A Novel Interpolation Method Based on Differential Evolution-Simplex Algorithm Optimized Parameters for Support Vector Regression

Support Vector Machine (SVM) is a machine learning method based on Structual Risk Minimization (SRM). In traditional SVM, different selections of hyper-parameters have a significant effect on the forecast performance. Differential evolution algorithm (DE) is a rapid evolutionary algorithm based on the real-code, which can avoid the local optimization by the differential mutation operation between individuals. Simplex searching algorithm is a direct searching algorithm which solves nonconstraint nonlinear programming problems. This paper introduces the application of SVM in the spatial interpolation in geosciences field. It proposes a new method of optimizing parameters of SVM based on DE algorithm and simplex algorithm. Firstly, DE algorithm is used to obtain the initial value of simplex algorithm, then the simplex local searching strategy is applied to optimize SVR parameters for the second time. Moreover, the spatial interpolation simulation is conducted on the standard dataset of SIC2004. The case study illustrates that the proposed algorithm has higher forecast accuracy and proves the validity of the method.