Accurate prediction of corrosion rates is of great significance for ensuring the safe operation of CO₂ transportation pipelines. In view of the current situation where there is a lack of a set of corrosion prediction models suitable for multi-impurities and different CO₂ transportation phases, a corrosion prediction model based on a multi-strategy improved dung beetle optimizer (MIDBO) and support vector regression (SVR) was proposed. To solve problems such as the slow convergence speed and imbalanced global exploration and local exploitation capabilities of the DBO algorithm, a multi-strategy improvement scheme was introduced: at the initial stage of the algorithm, Tent chaotic mapping and opposition-based learning strategies were integrated to initialize the population, enhancing population diversity and convergence speed; during the dung beetle breeding phase, a dynamic spiral search strategy was employed to improve the global search capability; in the later stage of the algorithm, Lévy flight and greedy strategies were combined to perturb the population and update positions, strengthening the algorithm’s ability to escape out of local optima. Simulation results show that, compared to the DBO algorithm and single-strategy improved DBO algorithms, the MIDBO algorithm exhibited superior optimization performance across various test functions, validating the effectiveness of the multi-strategy improvements. The MIDBO-SVR corrosion prediction model was compared with DBO-SVR, GWO-SVR, SSA-SVR, and POA-SVR models in terms of prediction accuracy. The results show that the MIDBO-SVR algorithm achieved mean absolute error (MAE), mean squared error (MSE), and mean absolute percentage error (MAPE) values of 0.010 0, 0.000 2, and 13.21, respectively, all of which were lower than those of the other four models. The MIDBO-SVR corrosion prediction model established demonstrated high prediction accuracy and strong practical significance, offering considerable potential for field applications.