0% found this document useful (0 votes)
48 views3 pages

DD - Dynsys: 1 Anatomy of A Dynamic System

This document discusses dynamic systems and techniques for modeling them, including dynamic mode decomposition (DMD). It provides the following key points: 1. The anatomy of a dynamic system includes the state, time, actuation, and a function relating their time derivatives. Modeling challenges include nonlinearity, unknown functions, high dimensionality, chaos, noise, and uncertainty. 2. DMD finds the best linear operator to map between snapshots of a dynamic system over time. It organizes data into matrices and uses singular value decomposition and eigendecomposition to identify dominant patterns and predict future states. 3. DMD decomposes the system into eigenmodes and eigenvalues to build an intuitive, low-dimensional model of the dominant dynamics

Uploaded by

Hari Appu
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
48 views3 pages

DD - Dynsys: 1 Anatomy of A Dynamic System

This document discusses dynamic systems and techniques for modeling them, including dynamic mode decomposition (DMD). It provides the following key points: 1. The anatomy of a dynamic system includes the state, time, actuation, and a function relating their time derivatives. Modeling challenges include nonlinearity, unknown functions, high dimensionality, chaos, noise, and uncertainty. 2. DMD finds the best linear operator to map between snapshots of a dynamic system over time. It organizes data into matrices and uses singular value decomposition and eigendecomposition to identify dominant patterns and predict future states. 3. DMD decomposes the system into eigenmodes and eigenvalues to build an intuitive, low-dimensional model of the dominant dynamics

Uploaded by

Hari Appu
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

DD_DynSys

December 23, 2018

1 Anatomy of a Dynamic system


In the Data-driven Dynamic systems with Machine Learning and Regression, the general anatomy
of a dynamic system is:
ẋ = f ( x, t, u, β)
ẋ - time derivative of x
x - state of the system
t - time
u - actuation(contolled)
FirstPrinciples ⇒ DATA

1.1 Challenges
1. Nonlinear f ()
2. Unknown f ()
3. High Dimensional
4. Chaotic, Transient
5. Noise, Stochastic
6. Multi-scale Dynamics
7. Uncertainity
8. Coupled Problems

1.2 Uses of models


1. Future State Prediction (Ensemble/Statistical)
2. Design/Optimization
3. Control (Actively modifying the behavior of the system)
4. Interpretability and Physical Intuition

1.3 Techniques
1. Regression(Linear and sparse)
2. Neural Networks(Deep Learning)
3. Genetic Programming

1
2 Dynamic Mode Decomposition(DMD)
2.1 Dynamic system of coupled spatial temporal nodes
2.1.1 Collect Data
The movie of the flow is broken down into a sequence of snapshots. The pixels of each snapshot
are stacked up vertically as a column

2.1.2 Organize into matrices


These vertically stacked pixels of each snapshots are then arranged in a matrix starting from intial
time.
The columns are essentially evolving in time along the dynamics of the system. Therefore, we
get thesetall skinny matrices. 
.. .. ..
. . . 
X=  x 1 x 2 · · · x m

−1 
.. .. ..
. . .
 
.. .. ..
. . . 
X′ =  x
 2 x 3 · · · x m

.. .. ..
. . .

2.1.3 DMD
DMD tries to find the best fit linear operator that advance X into X ′ .
X ′ ≈ AX
The transformation matrix A is written as:
A = X′ X†
DMD is used as a Diagnostics tool to predict the past and the future time states. The future
state can be predicted by: xk+1 = Axk
DMD is computed by the use of following decompositions:

Singular Value Decomposition(SVD) Identifying the dominant(low-dimensional data) coher-


ent patterns.
Instead of computing the gigantic A matrix, it is computed by taking the pseudo-inverse of
V ∗ , beacause it is easy to compute, since it is an unitary matrix. Followed by the multiplication of
Σ−1 and the complex conjugate of U.
X = UΣV ∗
X ′ = AX = AUΣV ∗
By projecting A to the dominant singular vectors U and U ∗ , a smaller matrix à is achieved. Ã
matrix gives the best-fit dynamical system of the dominant coherent patterns that evolve in time.
U ∗ X ′ VΣ−1 = U ∗ AU = Ã

Eigen Decomposition The eigen values(Λ) of A is equal to the ones od Ã. To compute the eigen
vectors W.
ÃW = WΛ
Φ = X ′ VΣ−1 W

2
A useful f ( x, y) should predict connections among derivatives.
y′ ∆y

x ′ ∆x data
∂y ∂ f /∂x
=
∂x f (x,y) ∂ f /∂y

You might also like