Dimensionally Consistent Learning with Buckingham Pi

Abstract

In the absence of governing equations, dimensional analysis is a robust technique for extracting insights and finding symmetries in physical systems. Given measurement variables and parameters, the Buckingham Pi theorem provides a procedure for finding a set of dimensionless groups that spans the solution space, although this set is not unique. We propose an automated approach using the symmetric and self-similar structure of available measurement data to discover the dimensionless groups that best collapse this data to a lower dimensional space according to an optimal fit. We develop three data-driven techniques that use the Buckingham Pi theorem as a constraint: (i) a constrained optimization problem with a non-parametric input-output fitting function, (ii) a deep learning algorithm (BuckiNet) that projects the input parameter space to a lower dimension in the first layer, and (iii) a technique based on sparse identification of nonlinear dynamics (SINDy) to discover dimensionless equations whose coefficients parameterize the dynamics. We explore the accuracy, robustness and computational complexity of these methods as applied to three example problems: a bead on a rotating hoop, a laminar boundary layer, and Rayleigh-Benard convection.

A Data-Driven Under Frequency Load Shedding Scheme in Power Systems

“…Thus, the challenge of adapting the linear representations to accommodate the parameters variations transforms into the challenge of accounting for the hidden variable in the model. Time-delay embedding provides an approach to augment these hidden variables, and under certain conditions, given by Takens' embedding theorem [25], the delay-augmented state yields an attractor that is diffeomorphic to the underlying, though unmeasured, full-state attractor [26]. Here, we design a deep neural network to learn a coordinate transformation from the delay embedded space into a new space where it is possible to represent the dynamics in a linear form and also to track the parameter variations in the system with input/output data.…”

Expressive architectures enhance interpretability of dynamics-based neural population models

“…The field of dynamical system identification has had great success in recovering governing equations from time series measurements, particularly in the physics domain [27,28,29]. However, existing approaches rely heavily on sparse, symbolic regression from states to derivatives [27], which poses a number of problems for spiking datasets.…”

From biological data to oscillator models using SINDy

“…Often such knowledge is not provided and either too many or not enough variables describing a system are measured, an example being the BZ reaction where one dimension is missing. Again, several extensions have been proposed that involve the use of auto-encoders (77,78) or delay embedding (70,79) to reconstruct dimensions or reduce the system. However, these approaches require high quality data to provide sufficient information to unfold missing dimensions or reduce existing ones, and therefore do not reflect data on dynamical systems in biology, where sometimes a variety of dimensions (e.g., fluorescent markers) are measured but suitable temporal data are lacking.…”

“…We could reduce the number of state variables to the most relevant ones, e.g. by using encoders (77,78). However, as shown especially in Chen et al (78), large amounts of high quality data are needed to identify the most relevant aspects of an observed dynamical system, while the set of variables we provide is also the relevant set of state variables for the model.…”