Sand Liquefaction Prediction by Using CatBoost Algorithm Combined with Optuna Framework
To solve the problem that some sand liquefaction prediction models built by using machine learning algorithms only achieved high accuracy in specific areas and had weak generalization ability,so as to expand applicability of the sand liquefaction prediction models and accurately predict sand liquefaction for better prevention and control of seismic hazards,a CatBoost-Optuna sand liquefaction prediction model was established on the basis of categorical feature boosting algorithm CatBoost combined with automatic hyper parameter optimization framework Optuna for parameter adjustment training.The seismic liquefaction dataset from the standard penetration test was divided into a training set and a test set,and prediction results of the established model were evaluated using five evaluation indexes.Evaluated results of multilayer perceptron and support vector machine sand liquefaction prediction models in the test set were compared,and the seismic liquefaction case data was used as a validation set to compare prediction effects of different prediction models.The results show that compared with multilayer perceptron and support vector machine sand liquefaction prediction models,the estab-lished model has larger evaluation indexes and better prediction effects in the test set.In the validation set,only pre-cision rate among evaluation indexes of the established model slightly decreases,and other evaluation indexes remain stable,while only recall rate among evaluation indexes of the comparison models remains stable,and other evaluation indexes decrease.Only prediction effects of the established model remain consistent with those in the test set,which further demonstrates superior generalization ability of the established model.
NETL
NSTL
万方数据
