Abstract: At present, the data regarding oil and gas pipeline corrosion rate prediction collected in oilfields is characterized by a limited number of samples, a limited variety of types, and the absence of linear regularity. In light of these characteristics, this paper intends to adopt an improved approach for accurately predicting the corrosion rate of oil and gas pipelines. Firstly, variational mode decomposition (VMD) is utilized to decompose various pipeline corrosion factors. This method decomposes the limited pipeline corrosion factors into new variables with distinct features, thereby making effective use of the collected data. Subsequently, kernel principal component analysis (KPCA) is employed to reduce the dimensionality of the decomposed new variables of the corrosion factors, a minimizing data
redundancy. Finally, the GM (1, N) power model with structural enhancements and optimized background - value calculation is used for modeling and prediction. The prediction results of the VMD - KPCA - OGM (1, N) are compared with those of the GRA - OGM (1, N), VMD - OGM (1, N), and GRA - OGM (1, N) - GABP. The results show that VMD addresses the end effect and mode mixing issues of the EMD during the decomposition of pipeline corrosion rate prediction data, enabling more effective extraction of each principal component. Simultaneously, compared to the PCA, the KPCA exhibits a more pronounced dimension - reduction effect on nonlinear data. The prediction model established by integrating these three methods demonstrates a higher training fitting degree and prediction
accuracy for the corrosion rate prediction data than other models. Its Mean Absolute Percentage Error (MAPE) is 1.4856%, the lowest among the four models. Consequently, this model proves to be an effective means of reflecting the variations in the corrosion rate of pipelines.

Key words: Variational mode decomposition, Kernel principal component analysis, Optimized GM (1, N) power model, Oil and gas pipelines, Corrosion rate prediction