Prediction and Reliability Analysis of the Pressuremeter Modulus of the Deep Overburden in Hydraulic Engineering Based on Machine Learning and Physical Mechanisms

Hanyu Guo, Deshan Cui, Qingchun Li, Qiong Chen, Lin Lai

In the process of large-scale water conservancy and hydropower station construction in the southwest region of China, obtaining the deep overburden pressuremeter modulus Em is of great significance for the calculation of foundation bearing capacity and dam foundation settlement. However, due to the complex nature of the soil properties in deep overburden layers, conducting deep-hole pressuremeter tests is challenging, time-consuming, and costly. In order to efficiently and accurately obtain the pressuremeter modulus of deep overburden, this paper takes the deep overburden in the river valley where a large hydropower station dam is located in the southwest region as the research object. It proposes a method based on data-driven prediction of the pressuremeter modulus and combines it with the physical mechanism to carry out the reliability analysis of the prediction results. By constructing a database of soil physical and mechanical parameters, including the pressuremeter modulus, the prediction performance of Random Forest (RF), Support Vector Regression (SVR), and BP Neural Network on the pressure modulus was evaluated. The Particle Swarm Optimization (PSO) was utilized for hyperparameter optimization to enhance the reliability of prediction results. The results indicate that the RF and PSO-RF models exhibit a comprehensive advantage for accurately predicting the pressuremeter modulus. The prediction results of the model for new data have a strong correlation with the results calculated by the Menard formula, which demonstrates the reliability of the model. Therefore, establishing the relationship between the conventional physical and mechanical parameters of deep overburden and the pressuremeter modulus, and predicting the pressuremeter modulus based on data-driven methods, has significant engineering value for obtaining the pressuremeter modulus of deep overburden efficiently, economically, and reliably. It also holds significant importance for the extended application of machine learning in the field of soil parameter prediction.