Neural Data-Enabled Predictive Control
Mircea Lazar
TL;DR
This work extends data-enabled predictive control (DeePC) to nonlinear systems by integrating deep neural networks (NNs). The key idea is that the last hidden layer of a deep NN forms a neural-space basis, with the output layer providing affine interpolation in that space, enabling offline training of the basis and online DeePC predictions. If the NN uses linear activations, the approach reduces to the original linear DeePC, while nonlinear activations yield a set-valued neural predictor that is regularized to balance bias and variance. The paper introduces computationally efficient variants (including Neural-DeePC-3) and demonstrates performance on a nonlinear pendulum, highlighting significant gains in online computation suitable for real-time control while preserving predictive accuracy.
Abstract
Data-enabled predictive control (DeePC) for linear systems utilizes data matrices of recorded trajectories to directly predict new system trajectories, which is very appealing for real-life applications. In this paper we leverage the universal approximation properties of neural networks (NNs) to develop neural DeePC algorithms for nonlinear systems. Firstly, we point out that the outputs of the last hidden layer of a deep NN implicitly construct a basis in a so-called neural (feature) space, while the output linear layer performs affine interpolation in the neural space. As such, we can train off-line a deep NN using large data sets of trajectories to learn the neural basis and compute on-line a suitable affine interpolation using DeePC. Secondly, methods for guaranteeing consistency of neural DeePC and for reducing computational complexity are developed. Several neural DeePC formulations are illustrated on a nonlinear pendulum example.