Support vector machine model for the relationship between physico-mechanical indexes and microstructure parameters of soft soil

摘要

Owing to peculiar geographical and geological conditions, the Pearl River Delta area is mainly made of quaternary soft clays. The mechanical behavior is largely controlled by the internal microstructure of soil. By means of measuring techniques such as Scanning Electron Microscope(SEM) and image processing technique, the paper sheds light on the microstructure characteristics of soft soil of Nansha area. Microstructure parameters such as porosity, total area of pores, number of pores, average area of pores, average diameter of pores and pore anisotropy have been obtained. Due to the nonlinear relationship between physico-mechanical indexes and microstructure parameters, Support Vector Machine (SVM), a new algorithm in the field of artificial intelligence, is introduced in this paper. SVM is based on Statistical Learning Theory and owns high generalization capability and can solve the problem of small samples, nonlinear, high dimensions and local minimization. The relationship between microstructure parameters and coefficient of compression, constrained modulus etc. are analyzed, the internal relationship between physico-mechanical indexes and microstructure parameters are established in the paper. Comparison was made between SVM method and Back Propagation neural networks (BP). The results indicate that SVM method can obtain high accuracy and generalization ability, and it can provide an important basis for geotechnical parameters analysis and choice of soft soil engineering.