A Graph Network-Based Cluster Dynamics Prediction Method

Abstract

The invention discloses a cluster dynamics prediction method based on a graph network, comprising: simulating self-driven cluster motion according to a Vicsek model to generate a data set; processing the data set into graph structure data through data preprocessing, including nodes, edges and Global attributes; implement preliminary feature extraction and data dimensionality reduction; use graph network blocks, select multi-layer perceptrons as update functions, and information is propagated on the graph through certain rules; information is restored to obtain graph-structured data; use stochastic gradient descent The optimization algorithm trains the graph network and automatically learns the parameters in the graph network. The invention integrates traditional physical science and deep learning technology, and can realize the simulation and reasoning of objects and relationships in the self-driven cluster system, and only determine the long-term evolution of the system from the initial position of the particles without any manual features.

Description

technical field [0001] The present invention relates to the technical field of deep learning, and more particularly, to a method for predicting cluster dynamics based on a graph network. Background technique [0002] Active matter systems are not bound by thermodynamic rules such as detailed equilibrium conditions or the fluctuation-dissipation theorem, and thus emerge rich and complex dynamical phenomena, such as cluster motions that exist widely on spatial scales in nature (e.g., animals, There are many examples in human societies (e.g., collective human behavior and social networks), and in multi-agent systems (e.g., multi-robot systems and multi-vehicle cooperative control). A typical flock motion model can simulate the motion of many forms of matter, including the motion of flocks of birds and fish. The model system that takes the swarm movement behavior as the research object is a collection composed of a large number of autonomous individuals, which are updated by co...

AI Technical Summary

Problems solved by technology

However, when traditional machine learning solves the above problems, it usually maps the graph structure data into a simple representation, which makes the topological information of the structure itself may be lost in the preprocessing stage, affecting the final prediction results

Images