Abstract:Based on generalized Boltzman-Gibbs statistical theory,the paper adopted a new disturbed modelproducedbytemperature-dependentpseudoCauchy distribution. Choosing modified acceptance probability formula and setting memory in practical computation,the best results can be searched out by memory and the simulated annealing algorithm suitable for inversion of time-lapse seismic data can be acquired. Through analysis of relations between seismic data and models,we chose proper objective function and flow for inversion of time-lapse seismic data and achieved good results through theoretical models tests. The new inversion algorithm hasbetter anti-noiseability,good stabilityand high-precision for inversion results,better finishing the inversion for small impedance changes in timelapse seismic data.