A case study of tunnel boring machines advance rate prediction using meta-heuristic techniques

Abstract

The advance rate (AR) of tunnel boring machines (TBMs) plays a pivotal role in evaluating their efficiency in tunnel engineering projects. This study focuses on the development of precise prediction models for TBM performance employing advanced algorithms, including gene expression programming, time series analysis, multivariate regression, artificial neural networks, particle aggregation algorithms, genetic algorithms, adaptive neural fuzzy inference systems, and support vector machines. The AR, serving as a performance metric, becomes the specific target for prediction models. A test database comprising 3597 datasets was curated from a tunneling project at the Sar Pol Zahab, Bazi Daraz water transfer tunnel. Utilizing 21 parameters as input variables, intelligent AR models were formulated based on comprehensive training and testing patterns, incorporating geological features and the key machine parameters influencing AR. Quantitative evaluation of the models involved statistical indicators such as root mean square error (RMSE), coefficient of determination (R²), and variance calculation. Comparative analysis based on RMSE, MAE, MAPE, VAF, and R² superior gene expression function models showed that the gene expression algorithm with 1.41, 0.66, 6.33, 98.88, and 0.95 ahead of the nose is better than other approaches. These results underscore the efficacy of the gene expression programming-based model, suggesting its potential to yield a novel functional equation for accurate TBM performance prediction.