Abstract: In view of the complex internal environment of reinforced concrete, there were many factors that affect the corrosion of steel bars, which led to the problem that the measurement accuracy of steel corrosion was not high. It was proposed to use the advantages of fast convergence characteristics and strong global search ability of the improved long-horned whisker algorithm (BAS) to initialize the weights and thresholds of the BP neural network. Based on the corrosion mechanism of steel bars, six influencing factors of concrete internal temperature, humidity, CO₂ content, pH, chloride ion content and corrosion potential were selected as input variables, and steel corrosion degree was selected as output variables, an IBAS-BP prediction model was established and the measured input datas were used for simulation verification. The results showed that as compared with the three models of BAS-BP, PSO-BP and GA-BP, IBAS-BP prediction model had a fast convergence rate, the maximum relative error of the test set was 0.025 and the coefficient of determination was 0.943 3.The IBAS-BP prediction model could more accurately predicte the corrosion degree of steel bars in reinforced concrete.