This paper introduces an enhanced state-space partitioning (ESSP) method designed for the efficient evaluation of complex power systems. The proposed algorithm integrates a fast Sorting technique (FST), Monte Carlo simulation (MCS), and a lightweight Q-Learning algorithm to facilitate reliable analyses of intricate power grids. Notably, the ESSP method employs an innovative adaptive dynamic threshold adjustment mechanism. By examining the correlation between the skewness of the fault probability distribution and computational resource consumption, the method optimizes the ceiling probability in real time within a Q-learning reinforcement framework, thereby achieving a dynamic balance between enumeration and sampling. A modified IEEE-RTS reliability test system was utilized to evaluate the effectiveness of the ESSP method. The results indicate that the ESSP method substantially improves computational efficiency while preserving evaluation accuracy and demonstrates superior adaptability and robustness across systems with varying reliability levels and dynamic operating states. This research provides an efficient tool to support online reliability evaluation and optimization decision-making in power systems.