This is method is recommended for the low resolution data with the resolution larger than 1.2 A and with no solution by other methods such as shelXT, XD, XS and superflip etc.

Simulated Annealing (SA), as well as similar procedures like grid search, Monte Carlo, parallel tempering, genetic algorithm, etc., involves the generation of a random sequence of trial structures starting from an appropriate 3D model. The parameters defining the model are modified until a good match between calculated and observed structure factors is found. The information about chemical knowledge of molecules is actively used to reduce the number of parameters to be changed: bond distances and angles are usually known and kept fixed. Global parameters defining the orientation and the position of each fragment are varied during the procedure, together with internal parameters like torsion angles.

cited from https://www.ba.ic.cnr.it/softwareic/sir/simulated-annealing/