Using Genetic Programming to Predict Plastic Strain in Subgrade Soils Under Repeated Loading

Accumulated plastic strain in subgrade soils under repeated loading is an important factor for the design and analysis of flexible pavement. Accurate prediction of the plastic strain is dependent on the constitution of a predictive model. This study discusses the plastic strain properties in subgrade soil under repeated loading and develops a predictive model using a genetic programming (GP) method. Repeated load triaxial tests were conducted under various confining pressures, stress levels, and moisture contents. To develop the predictive model, a total of 475 records were randomly divided into three datasets (159, 158, and 158 records) for training, validation, and application purposes. The results showed that the GP method was applied successfully to develop a predictive model that uses stress level, load repetition, dry unit weight, liquid limit, plastic index, and clay content as inputs and has plastic strain as the output. The R² values for the training, validation, and applied datasets were 0.9551, 0.9494, and 0.9489, respectively. This study demonstrates and confirms the superior performance of GP over traditional prediction methods when applied to predict plastic strain in subgrade soils. The model and algorithms proposed in this study provide a good foundation for further development when more robust datasets of other types of soils are available.