Modeling of true triaxial strength of rocks based on optimized genetic programming

Abstract

The strength of a rock is the main factor affecting the stability of an engineered rock mass. As laboratory testing requires sophisticated equipment and considerable time to determine rock strength, prediction models are needed for establishing rock strength criteria. Genetic programming (GP) is a soft computing technology often used to address rock mechanics and engineering challenges. However, GP also has limitations, such as a long running time, complex individual growth without a corresponding fitness improvement, and difficulty in finding the optimal solution. Therefore, we conducted this study by applying a dynamic restriction on individual size, local search of the neighborhood of the optimal individual, and multithreaded evaluation to optimize GP and guarantee the accuracy of the results and to build a prediction model for the true triaxial strength involving different rock types. The results showed that the restriction dynamically changes to restrict the redundant bloat of strength individuals without a corresponding fitness improvement; using local search rules can effectively find individuals with high fitness, so the strength predicted by the system was in good agreement with the measured strength. We also found the predicted strength was suitable for fitting the rock strength criteria. Using this multithreaded evaluation sped up the operation of the algorithm and produced accurate predictions; and for complex problems, increasing the threads had a more pronounced effect on the runtime and fitness improvements. Based on the Sobol global sensitivity analysis, we analyzed the influence of each prediction parameter on the true triaxial strength of rocks. Combined with the statistical assessment indices involving sum of the absolute error, mean, a10-index, and regression determination coefficient, the predictions of the optimized GP model that we established in this study were more accurate than those of multiple regression analysis.