/Hurricane-Path-Prediction

Hurricane Path Prediction using a Recurring Neural Network

Primary LanguagePythonMIT LicenseMIT

Hurricane-Path-Prediction

Hurricane Path Prediction using a Recurring Neural Network

In this repository you will find the code necessary to train a LSTM-based Neural Network to predict the next position of a hurricane.

The Data

The data used was retrieved from the NOAA HURDAT2 database.

As explained on their website it contains the geographic progression data of each hurricane since 1851 (currently am using a version downloaded in 2020, only goes up to 2015)

The hurricanes' coordinates are available for the duration of the hurricane at 4 times in the day : 0000, 0600, 1200, 1800.

The Model

I built the model using PyTorch.

It's a very shallow model, as I found out the resultss only worsened when I deepened the LSTM part.

  • 1 Input Layer with Batch Normalization

  • 1 LSTM Layer with 256 hidden features

  • 1 Dense (Fuully Connected Layer)

QUICKSTART

Installation

git clone https://github.com/Zhack47/Hurricane-Path-Prediction.git

cd Hurricane-Path-Prediction/

pip3 install -r requirements.txt

Training

python3 train.py

Testing

python3 test.py