Abstract:Physics-informed neural networks (PINNs) incorporate established physical principles into the training of deep neural networks, ensuring that they adhere to the underlying physics of the process while reducing the need for labeled data, since the desired output is not a prerequisite for physics-informed training. For modeling systems described by Ordinary Differential Equations (ODEs), traditional PINNs typically take continuous time as an input variable and produce the solution to the corresponding ODE. However, in their original form, PINNs neither accommodate control inputs nor do they effectively simulate variable long-range intervals without experiencing a significant decline in prediction accuracy. In this context, this work introduces a novel framework known as "Physics-Informed Neural Nets for Control" (PINC). PINC presents an innovative PINN-based architecture tailored to control problems, capable of simulating longer-range time horizons that are not predetermined during training. This increased flexibility sets it apart from traditional PINNs. The variable simulation time is achieved by adding inputs to the PINC network that convey the initial condition and the control signal for a particular time interval. Simulating variable long-range intervals involves running the PINC net across a sequence of shorter intervals. In this autoregressive process, the network predictions are linked in a self-feedback mode, with the initial state (input) of the next interval set to the last predicted state (network output) of the previous interval. We showcase the effectiveness of our proposal in identifying and controlling three nonlinear dynamic systems: the Van der Pol oscillator, the four-tank system, and an electric submersible pump. Crucially, these experiments demonstrate that learning the dynamics of these systems can be achieved without relying on any sample collected from the actual process, and it offers faster inference speed compared to numerical simulations.

Control of Partial Differential Equations via Physics-Informed Neural Networks

An extended physics informed neural network for preliminary analysis of parametric optimal control problems

An Analysis of Physics-Informed Neural Networks

Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations

Evaluating Error Bound for Physics-Informed Neural Networks on Linear Dynamical Systems

Residual-based error bound for physics-informed neural networks

Physics-Informed Representation and Learning: Control and Risk Quantification

On the Stability and Convergence of Physics Informed Neural Networks

Estimates on the generalization error of Physics Informed Neural Networks (PINNs) for approximating PDEs

Physics-Informed Neural Network Policy Iteration: Algorithms, Convergence, and Verification

Estimates on the generalization error of Physics Informed Neural Networks (PINNs) for approximating a class of inverse problems for PDEs

Physics Informed Deep Learning (Part I): Data-driven Solutions of Nonlinear Partial Differential Equations

Neural Control of Parametric Solutions for High-dimensional Evolution PDEs

Characterizing and Mitigating the Difficulty in Training Physics-informed Artificial Neural Networks under Pointwise Constraints

Physics-informed Neural Networks to Model and Control Robots: a Theoretical and Experimental Investigation

Physics-informed neural nets for control of dynamical systems

Error estimates for physics informed neural networks approximating the Navier-Stokes equations

Exact and approximate error bounds for physics-informed neural networks

Learning from Integral Losses in Physics Informed Neural Networks

A PINN approach for the online identification and control of unknown PDEs

A Unified Framework for the Error Analysis of Physics-Informed Neural Networks