目的 研究输油金属管道内腐蚀预测模型，以改善管道维修策略和提高使用期限。方法 分析输油管道内腐蚀原因，归纳腐蚀因素，采用主成分分析法对腐蚀因素进行优选，摒弃相关联但腐蚀贡献率较低的因素，以最大化表征腐蚀原因及减少不必要的处理过程。将贡献率较大的腐蚀因素作为支持向量机预测模型的输入变量，以腐蚀率作为目标输出，建立管道腐蚀预测模型。针对支持向量机参数选取问题，应用鲶鱼粒子群算法进行寻优，以提高预测精度。以20#钢管为例进行了模型验证，并与其他预测模型进行了对比和分析。结果 主成分分析筛选得到的管道内腐蚀的主要因素有：H2S、CO2、Cl?、酸碱值、压力、介质温度、流速。通过鲶鱼粒子群算法改进的支持向量机模型的预测与实际值的平均相对误差为2.82%，相关性系数值为0.9955，均优于其他三种预测模型。结论 金属管道的内腐蚀由多个腐蚀因素共同作用形成，采用主成分和粒子群优化的支持向量机模型能够精确预测内腐蚀率，对金属管道维修和管理的借鉴性高。

The work aims to improve the maintenance strategy and service life by studying the prediction model for inner-corrosion of metal pipelines. The causes and factors for inner-corrosion of pipeline were analyzed and summed up. The corrosion factors were filtered by the principal component analysis. Those factors related to each other but contributing to low corrosion were abandoned. The causes for corrosion were characterized at maximum and some unnecessary analysis processes were reduced. The factors contributing to large corrosion were used as the input variables to the support vector machine prediction model, and the corrosion rate was used as the target output to build the pipeline corrosion prediction model Aiming at parameter selection of support vector machines, catfish particle swarm optimization algorithm was applied to improve the prediction accuracy. 20# pipeline was taken as an example to verify the model. The model was contrasted and compared with other prediction models. The main factors for inner-corrosion of pipeline selected by the main component analysis were H2S, CO2, Cl?, pH, pressure, medium temperature and current velocity. The average relative error between the predicted value of support vector machine model improved by catfish particle swarm optimization algorithm and the actual value was 2.82% and the correlation coefficient value was 0.9955, which were better than those in the other three prediction models. The inner-corrosion of pipeline is formed by multiple corrosion factors, and the inner-corrosion rate can be predicted accurately by the support vector machine model adopting main component analysis and particle swarm optimization, which provide a reference for the maintenance and management of the metal pipelines.