利用多相动态腐蚀检测装置测定不同工况下油气水集输管道的均匀腐蚀速率，以此结果为样本建立径向基函数人工神经网络预测模型，并对管道均匀腐蚀缺陷进行评价。结果表明：当以硫化氢含量、二氧化碳含量、水含量、钙离子含量、镁离子含量、氯离子含量、温度、压力、流速等为输入信号，以管道均匀腐蚀速率为输出信号时，9-29-1型径向基函数人工神经网络结构合理且准确度良好，经过4 450次迭代后，预测模型的均方误差为0.000 9，小于允许收敛误差限0.001 0，在训练阶段、验证阶段与测试阶段线性拟合的决定系数分别为0.993、0.973、0.969，预测值和期望值具有较高相关性，同时还依据该模型找出了9段存在均匀腐蚀风险的管道。

The uniform corrosion rates of oil-gas-water gathering pipeline under different working conditions were measured by a multiphase dynamic corrosion detection device. A radial basis function artificial neural network prediction model was established using the data from above mentioned experiment as samples. The average corrosion defects of the pipeline were also evaluated. The results show that 9-29-1 radial basis artificial neural network had a reasonable structure and good accuracy when hydrogen sulfide content, carbon dioxide content, water content, calcium ion content, magnesium ion content, chloride ion content, temperature, pressure and flow rate were used as input signals and uniform corrosion rate was used as output signal. A mean square error of 0.000 9 lower than the specific convergence tolerance of 0.001 0 was obtained after 4 450 iterations. The coefficients of determination of linear fitting in training, verification and testing stages were 0.993, 0.973 and 0.969, respectively, demonstrating high relevance between the predicted values and the desired values. Meanwhile, according to the model, 9 sections of pipelines with uniform corrosion risk were found.