Research on collision avoidance for USV (unmanned surface vessel) is crucial for achieving intelligent ship navigation. In practical applications, the motion of USV is complex and highly nonlinear. This paper proposes a collision avoidance method based on MFAC (Model-Free Adaptive Control). Initially, the method assesses collision risk, historical heading, and rudder angle information of the USV. Based on this historical data, a control algorithm is established. Subsequently, the method determines the desired heading of the USV according to the collision risk. Finally, the USV is steered to avoid collision using the control algorithm based on the desired heading. In simulation, experiments were conducted on encounter, overtaking, and crossing situations. Results demonstrate that the collision avoidance control algorithm responds rapidly, adjusts heading and speed in real-time, and it has advantages over traditional artificial potential field methods in complex environments. Successfully navigates the USV to comply with the COLREGs (International Regulations for Preventing Collisions at Sea). This ensures safety and traffic efficiency in maritime navigation.