Abstract: In order to improve the prediction accuracy of the corrosion rate of condensate gas field gathering pipelines, an internal corrosion prediction model based on GRA and RFR was constructed to analyze the causes of internal corrosion. The GRA was used to select characteristic variables as inputs of RFR, and the internal corrosion rate was used as target factor output. Taking the data of Yakela condensate gas field as an example, the PRF prediction model was verified by comparison with the Back Propagation(BP)neural network and Support Vecor Machine(SVM) prediction model. The results show that the main factors of corrosion in pipelines obtained by GRA are CO₂ volume fraction, Cl^-concentration, pressure, temperature and flow rate, that the root mean square error, average relative error of the RFR prediction model are lower than those of BP neural network and SVM prediction model, and that determination coefficient of RFR prediction model is 96.48%.

Key words: condensate gas field, corrosion factors, grey relational analysis (GRA), random forest regression (RFR), corrosion rate