2020/04/01 by Benjamin Karg, Karg, Benjamin, Sergio Lucia +1 · 1 citation
Engineering · #Advanced Control Systems Optimization #Control Systems and Identification #FOS: Electrical engineering #FOS: Mathematics #Fault Detection and Control Systems #Optimization and Control (math.OC) #Systems and Control (eess.SY) #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2004.00521
openalex publication_date 2020/04/01 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28
Neural networks can be used as approximations of several complex control\nschemes such as model predictive control. We show in this paper which\nproperties deep neural networks with rectifier linear units as activation\nfunctions need to satisfy to guarantee constraint satisfaction and asymptotic\nstability of the closed-loop system. To do so, we introduce a parametric\ndescription of the neural network controller and use a mixed-integer linear\nprogramming formulation to perform output range analysis of neural networks. We\nalso propose a novel method to modify a neural network controller such that it\nperforms optimally in the LQR sense in a region surrounding the equilibrium.\nThe proposed method enables the analysis and design of neural network\ncontrollers with formal safety guarantees as we illustrate with simulation\nresults.\n