Framework for development and evaluation of machine learning models for solar flare prediction.
Machine Learning, Python, Tensorflow, Solar flare, Convolutional Neural Network
The Institute for Data Science (I4DS) curated a machine learning dataset and benchmark for the prediction of solar flares. The benchmark includes an implementation of a simple baseline algorithm (e.g. Zero-rule). In this project we worked towards the goal of training better machine learning models which make use of this dataset.
Solar flares are intense bursts of radiation which can disrupt the power grids of a continent, shut down the GPS system or irradiate people exposed in space. Developing systems for predicting solar flares would allow us to precisely aim our observation instruments at upcoming events, and eventually allow countermeasures to be taken in time against worst-case scenarios.
To enable our and future investigations, we built a software framework. It is built to train and evaluate Keras machine learning models on the solar flare dataset. The primary design goal of this framework was to fully separate the definition of a model from all of the code required to train and evaluate it. Now, new models can be trained and evaluated, thus compared to other models, with minimal overhead. The framework provides a few settings regarding the initialization of the environment, the loading and preprocessing of the data and the detail of the evaluation, however, all of the complexity required in these areas is encapsulated and hidden away by sensible defaults. We developed and evaluated several models with various architectures and differing hyperparameters, using our framework described above.
Project duration: 1 Semester
Man hours invested: 360h (180h per person)
Team size: 2 Students
Institute for Data Science | FHNW
Siffer Florian
Ackermann Patrick
Csillaghy André (andre.csillaghy@fhnw.ch)
Felix Simon (simon.felix@fhnw.ch)