A significant number of transient astronomical events are discovered through multi-wavelength and multi-messenger observatories, which require rapid follow-up observations by astronomical satellite. Efficient observation scheduling is crucial for handling time-sensitive Target of Opportunity (ToO) tasks, which demand quick responses under dynamic and uncertain conditions. To address this challenge, we propose a Genetic Programming-based Dynamic Scheduling algorithm (GPDS) for astronomical observations. The GPDS algorithm utilizes genetic programming, a type of hyper-heuristic method, to evolve and optimize Heuristic Rules (HRs) for scheduling tasks, considering both local and global information simultaneously. The best-evolved HR is integrated into a timeline-based scheduling process, which makes decisions at each decision point to generate a feasible schedule. In experimental scenarios with varying task scales, the GPDS algorithm significantly outperforms scheduling methods based on classic HRs, such as First Come First Served (FCFS), Shortest Job First (SJF) and Earliest Deadline First (EDF), particularly in terms of total reward. This demonstrates that the GPDS algorithm is both adaptable and robust, making it an efficient and reliable solution for dynamic task scheduling in satellite missions, especially under rapidly changing observation scenarios.