Prediction of Asteroids Orbit Paths and Collision

Introduction

During this semester I researched creating a Predictive model for asteroid paths in hopes of predicting collisions. I went about this in two different ways. I created one model based on an article I read on predicting common behavior movement with linear regression as I wanted to see if this idea would carry over to orbit paths. The next model was based off of measuring correlated data provided by ephemeris charts and using it to create a prediction model

Tools and methods

The data modeling, training, and testing was done using jupyter notebook in Visual Studio code. In the notebook I would access the data provided by Jet Lab Propulsion small body database system and there ephemeris chart system by using astroquerys’ sbdb and horizon extensions. It would then be modeled into numpy dataframe, where data would be structured for the models and the unnecessary data would be removed. Then scikit learn Linear regression models where used to fit the data and predict the location of the asteroids. The result was then plotted with matplot.lib

Results

The first method views a group of asteroid  longitude and latitude positions comparing them to where it will be at the next observed point in time. This prediction method was based off of a trajectory clustering model from the article Predicting Common Movement. This model was able to group together the movements of ten asteroids and predict a matrix of arrows that show where the data is moving is most likely to move in a short distance.

The second method was built off of correlation between asteroid data and it longitude and latitude coordinates.  It was able to fitted and used for predictions of orbit paths however the margin of stander error(mse) for the longitude coordinate was high around 6000 during some test, this compared to the latitude which score a mse of .01 on some tests. Led to the predicted orbit having exaggerated movements along the x-axis.

Conclusions

The first model had shown the capability to help generalize when a group of asteroids will be in close proximity to each other. This could be used to generalize if orbit paths are likely to overlap in short term setting. As the generalization this method makes does not work well with the orbit paths after mapping a full orbit due to overlap.

The correlation model showed promise in the ability to predict an orbital path over a long term. The inaccuracy of the longitude may cause it some problems, but due to the much better accuracy of the latitude shows that the modeling could improve upon further training. Leaving room for the predictions to become more accurate.

Video link to elevator Pitch

Literature Cited

Bakibayev, Timur. “Predicting Common Movement.” Medium, Analytics Vidhya, 25 Aug. 2020, medium.com/analytics-vidhya/predicting-common-movement-b225569bee91.

Davegn. “Asteroids Potential Hazards Prediction.” Kaggle, Kaggle, 30 June 2021, www.kaggle.com/code/davegn/asteroids-potential-hazards-prediction.


Poster

css.php