Search

It is of great significance to accurately evaluate the end-bearing capacity of rock-socketed piles based on multiple parameters. This study presents a hybrid model coupling extreme gradient boosting (XGBoost) with Bayesian optimization (BO) method for predicting the end-bearing capacity of rock-socketed piles. 138 data samples collected from the literature were used to construct the model. Five parameters, unconfined compressive strength of intact rock σc, geological strength index GSI, pile length within the soil layer Hs, pile length within the rock layer Hr and pile diameter B, are used as the input variables. The BO-XGBoost model was compared with two empirical formulas, as well as random forest (RF), gene expression programming (GEP), back propagation neural network (BPNN) and group method of data handling (GMDH). The results show that the coefficient of determination, root mean squared error and mean absolute error of the hybrid BO-XGBoost model for all datasets are 0.963, 0.634 and 0.240, respectively, and the accuracy order of these models is BO-XGBoost > RF > GEP > BPNN > GMDH > empirical formulas. In addition, the sensitivity analysis results show that σc is the most important parameter for predicting the end-bearing capacity of rock-socketed piles. Highlights: Extreme gradient boosting approach is used for predicting the end-bearing capacity of rock-socketed piles. Bayesian optimization is used to optimize the hyperparameters of the extreme gradient boosting model. Compare the proposed model with two empirical formulas as well as four other artificial intelligence models. [ABSTRACT FROM AUTHOR]

Shear strength parameters play a crucial role in the design and construction of rock slopes, underground openings, tunnels, excavations, and foundations. Determination of cohesion (c) and angle of internal friction (φ) through laboratory tests (e.g., triaxial tests) and in situ tests is time-consuming and expensive; therefore, it is valuable to accurately predict the shear strength parameters of rock materials using less expensive and more reliable methods. In this study, 4 artificial intelligence models, namely group method of data handling (GMDH), gene expression programming (GEP), bidirectional long short-term memory network (BILSTM), and random forest (RF), are proposed to predict the rock shear strength parameters c and φ. A database of 199 sets of experimental data from 4 rock types was used to construct the proposed models. The results show that the accuracy of the RF model is superior to the other three models, with coefficients of determination of 0.9933 (c) and 0.9727 (φ) on all datasets, respectively. To further improve the prediction accuracy of the RF model, six metaheuristic algorithms, namely particle swarm optimization (PSO), bald eagle search (BES), marine predators algorithm (MPA), northern goshawk optimization (NGO), golden jackal optimization (GJO), and dung beetle optimizer (DBO), were used to optimize the hyperparameters of the model. The results show that the accuracy of the six hybrid RF models is higher than that of the single RF model. Among the 6 hybrid RF models, the hybrid DBO–RF model is superior to the other hybrid RF models. In addition, the factors affecting the shear strength of rock materials are analyzed through the parameter sensitivity analysis. Highlights: Four artificial intelligence models are used to predict the shear strength parameters of rock materials. Six metaheuristic algorithms are used to optimize the hyperparameters of the random forest model. The prediction performance of the six hybrid random forest models is compared. The factors influencing the shear strength of rock materials are analyzed by parameter sensitivity analysis. [ABSTRACT FROM AUTHOR]

Fiber-reinforced polymer (FRP) is an excellent building material for strengthening concrete structures, but it is difficult to accurately evaluate the bond strength of FRP-to-concrete due to the influence of various parameters. In this study, a novel hybrid model which combines particle swarm optimization (PSO) with random forest (RF) was proposed to predict the bond strength of FRP-to-concrete. The PSO algorithm was used to optimize the hyperparameters of the RF model. A total of 749 specimens collected from the literature were used to develop the proposed PSO-RF model. Each sample contains 11 parameters required for the model. These 11 parameters are (1) the compressive strength of concrete, (2) the tensile strength of concrete, (3) the width of concrete specimen, (4) the maximum aggregate size of concrete, (5) the tensile strength of FRP, (6) the thickness of FRP, (7) the elastic modulus of FRP, (8) the tensile strength of adhesive, (9) the bond length of FRP, (10) the bond width of FRP, and (11) the bond strength of FRP-to-concrete. The proposed PSO-RF model was compared with other machine learning models as well as ten empirical equations. Six statistical indices, namely root mean squared error (RMSE), mean absolute error (MAE), coefficient of determination (R2), Nash-Sutcliffe efficiency coefficient (NSE), Willmott's Index of Agreement (WIA), and Legates-McCabe's Index (LM) were used to evaluate the prediction performance of the abovementioned models. The results show that the RMSE, MAE, R2, NSE, WIA and LM values of the PSO-RF model are 1.529 kN, 0.942 kN, 0.986, 0.984, 0.996 and 0.892, respectively, for the training datasets and 2.672 kN, 1.967 kN, 0.963, 0.961, 0.989 and 0.761, respectively, for the test datasets. It can be concluded that the proposed PSO-RF model has the best comprehensive performance in predicting the bond strength of FRP-to-concrete. In addition, the sensitivity analysis of the PSO-RF model was also conducted in this study. [ABSTRACT FROM AUTHOR]

In this study, three artificial intelligence models, namely group method of data handling, gene expression programming and random forest, are proposed to predict the ultimate bearing capacity of concrete filled steel tube stub columns. A total of 220 data samples collected from the literature was used to construct the three models. Five statistical indices were used to evaluate the performance of the three models and the other five existing design codes and two reference models. Compared with the optimal model among the seven existing models, the coefficient of variation, mean absolute percentage error, root relative squared error and integral of absolute error values of all datasets of the three models (i.e., group method of data handling, gene expression programming and random forest) were decreased by 46.47%, 56.49%, 71.35% and 65.78%; 49.82%, 59.49%, 66.88% and 64.51%; 79.27%, 84.83%, 78.64% and 83.71%, respectively; while the determination coefficient values of all datasets of the three models were increased by 7.36%, 6.79% and 8.22%, respectively. The results show that the predicted results of the three models agree well with the experimental results, and the random forest model has the best comprehensive performance. [ABSTRACT FROM AUTHOR]

The creep index plays an important role in calculating the long-term settlement of natural soft clays, so it is vital to determine the creep index quickly and accurately. However, the prediction accuracy of the existing creep index models is low. This study presents seven gene expression programming (GEP) models by using different combinations of the liquid limit wL, plasticity index Ip, void ratio e and clay content CI as input variables for the prediction of creep index. A total of 151 datasets were collected from the available literature for building and testing the GEP models. The proposed GEP models were compared with two machine learning (ML) models (i.e., back propagation neural network and random forest) and five conventional empirical models in terms of three statistical indicators. The research results showed that the prediction performances of the two proposed GEP models (i.e., with combinations C I - w L - e and C I - I p - w L - e as input, respectively) surpass those of the five conventional empirical models and two ML-based models, recommended for predicting the creep index of natural soft clays in engineering practice. [ABSTRACT FROM AUTHOR]

Accurate predictions of significant wave heights are crucial in the field of marine and ocean engineering. This paper presents two hybrid intelligent models, EN-1 and EN-2, that combine random forest (RF) and extreme learning machine (ELM) methods through direct linear weighting and gene expression programming-based nonlinear weighting methods, respectively, for the prediction of weekly mean significant wave heights. The performances of the EN-1 and EN-2 models were compared with those of 11 reference models. The results showed that the coefficient of determination (R2), mean absolute error (MAE), root mean squared error (RMSE) and mean absolute percentage error (MAPE) of the EN-1 and EN-2 models were 0.9850, 0.0567, 0.0736 and 3.7107%, 0.9940, 0.0335, 0.0487 and 2.2373%, respectively, for the training datasets and 0.9802, 0.0627, 0.0847 and 4.1461%, 0.9754, 0.0707, 0.0966 and 4.2614%, respectively, for the testing datasets, indicating that the EN-1 and EN-2 models have good predictive performance and can be effectively used for the prediction of weekly mean significant wave heights. In addition, a sensitivity analysis was conducted to investigate the influence of the input variables on the model performance. • Two hybrid models are proposed for the prediction of weekly mean significant wave heights. • The differences of the two weighting methods in improving the performance of the model are compared. • Sensitivity analysis is conducted to investigate the influence of input variables on the model performance. [ABSTRACT FROM AUTHOR]

• Three AI models for predicting the soil thermal conductivity are proposed. • Compared the three AI models with five existing models. • A web page program for the proposed models was built. In this study, three artificial intelligence models, namely group method of data handing (GMDH), multi expression programming (MEP) and random forest (RF), are proposed to predict soil thermal conductivity. A large database of 444 data samples collected from existing literature was used to construct the three models. Three statistical indexes were used to evaluate the performance of the proposed three models and the other five existing models. The results show that the predicted results of the three models are in good agreement with the experimental results, and the RF model has the best comprehensive performance. Therefore, RF model is recommended to predict the thermal conductivity of unsaturated soil. To facilitate practical engineering applications, we have built a web program for the three proposed models, and users can use these three models by visiting the website. [ABSTRACT FROM AUTHOR]

A low deviation creep coefficient prediction model is vital for engineering practice related to soft clays. However, current empirical methods for calculating creep coefficients are not sufficiently reliable. This study presents a new hybrid intelligent model that couples random forest (RF) and support vector machine (SVM) by weighting for prediction of creep coefficient of soft clays. The genetic algorithm (GA) is used to search the optimum hyperparameters of the proposed model. A total of 151 datasets collected from the literature were used to construct the proposed model. Limited by the size of the dataset, k-fold cross-validation is used in the optimization processes. The obtained results show that the forecasting performances of the proposed model surpass those of the empirical methods, recommended for the prediction of creep coefficient in practical engineering. In addition, for the convenience of practical engineering application, we have built a web page program for the proposed hybrid model, and the users can use the model by visiting the website. • A hybrid intelligent model coupling random forest and support vector machine was proposed to predict the creep index. • Genetic algorithm was used to search the optimum hyperparameters of the proposed models. • Robustness and parametric analysis were conducted to verify the reasonableness of the hybrid model. [ABSTRACT FROM AUTHOR]

This paper presents two artificial intelligence techniques (e.g., gene expression programming (GEP) and random forest (RF)) for predicting the bond strength of near-surface-mounted (NSM) fiber-reinforced polymer (FRP) strips or rods bonded to concrete. Experimental data from 145 direct pullout tests collected from the literature and five parameters, namely, the bond length, FRP axial rigidity, groove depth-to-width ratio, epoxy tensile strength, and concrete compressive strength, were used to develop the GEP and RF models. A comparison was conducted between the proposed GEP and RF models and two existing empirical models, namely, Seracino's model and Zhang's model, and six statistical indices were used to evaluate the performance of these four models. The results show that the proposed GEP and RF models had higher coefficient of determination (R2) values and lower root mean squared error (RMSE), mean absolute error (MAE), root relative squared error (RRSE), mean absolute percentage error (MAPE), and integral absolute error (IAE) values than the two existing empirical models. Finally, a detailed parametric study was conducted to investigate the influence of each input variable on the bond strength. The results showed that the bond strength increased with increasing bond length, FRP axial rigidity, groove depth-to-width ratio, and concrete compressive strength, while the epoxy tensile strength had little effect on the bond strength. [ABSTRACT FROM AUTHOR]