Abstract: In response to the challenges of large storage requirements and poor real-time performance faced by path planning for multirotor UAVs (unmanned aerial vehicles) in cluttered, large-scale environments, an efficient path planning method is proposed. Firstly, grid occupancy state construction is transformed into a classification task by the random mapping method (RMM), and thus a continuous implicit linear model storing only the parameter vector is learned, enabling millisecond-level rapid environmental queries. Secondly, a heuristic function centered on the occupancy change rate is introduced to solve the problems of traditional Kinodynamic A^* algorithms, such as poor search efficiency caused by only considering geometric distances. Unlike traditional methods, the proposed method utilizes environmental gradient information to preemptively identify and avoid high-risk areas, reducing the expansion of ineffective nodes. Finally, a closed-form ESDF (Euclidean signed distance field) method is proposed, transforming the mapping of spatial coordinates to obstacle distances into an RMM regression task, achieving an analytical expression for the ESDF and avoiding the clutteredness and resolution limitations of traditional interpolation calculations, which significantly enhances the efficiency and accuracy of distance field construction. Simulation results show that the proposed method reduces the search time by an average of 79.73% and lowers the number of accessed nodes by 23.69% across different resolutions (0.3, 0.2, 0.1 m) compared to the traditional Kinodynamic A^* algorithm; ESDF construction time is reduced by over 99%. Actual flight tests further validate the efficiency and robustness of the proposed method.