Prediction Model of Material Corrosion Rate in Soil Based on PCA and AGO-SVM

Fu Dongmei¹,Fu Zhenze¹,Liu Deyong¹,Li Xiaogang＊^2,3

(1、School of Information and Engineering, University of Science and Technology Beijing, Beijing, 100083,China)

(2、School of Materials Science and Technology, University of Science and Technology Beijing, Beijing 100083, China)

(3、Beijing Key Laboratory for Corrosion, Erosion and Surface Technology , Beijing 100083，China)

Abstract: Support vector machine (SVM), which is a novel machine learning method that based on statistical learning theories, is a more effective way in small-sample-data analysis than traditional statistical methods. However, it often fails when the data are correlated and with high dimensions. In soil environment, due to the consideration of the experimental cost and environmental conditions, samples of material corrosion data are limited and affected by many factors. These features determined that soil corrosion data was typical small sample and high dimensional data, and they were strongly correlated with each other, thus SVM was not proper to be used directly in the modeling. This paper preprocessed the corrosion data with principal component analysis (PCA) method, by which the dimension of the input samples was reduced from 13 to five without the large loss of data information. In addition, for the reason that material corrosion data was characterized with time sequence, an AGO-SVM modeling method was postulated, which integrates the accumulated generating operation (AGO) with SVM method. This method abbreviated the influence of random disturbance factors resulting from original data, and revealed the principle of discrete data, which enhanced the regularity of data. Resultantly, a new sequence with a monotonic increase trend can be achieved, which was also convenient for SVM learning. Developed by this method, soil corrosion rate forecasting model was established, which preserved a better prediction effect than SVM.

Key words: SVM，PCA，AGO，prediction model

* Contract Email: lixiaogang99@263.net

** Acknowledgements

The authors are grateful for the financial support from the National R&D Infrastructure and Facility Development Program of China (registration number: 2005DKA10400), from the National natural science fund project, No: 50499336-10