Due to the torque characteristics of the propeller during rotation, the unmanned underwater vehicle (UUV) will produce a roll effect, which is unfavourable for UUV that require precise detection. How to design a control algorithm to weaken or even eliminate the roll effect is a topic worthy of study. This paper first analyzes the roll effect of the UUV, designs a rudder force distribution strategy for the cross-shaped structure, and proposes a model-free adaptive roll suppression based on the online/offline I/O data-driven reorganization state iterative learning (IL-MFAC) method. This method consists of iterative learning control law, parameter updating law and reset criterion. It is designed only using the online/offline I/O parameters of the controlled system and is a typical data-driven control algorithm. Finally, through numerical simulation, the effectiveness and superiority of this algorithm in UUV roll suppression were confirmed.