Underwater Modular Self-Reconfigurable (UMSR) robots, made up of many identical modules, can adjust their configurations to multiple underwater environments or tasks. They can be used in many complex underwater occasions where the ROVs/AUVs can't work well. However, their reconfiguration is difficult because this has to involve many connection adding and removing operations which are difficult to be executed in the underwater environment. To reduce the times of these operations, we propose the theory of Topological Transformation (TT), which includes some definitions in TT, the basic approach to TT, and the Genetic Algorithm (GA) based solution for the optimal TT.