Intelligent information-based construction in tunnel engineering based on the GA and CCGPR coupled algorithm. (June 2019)

• Record Type: Journal Article
• Authors: Liu, Kaiyun; Liu, Baoguo
• Is Part Of: Tunnelling and underground space technology. Volume 88(2019)
• Page Start: 113
• Page End: 128
• Publication Date: 2019-06
• Subjects: Tunnel engineering -- Information-based construction -- Gaussian process regression -- Genetic algorithm -- Displacement prediction -- Optimization of preliminary design; Tunneling -- Periodicals; Underground construction -- Periodicals; Tunnels -- Periodicals; Underground areas -- Periodicals; 624.193
• DOI: 10.1016/j.tust.2019.02.012
• Languages: English
• ISSNs: 0886-7798

Abstract

Highlights: A genetic algorithm and combined covariance Gauss process regression coupled algorithm (GA-CCGPR) is proposed. A new displacement back analysis method based on GA-CCGPR algorithm was developed. A new prediction method of long-term displacement based on GA-CCGPR algorithm was developed. An optimization method of preliminary design based on GA-CCGPR algorithm was proposed. A novel information-based construction method of tunnel based on GA-CCGPR algorithm is proposed. Abstract: Based on the construction of the Beikou tunnel, a genetic algorithm (GA) and combined covariance Gaussian process regression (CCGPR) coupled algorithm (GA-CCGPR) were introduced for information-based construction in tunnel engineering. In the initial monitoring period, GA-CCGPR algorithm is used to execute displacement back analysis and predict displacement of surrounding rock; after entering the long-term monitoring period, GA-CCGPR algorithm for prediction of surrounding rock displacement is established by training with the measured displacement data directly. One GA and support vector regression (SVR) coupled algorithm (GA-SVR) is introduced in this study for comparison with the GA-CCGPR algorithm. The application results of the Beikou tunnel show that the maximum relative error and maximum average relative error of displacement prediction for 3 continuous excavation steps based on the GA-CCGPR algorithm are 14.13% and 11.26% in the stage of inversion prediction, respectively. Correspondingly, these two indexes of GA-SVR algorithm reach as high as 34.78% and 23.52%, respectively. The displacement prediction results of the surrounding rock indicate that the maximum relative error and maximum average relative error of the GA-CCGPR algorithm are only 10.64% and 3.59% in long-term monitoring period, respectively; however, these two indexes based on the GA-SVR algorithm are 46.3% and 6.18%, respectively. In addition, the computational time of the SVR algorithm is 3–4 times higher than that of the CCGPR algorithm in both the sample training and parameter identification in initial monitoring period. Similarly, the time required for the SVR algorithm to complete the sample training process is 3–4 times that of the CCGPR algorithm in long-term monitoring period. Finally, an optimization method for the preliminary support parameters was proposed based on the GA and CCGPR coupled algorithm presented in this paper to form a complete information-based construction method for tunnel engineering. With the advantages of fast processing, simple operation and high accuracy, this method can be widely used in tunnel engineering and similar applications.