The work aims to predict service life of pipes by building a prediction model of corrosion rate for metal pipes. Corrosion process of metal pipes under the effect of CO2 or H2S was analyzed, chemical reaction equation of pipe corrosion was given, a mathematical model of corrosion rate was built for metal pipes by using BP neural network, and the prediction model was optimized in the improved method of particle swarm optimization. Taking 45# metal pipe as an example, corrosion rate of the pipe was simulated and verified with the help of Matlab software, and was compared with the experimental measurements for analysis. The corrosion rate of metal pipes increased with the increase of CO2 or H2S pressure. The simulation results showed that the maximum corrosion rate of CO2 and H2S was 7.20×10-5 mm/h and 5.76×10-5 mm/h, respectively, while the experimental results showed the maximum corrosion rate of CO2 and H2S was 7.14×10-5 mm/h and 5.65×10-5 mm/h, respectively. The relative error caused by the improved BP neural network was less than 5%. For metal pipes under different pressure conditions, corrosion rate can predicted approximatively by using the improved BP neural network prediction model, which provide a reference basis for replacement of metal pipes.