Alert button

Physics-guided Neural Networks (PGNN): An Application in Lake Temperature Modeling

Feb 20, 2018
Anuj Karpatne, William Watkins, Jordan Read, Vipin Kumar

Figure 1 for Physics-guided Neural Networks (PGNN): An Application in Lake Temperature Modeling
Figure 2 for Physics-guided Neural Networks (PGNN): An Application in Lake Temperature Modeling
Figure 3 for Physics-guided Neural Networks (PGNN): An Application in Lake Temperature Modeling
Figure 4 for Physics-guided Neural Networks (PGNN): An Application in Lake Temperature Modeling

Share this with someone who'll enjoy it:

This paper introduces a novel framework for combining scientific knowledge of physics-based models with neural networks to advance scientific discovery. This framework, termed as physics-guided neural network (PGNN), leverages the output of physics-based model simulations along with observational features to generate predictions using a neural network architecture. Further, this paper presents a novel framework for using physics-based loss functions in the learning objective of neural networks, to ensure that the model predictions not only show lower errors on the training set but are also scientifically consistent with the known physics on the unlabeled set. We illustrate the effectiveness of PGNN for the problem of lake temperature modeling, where physical relationships between the temperature, density, and depth of water are used to design a physics-based loss function. By using scientific knowledge to guide the construction and learning of neural networks, we are able to show that the proposed framework ensures better generalizability as well as scientific consistency of results.

* submitted to ACM SIGKDD 2018  
View paper onarxiv icon

Share this with someone who'll enjoy it: