sensors are selected from each cluster to predict based on Long Short-Term Memory (LSTM) neural network. The prediction results well agree with the measured value and the prediction accuracy is reaching to 99.3%. Compared with other sensor selection ways, the data of representative sensors exhibits good representativeness and effectiveness. Finally, the prediction result after data update is more consistent with the monitoring data than the prediction result without data update. Increasing the data update frequency improves the accuracy of the prediction results in practical engineering application.