论文标题
部分可观测时空混沌系统的无模型预测
Invariants and chaos in the Volterra gyrostat without energy conservation
论文作者
论文摘要
储层计算是预测湍流的有力工具,其简单的架构具有处理大型系统的计算效率。然而,其实现通常需要完整的状态向量测量和系统非线性知识。我们使用非线性投影函数将系统测量扩展到高维空间,然后将其输入到储层中以获得预测。我们展示了这种储层计算网络在时空混沌系统上的应用,该系统模拟了湍流的若干特征。我们表明,使用径向基函数作为非线性投影器,即使只有部分观测并且不知道控制方程,也能稳健地捕捉复杂的系统非线性。最后,我们表明,当测量稀疏、不完整且带有噪声,甚至控制方程变得不准确时,我们的网络仍然可以产生相当准确的预测,从而为实际湍流系统的无模型预测铺平了道路。
The model of the Volterra gyrostat (VG) has not only played an important role in rigid body dynamics but also served as the foundation of low-order models of many naturally occurring systems. It is well known that VG possesses two invariants, or constants of motion, corresponding to kinetic energy and squared angular momentum, giving oscillatory solutions to its equations of motion. Nine distinct subclasses of the VG have been identified, two of which the Euler gyroscope and Lorenz gyrostat are each known to have two constants. This paper characterizes quadratic invariants of the VG and each of its subclasses, showing how these enjoy two invariants even when rendered in terms of a non-invertible transformation of parameters, leading to a transformed Volterra gyrostat (TVG). If the quadratic coefficients of the TVG sum to zero, as they do for the VG, the system conserves energy. In all of these cases, the flows preserve volume. However, physical models where the quadratic coefficients do not sum to zero are ubiquitous, and characterization of invariants and the resulting dynamics for this more general class of models with volume conservation but without energy conservation is lacking. This paper provides the first such characterization for each of the subclasses of the VG in the absence of energy conservation, showing how the number of invariants depends on the number of linear feedback terms. It is shown that the gyrostat with three linear feedback terms has no invariants. The number of invariants circumscribes the possible dynamics for these three-dimensional flows, and those without any invariants are shown to admit rich dynamics including chaos. This gives rise to a broad class of three-dimensional volume conserving chaotic flows, arising naturally from model reduction techniques.