The Startup
Published in

The Startup

Model-Based Control using Neural Network: A Case Study

Control simulation of a mechanical system using a neural network-based model predictive control algorithm

[Photo by Mathew Schwartz]

Motivation

Problem

nonlinear MIMO system [Illustration by Author]

Approach

Model

Explainable Artificial Intelligence [DARPA]
Comparison of general RL algorithms [article]

Neural Network Model

A neural network as a function approximator [1]

Neural Network Design

Block diagram of my neural network design
LReLU Activation function
You can’t just do this. That is NOT how it works! [source]

Model Predictive Control (MPC)

MPC strategy [TU Delft]
Do you see the line plots? This is what I mean by “smooth”. [xkcd]

Evaluation Metrics

Control Strategy

The Reinforcement Learning (RL) Framework

General RL framework [Mathworks]

The major work in this article is online learning of the local dynamics. This means that the neural network is not pre-trained. Instead, it is continuously learning the dynamics function of the system.

Ordinary differential equations describing the system motion
Uncontrolled random system trajectories
Single-step prediction [black: ground truth; blue: predicted]
10-step-ahead prediction [black: ground truth; blue: predicted]
Initial state conditions at 40 seconds

Final Results

RMSE= 0.0430 m, I= 382.95 kNs, ΔI= 818.62 kNs
RMSE= 0.0814 m, I=348.94 kNs, ΔI= 873.65 kNs

TL;DR

Future Work

References

--

--

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store