The improvement of sobol´ sensitivity analysis method

Sobol´ sensitivity analysis method is widely used as a classical approach in global sensitivity analysis. this paper improves the integral used for calculating conditional variance, and proves that it can reduce error in the view of Monte Carlo method; besides, Sobol´ low discrepancy sequence is replaced by another quasi-random numbers with lower star discrepancy generated by uniform design to increase accuracy of Quasi-Monte Carlo method. A numerical example verifies that both measures can effectively improve accuracy of estimation of sensitivity index for Quasi-Monte Carlo method.