Group agent trajectory prediction method and system with low computational complexity and domain adaptability

The invention provides a swarm agent trajectory prediction method and system with low computational complexity and domain adaptability, and the method comprises the steps: S1, obtaining an agent trajectory and an initial correlation, respectively representing the agent trajectory and the initial correlation by nodes and edges, and constructing a spatial edge feature directed graph and a time edge feature directed graph; s2, performing node-edge feature fusion processing on the spatial edge feature directed graph and the time edge feature directed graph to obtain enhanced spatial interaction features and enhanced time interaction features; s3, adaptively capturing a space global interaction feature and a time global interaction feature for the enhanced space interaction feature and the enhanced time interaction feature respectively by using a mixed feature attention network; s4, constructing a spatial adjacency matrix and a time adjacency matrix of a residual aggregation module based on the spatial global inte