To address the computational challenge known as "curse of dimensionality" inherent in traditional branch and bound algorithms for large-scale power grid unit commitment problem, an optimization method for unit commitment state iteration path search is proposed. To avoid the loss of the optimal solution caused by simplifying the problem and reducing the feasible region, the determination of the unit state scheme is divided into two stages: deep traverse and breadth iteration. On the basis of initial solution, the unit dynamic priority list is used as the search direction for the unit state iteration path. Through deep traverse, the optimal shutdown redundant unit and corresponding shutdown time are determined, and breadth iteration is used to expand the feasible region of the problem to improve the optimality of the solution. A comparative case study is conducted on IEEE 118 system and ACTIVSg10k system. The results indicate that the proposed method can reduce the scale of the problem, decrease the number of unit state attempts, and achieve efficient search iteration of the unit state, exhibiting fast computational speed and high efficiency, and demonstrating practicality and effectiveness in solving large-scale unit commitment problems.