Abstract: Aiming at the frequent occurrence of CO₂ corrosion in submarine multiphase flow pipelines and the difficulty of detection, an extreme learning machine(ELM) prediction model based on lasso regression(LASSO) and multi-group beetle beetle optimization algorithm(MBAS) was established to improve the forecasting efficiency and forecasting accuracy.LASSO regression screens corrosion influencing factors, extracts key indicators, and reduces the predictive input dimension; MBAS was used to revise ELM input weights and hidden layer thresholds to avoid instability caused by random settings.Taking 50 sets of data from a submarine oil and gas pipeline in eastern Hainan as an example, MATLAB simulation was used to analyze the prediction results and compare with the other two models.The results show that temperature, pH value, fluid flow rate and CO₂ partial pressure are the key factors affecting the corrosion of this type of pipeline.The prediction results of the LASSO-MBAS-ELM model have a higher degree of fit with the actual value.The root mean square error, average absolute error and average absolute percentage error are 0.089%, 0.079% and 3.068%, respectively, which are better than the comparison model.The proposed method has good reliability and stability even with limited data.It provides a new idea for accurately grasping the corrosion status of submarine pipelines, and at the same time provides a reference basis for the daily operation and maintenance of marine oil and gas transportation systems.