Abstract: Simulations of underground cavern group construction are important to analysis of the construction process. It is difficult to achieve high-accuracy simulation of slag transportation time by using traditional simulation models that suffer from problems such as large subjective errors in quantifying the influence of transportation machinery failure on construction progress. This study develops a new simulation model of underground cavern group construction to predict this failure using a M5P-SVR model. This model improves the accuracy in two aspects. Firstly, it improves the slag module adopted in the traditional CYCLONE model and uses a transportation simulation circuit to calculate slag transportation time with an improved accuracy. Secondly, it considers certain external factors more rationally, adopts the simple, effective training rules of the M5P model tree, and is equipped with the support vector machine regression (SVR) that can effectively solve those nonlinear prediction problems that are based on small sample dataset. Combining these two advantages, we develop the M5P-SVR model based on a prediction method of transportation machinery faults, and verify its significant improvement on prediction accuracy using a cross validation method. Finally, our new model is applied to the simulations of a real project and compared with the traditional method, verifying its accuracy and superiority and the effectiveness of M5P-SVR.

Key words: underground cavern group, construction simulation, M5P-SVR, machinery failure, slag transportation