Abstract: Compared with steel bars, FRP bars have excellent properties such as corrosion resistance, low magnetic properties and fatigue resistance, making their applications in unfavorable engineering environments more and more common. However, the current predictive models for the bond behavior between FRP bars and concrete exhibit limited applicability and low predictive accuracy. Therefore, this paper collected 170 sets of hinged beam test data. Utilizing particle swarm optimization (PSO) to enhance the random forest (RF)algorithm, a model for predicting the bond strength between FRP bars and concretewas developed. The PSO-RF model was compared with the RF model and five existing models, and the predictive performance of these models was evaluated using statistical indicators such as the coefficient of determination R² and the mean absolute error MAE. The PSO-RF model demonstrated an R² value of 0.939 6 and an MAE value of 1.014 3, which can provide a valuable reference for the applications of FRP bars in concrete. Compared with the existing models, the R² and MAE values of the PSO-RF model were improved by 141.9% and 81.3%, respectively. The results of the analysis on parameter importance within the model indicated that bond length and compressive strength of concrete are two significant factors influencing the model’s predictive outcomes, with importance coefficients of 22.35% and 18.3%, respectively.

Key words: bond strength, FRP bars, concrete, particle swarm optimization, random forest, composites