首页>
外文期刊>Materials and structures
>Random forest-based modelling of parameters of fractional derivative concrete creep model with Bayesian optimization
【24h】
Random forest-based modelling of parameters of fractional derivative concrete creep model with Bayesian optimization
展开▼
机译：Random forest-based modelling of parameters of fractional derivative concrete creep model with Bayesian optimization
Abstract Although fractional derivative models (FDM) have been extensively studied in the calculation of concrete creep, all studies’ focus is only limited to the FDM parameters estimation under existing creep data. However, conducting tests on the creep of concrete with multiple influencing variables is time-consuming and costly. This paper presents a random forest (RF)-based model for the parameters (viscosity coefficient η and fractional derivative order α) of the FDMs based on the updated Infrastructure Technology Institute of Northwestern University (NU-ITI) database with 748 sets by the Levenberg–Marquardt method, and 10 influential variables in the database. Further, a Bayesian optimization-based random forest (BO-RF) model to more accurately and efficiently predict η and α of the FDM is also proposed. The results show that Bayesian optimization can effectively search for the optimal hyper-parameters of the RF model and achieve high prediction accuracy with determination coefficients 0.972 and 0.965 on the testing sets, respectively. Moreover, the statistical values of CEB coefficient of variation and CEB mean square error method for the FDM model based on BO-RF prediction results can achieve 13.03 and 3.96, respectively. Finally, by measuring the relative importance of the variables, it is observed that the loading time t0 has the highest importance for both η and α, followed by temperature for η and aggregate cement ratio for α.
展开▼
