Virtual power plants face problems such as low efficiency in multi-energy coordination. This study proposes a fusion framework based on graph neural network (GNN) and proximal policy optimization (PPO) algorithm. By constructing a graph structure topology of energy equipment to represent its spatial correlation, the graph convolution layer and multi-head attention mechanism of GNN are used to extract the spatiotemporal coupling characteristics of energy resources. The extracted high-dimensional features are then input into the policy network of the PPO algorithm. The dominance evaluation of the value network is combined to achieve gradient optimization of the dynamic scheduling strategy, and strategy parameters are adjusted in real time by designing an adaptive mechanism based on a sliding time window.