Title:

Classification of Auroral Images and Uncertainty:A Bayesian Machine Learning Approach

The aurora is a luminous phenomenon of Earth's upper atmosphere, ignited by the interaction of charged particles from Earth's Magnetosphere with the neutral atoms of the atmosphere, particularly during geomagnetic storms and substorms. The aurora is often accompanied by currents induced in the conductors on the Earth's surface, called Geomagnetically Induced Currents (GICs). GICs can lead to voltage dips, elevated reactive power demand, transformer overheating, or malfunction of the electric power devices or power grids. The literature consists of studies where Machine Learning (ML) techniques have been used to classify space weather data such as auroral images. We are studying the role that classified auroral images can play in predicting GICs to mitigate the risks presented by future Geomagnetic storms. While ML holds great promise, such models' results could be unreliable due to uncertainty. Uncertainty refers to situations involving imperfect knowledge and is inherent in a stochastic and partially observable environment. For example, auroral images are classified into predefined labels, and there is no consensus within the scientific community regarding such labels. Also, the captured images are subject to noise and future environmental variability. An ML model generates an optimal solution based on the training data. However, if the data and model parameters' uncertainty is not considered, then such optimal solutions have a high risk of real-world deployment failure. Uncertainty becomes especially relevant when real-time data is used to train the model over historical data. For this ongoing work, a Bayesian ML model has been developed to classify prelabeled auroral images. The model leverages an interpretation of probability called Bayesian statistics to quantify data and parameter uncertainty and classify the images with a confidence level of 95%. We further propose how the implemented ML model can be trained using real-time data on a local system via edge computing (Edge AI).

Name

Comment*

Comments are viewable only by submitter

Captcha*