In order to solve the problems of low direction-finding accuracy, insufficient signal separation capability and limited resolution caused by unreasonable subarray allocation in spatial smoothing techniques for coherent sources, an improved Multi-Strategy Collaborative Simulated Annealing (MSC-SA) algorithm is proposed for subarray partition. The global optimization ability and computational efficiency of the algorithm are improved by preserving the initial solution and introducing steering vector condition number to guide the improvement of neighborhood structure. And the adaptation function based on spatial resolution is constructed for comprehensively consideration of the direction-finding accuracy and algorithm performance. Through three sets of simulation experiments, compared with the existing methods, the improved MAC-SA algorithm demonstrates faster convergence and superior performance in direction-finding accuracy, signal separability, and global optimization capability. The proposed algorithm effectively improves the accuracy and reliability of signal direction finding in complex electromagnetic environment.