Hybrid Physics and Deep Learning Model for Interpretable Vehicle Simulation[go to overview]
Physical motion models offer interpretable estimates for the motion of vehicles. However, certain parameters of these models, such as those related to aero- and hydrodynamics, are expensive to measure and are often only roughly approximated, which reduces the estimation accuracy. Recurrent neural networks achieve high estimation accuracy at low cost, as they can use cheap measurements collected during the routine operation of the vehicle. However, they are not interpretable, unlike physical models. To ensure highly accurate and interpretable simulation without a need for expensive measurements of physical parameters, we propose a hybrid approach combining deep learning and physical motion models. Our approach assumes that only easily accessible parameters are available, whereas the remaining ones are indirectly estimated by a recurrent neural network. We have evaluated our approach for the use case of ship motion under environmental disturbances. The results show that the hybrid model increases the interpretability without decreasing the accuracy compared to existing deep learning approaches, which are typically accurate but not interpretable.
19.12.19 - 10:15