Data Engineering and Data Science

5Application of Data Science in Macromodeling of Nonlinear Dynamical Systems

Nagaraj S.¹*, Seshachalam D.¹ and Jayalatha G.²

¹ BMS College of Engineering, Bengaluru, India

² RV College of Engineering, Bengaluru, India

Abstract

Systems trained on data to predict the behaviour of the real-world physical system models are giving faster and more accurate simulation results than the nonlinear mathematical models. The reason for this is that the nonlinear structure of real-world problems with complexities cannot be handled by linearized models, and nonlinear models take long evaluation time, which includes redundant parameter evaluations. In this work, the neural network modeling of nonlinear dynamical system using supervised learning technique and proper orthogonal decomposition has been addressed.

The work is validated using two numerical models. The biological neuron spiking and ring oscillator models are considered here. They are trained with multiple temporal swatches evolved on different parametric variation and trailed out at different time instances. Neural network tries to learn such models by analysing snapshot observations from active systems. Trained neural network is able to make better prediction for validating data which is different from the training data set. The complexity of the data set is reduced by applying the POD on the original data set. This demonstrates that the neural network is successfully utilised to macro model the autonomous dynamical system.

Keywords: ...