This paper proposes a Monte Carlo Tree Search algorithm based on p-rollout that leverages the tree structure of MCTS to significantly enhance search performance in high-dimensional state spaces. The EOS task scheduling problem is modeled as a Markov Decision Process, with newly designed state space, action space, and reward function. A pretrained PPO network, referred to as p-rollout, replaces the traditional random rollout in MCTS. Experimental results demonstrate that this algorithm effectively improves task scheduling accuracy, enables reasonable task structuring, and ensures the completion of scheduling under resource constraints. Ultimately, the proposed approach provides near-optimal planning solutions without compromising spacecraft safety or computational efficiency.