Abstract: In order to extend back analysis means and improve the efficiency and accuracy, an improved particle swarm optimization(PSO) with variable inertia weight and synchronous changing learning factor was carried out and the program was completed, the advantage of the enhanced method was tested by Sphere and Rastrigrin functions. Then, a back analysis program GeoPSOInverse.m was developed by the software MATLAB, in which the finite element code of ABAQUS was embedded. At last, the parameters of creep constitutive model of a mudstone with irreversible strain as variable and implicit time in formula were back analysied. The results show that this improved PSO method is a very good inverse analysis method and its efficiency is quite good, and can be proceeded in practical application for complex engineering.