Forecasting failure load of Sandstone under different Freezing-Thawing cycles using Gaussian process regression method and grey wolf optimization algorithm

Abstract

The stability analysis of rock is an important basis to ensure the safe exploitation of underground resources and the reliable operation of space engineering. Failure load is one of the most important parameters on which the stability analysis of rocks can be performed. Rocks are constantly subjected to the freezing-thawing (F-T) weathering phenomena. F-T cycles can significantly affect the failure load of rocks and cause geological disasters in cold regions. As a result, the effects of F-T cycles on rock mechanical characteristics should be carefully investigated. Weak, heavily fractured, worn, and thinly bedded rocks, on the other hand, can never provide enough high-quality cores. Aside from being difficult, time-consuming, and costly, this test also entails destructive testing. Therefore, the prediction of failure load under different F-T cycles utilizing simplified indirect test procedures has been emphasized. In this study, a Gaussian process regression (GPR) technique was developed to predict the failure load of rock samples, utilizing 300 datasets obtained from the laboratory tests for Sandstone under different F-T cycles (0,1,4,8,16,32). Seven input parameters effective on the failure load were considered in the datasets. The grey wolf optimization (GWO) algorithm was used to fine-tune the GPR’s hyperparameters. The evaluation indices results showed that the GPR model developed by this study has a high potential ability in the prediction of failure load. The stepwise method was used for feature selection purposes. Confining pressure and porosity parameters were selected as the most effective features on the failure load. Considering only the confining pressure and porosity parameters in the model, the accuracy of the model was increased as the RMSE and R2 were obtained by about 5.57 and 0.9885, respectively. Also, the performance of the GPR model was compared to the other machine learning methods. Finally, the GPR model was recommended to predict the failure load of rock samples. This work’s significance is that it allows geotechnical engineers to accurately estimate the failure load rock samples under F-T cycles. In this way, instead of preparing more samples and waiting a long time to prepare these samples for the same laboratory test, we can use the trained model and minimize the time and costs required for laboratory tests.