The path planning problem for mobile robots has garnered significant attention in recent years. While existing algorithms based on a deep $Q $ network have exhibited a certain degree of effectiveness, they often struggle with excessive randomness in action selection and slow convergence during training. Moreover, existing algorithms are less involved in quantitative analyses of dynamic settings. Therefore, a path planning algorithm based on a double deep $Q $ network (DDQN) is introduced, featuring a unique time-series input structure. Furthermore, a distinctive experience distribution strategy is proposed, which optimizes network training efficiency by distributing experiences from different experience pools at different training stages. The proposed algorithm is evaluated in both static and dynamic environments. Compared to the DDQN method, the proposed algorithm reduces training time by 50% and increases the success rate by 9.6%.