Prediction method for internal corrosion rate of gas pipeline based on RS-ISOAKELM model

[Objective] Surface gas transmission pipelines suffer serious internal corrosion. To ensure the service safety of pipelines, it is necessary to accurately predict their internal corrosion rate. [Methods] The main control factors that affect corrosion were screened based on the Rough Set(RS) theory, and the reconstructed data set was used as the input while the corrosion rate was used as the output to train the Kernel Based Extreme Learning Machine(KELM) model. Additionally, the Improved Seagull Optimization Algorithm(ISOA) was used to optimize the model hyperparameters, and a prediction method of internal corrosion rate based on the RS-ISOA-KELM model was proposed. The prediction precision was compared with that of other combination models, and the long-term prediction effect and model universality were analyzed. [Results] The convergence analysis for the ISOA algorithm was conducted on five benchmark functions including Sphere, Schaffer, Rosenbrock, Rastrigin, and Griewank, and it was found that the ISOA algorithm had better performance in terms of solution precision and calculation stability. The model was verified using actual operating data from a gas field block. For the selected data set, the temperature, CO2 partial pressure, H2S partial pressure, flow rate, Cl-content, moisture content, and corrosion inhibitor residual concentration were crucial factors that affect internal corrosion. Among them, H2S partial pressure, flow rate, and corrosion inhibitor residual concentration had the largest weight. The RS-ISOA-KELM model was used to predict the corrosion rate, and its average relative error was 1.498%, root mean square error was 0.002 1 mm/a, and coefficient of determination was 0.999 3, which were all better than other common comparison models. [Conclusion] The combined model has strong generalization performance and high prediction precision. By expanding and updating the original database,it can predict the medium-and long-term corrosion rate of pipelines.With different corrosion parameters,data sizes,and training test ratios,the model still maintains a good prediction effect.