Traditional Artificial Bee Colony (ABC) algorithms exhibit inadequate optimization capabilities when addressing complex optimization problems. To further enhance the performance of conventional ABC algorithms, researchers have continuously updated the original algorithm, proposing new improved versions such as GABC and IABC. However, these improvements still face issues with slow convergence rates. In response to these challenges, this paper introduces a novel improved algorithm, named Nonlinear Dual Search Artificial Bee Colony (NDS-ABC). In NDS-ABC, nonlinear function variations replace the linear transformations of dynamic weight factors found in the original IABC. The entire search process is divided into two phases, enabling dual search capabilities. This algorithm is applied to eight benchmark functions to analyze its performance. Experimental results indicate that the NDS-ABC algorithm demonstrates superiority in terms of optimization capability and convergence speed compared to traditional ABC and improved IABC algorithms, particularly in tackling multimodal separable problems.