Three-Body Hidden Forces Benchmark Dataset: 1,200+ Simulated Chaotic Trajectories with Controlled Perturbations

This dataset contains over 1,200 high-resolution simulated three-body dynamical trajectories generated for anomaly detection and scientific machine learning research. It includes clean Newtonian simulations alongside five categories of controlled unmodeled perturbations: hidden mass, non-inverse-square gravitational interactions, drag forces, impulsive disturbances, and time-varying gravitational constants. Each sample is stored in compressed NumPy (.npz) format and contains trajectory state data (positions and velocities) suitable for temporal modeling and representation learning. The dataset is designed to benchmark anomaly detection methods, particularly Neural ODEs and dynamical latent models, in chaotic nonlinear systems where hidden or unmodeled forces may alter expected behavior. It can be used for supervised anomaly classification, out-of-distribution generalization studies, and scientific AI experimentation in chaotic dynamical systems.本数据集包含1200余条高分辨率三体动力学模拟轨迹，专为异常检测与科学机器学习研究构建。数据集涵盖纯净牛顿力学模拟样本，以及五类受控未建模扰动：隐藏质量、非平方反比引力相互作用、阻力、脉冲扰动与时变引力常数。每个样本均以压缩NumPy（.npz）格式存储，包含适用于时序建模与表征学习的轨迹状态数据（位置与速度信息）。本数据集旨在为混沌非线性系统中的异常检测方法提供基准测试，此类系统中隐藏或未建模的作用力可能改变预期行为，尤其适配神经常微分方程（Neural ODEs）与动力学隐变量模型的相关研究。本数据集可应用于监督式异常分类、分布外泛化研究，以及混沌动力学系统中的科学人工智能实验。