Study on seawater metal corrosion modeling based on Partial Least-Square Regression

The seawater environmental factors are high dimensional, with strong correlations, while the metal corrosion data are small sample. It makes metal seawater corrosion modeling difficult. The scheme of metal seawater corrosion modeling is given out based on Partial Least-Squares Regression. Adapting small sample, the multi-input, multi-output models for more than 20 metals are established to describe the metal seawater corrosive behavior comprehensively. In purpose of reduction the error of certain outputs individually, an improved components extracting principle is proposed, which combines the cross validation with individual relative percentage error judgment. It effectively reduces the error of certain appointed outputs to meet the engineering precision requirement. The modeling procedure facing small sample is simple and convenient, effective, error controllable individually. It could provide the balance between the modeling precision and model prediction accuracy. Therefore it is applicable for metal seawater corrosion modeling and the other situation similar.