Abstract: In order to accurately predict the residual life of submarine oil and gas pipelines subject to corrosion, a GM-RBF neural network corrosion rate prediction model based on error compensation principle was built. Firstly, a gray model (GM) of corrosion rate was established, and the gray prediction value of corrosion rate was taken as the input to radial basis (RBF) neural network, and the residual was taken as the output. The neural network was trained to obtain the error compensator. The new grey prediction value was compensated, and the final prediction value of the corrosion rate was obtained. The annual corrosion depth was calculated according to the prediction results, and the residual service life of the pipeline was calculated in combination with the residual strength criterion. The effectiveness of the model was verified by using the data on a certain subsea pipeline take as an example. The results show that the relative error of single-use GM model prediction is 17.48%, the relative error predicted by GM-RBF model is 6.37%, and the remaining life of the pipeline is predicted to be 5.4 years, and that the GM-RBF model improves the prediction accuracy and can make a better description of the trend in the corrosion development.

Key words: seabed oil and gas pipelines, corrosion remaining life, error compensation, grey model-radial basis function (GM-RBF), corrosion rate