Real-time Human Response Prediction Using a Non-intrusive Data-driven Model Reduction Scheme

TL;DR

This paper introduces a novel two-step data-driven model reduction method that combines classic MOR techniques with LSTM neural networks to accurately predict time-dependent parameterized ODEs in real-time applications.

Contribution

It presents a new approach integrating MOR with LSTM to handle time-varying parameters, extending previous methods that only addressed constant parameters.

Findings

High accuracy in predicting responses to time-varying accelerations.

Real-time capability demonstrated on a car occupant model.

Effective approximation of parameterized ODEs with time-dependent parameters.

Abstract

Recent research in non-intrusive data-driven model order reduction (MOR) enabled accurate and efficient approximation of parameterized ordinary differential equations (ODEs). However, previous studies have focused on constant parameters, whereas time-dependent parameters have been neglected. The purpose of this paper is to introduce a novel two-step MOR scheme to tackle this issue. In a first step, classic MOR approaches are applied to calculate a low-dimensional representation of high-dimensional ODE solutions, i.e. to extract the most important features of simulation data. Based on this representation, a long short-term memory (LSTM) is trained to predict the reduced dynamics iteratively in a second step. This enables the parameters to be taken into account during the respective time step. The potential of this approach is demonstrated on an occupant model within a car driving scenario. The reduced model's response to time-varying accelerations matches the reference data with high accuracy for a limited amount of time. Furthermore, real-time capability is achieved. Accordingly, it is concluded that the presented method is well suited to approximate parameterized ODEs and can handle time-dependent parameters in contrast to common methods.