Small sample prediction of external corrosion rates of buried pipelines

To address the shortcomings of existing linear regression models, single-support vector machines and genetic algorithm-supported vector machines in predicting pipeline corrosion rate, the sparrow search slgorithm-supported vector machine (SSA-SVM) corrosion rate prediction model was developed by using sparrow search algorithm and selecting dominant factors including total salinity, oxidation-reduction potential, pH value, chloride ion concentration, nitrate radical concentration, dissolved oxygen content and natural corrosion potential as input variations. SSA-SVM corrosion rate prediction model has a coefficient of determination R₂ of 0.9919, which is higher than that of the linear regression model (0.7189), single-support vector machine (0.8442) and GA-SVM (0.9137). The root-mean-square error is 0.0686 mm/a, which is lower than 0.1166, 1.7745 and 0.1183 mm/a, the values of the other three models. The average absolute error is 0.0902 mm/a, which is lower than 0.1474, 1.7056 and 0.0977 mm/a, the values of the other three models.. The average relative error is 3.94%, which is lower than 25.59%, 32.29% and 6.42%, the percentage of the other three models.Using this model, 8 sets of data were chosen at random to predict the external corrosion rate of buried pipeline B. Compared to real annual corrosion rate on site, the prediction accuracy was 0.9642, higher than 0.6690, the prediction accuracy of the genetic algorithm-supported vector machine, indicating that the model can be used to predict the external corrosion rate of underground pipelines in oil fields, offering data support for safe operation of the buried pipelines.