Objective This study aims to enhance the prediction accuracy of pipeline corrosion rates, evaluate the residual strength and remaining life of pipelines, and develop targeted anti-corrosion strategies.

Methods A corrosion rate prediction model was established on the basis of Ensemble Empirical Mode Decomposition (EEMD)-Least Absolute Shrinkage and Selection Operator (LASSO) and optimized First Order Multidimensional Grey Model i.e., GM(1, N). First, the EEMD algorithm was employed to enhance the diversity of the original independent variables, mitigate the impact of sequence fluctuations on prediction results, and decompose data information layer by layer based on frequency scaling. Next, the LASSO algorithm screened independent variables to minimize correlation and redundancy among sequences. Finally, linear correction and grey action were incorporated to optimize the GM(1, N) model by converting the differential equation into a difference equation. The input variables were then substituted into the optimized GM(1, N) model for training and prediction.

Results In the case presented in this paper, the weight of influence related to CO₂ partial pressure and the amount of sulfate-reducing bacteria (SRB) was greater, with the corrosion process primarily controlled by CO₂. A total of eight independent variables, including decomposition and residual variables, were screened out. Compared with the influencing factors identified by the grey correlation method, the LASSO algorithm offered a more objective and scientific approach. The EEMD-LASSO-GM(1, N) model achieved an average relative error of 1.25％ and a standard deviation of 0.72, surpassing the accuracy of traditional GM(1, N), EEMD-traditional GM(1, N), EEMD-optimized GM(1, N), and EEMD-principal component analysis (PCA)-optimized GM(1, N). Furthermore, the EEMD-LASSO-GM(1, N) model demonstrated the highest prediction accuracy and superior generalization and robustness compared to literature data. It also exhibited good adaptability to variations in corrosion influencing factors and data quantities.

Conclusion The research findings offer a theoretical foundation and practical reference for predicting multi-factor coupled pipeline corrosion behavior and corrosion rate development trends.