Traditional emergency inspection methods for power grid disasters primarily rely on manual inspection and single data processing techniques, which suffer from issues such as low efficiency and slow response. Although UAV-based emergency inspection technologies for power grids can enhance assessment timeliness, existing methods struggle to balance obstacle avoidance efficiency with global optimality in complex terrains and lack re-al-time disaster data processing capabilities. To investigate optimal path generation and dynamic prediction methods for UAV emergency inspection, this paper first introduces an adaptive heuristic-weight and turning-cost A star (AHWTA*) algorithm. This algorithm adapts heuristic weights dynamically and incorporates a turning cost opti-mization mechanism to reduce redundant turning points, thereby solving the UAV path planning problem. Fur-thermore, an adjacent-feedback cycle reservoir with cross jumps (ACRCJ) model is proposed based on echo state network (ESN), leveraging its short-term memory characteristics and high-dimensional mapping capabilities to achieve rapid processing of disaster data. Experimental results demonstrate that the combined AHWTA* algorithm and ACRCJ model enable intelligent generation and dynamic prediction of UAV inspection paths in power grid disaster scenarios, outperforming other comparative models in terms of performance.