Air pollution forecasting with Numerical modeling and ML

• This project is designed to predict the levels of air pollution in South Korea by using numerical modeling and machine learning techniques.
• Convection-diffusion model, LSTM
• Sep. 2, 2020 ~ Oct. 8, 2020

Scientific computing lab project for academic conference

• This repo is maintained by 오서영, 신영민
• Silver award in natural science academic conference

| Summary | Presentation |

Process

1. Refining location (latitude/longitude), wind (directional/speed) and air pollution dataset | Code

2. Visualization with MATLAB, Simple implementaion of numerical modeling with refined dataset | Code

It is realistically impossible to obtain wind and air pollution data at all points due to problems such as cost and time.
So we come up with a way to get empty space data through interpolation.

3. Cubic Interpolation and Inverse Distance Weighted | Code

We apply cubic interpolation to wind vector dataset by using scipy. It returns the value determined from a peicewise cubic, continuously differentiable and approximately curvature-minimizing polynomial surface.
Also, We use Inverse Distance Weighted (IDW) to air pollution dataset. IDW is an interpolation method that computes the score of query points based on the scores of their k-nearest neighbours, weighted by the inverse of their distances.

4. Convection-diffusion equation with interpolated dataset | Code

Central difference method, Neumann boundary condition

Convection-diffusion equation and discretized one

5. Long Short Term Memory (LSTM) | Code

Simple RNN, Simple LSTM, Stacked LSTM

Results - Visualization

1. Mathematical Modeling - Convection-diffusion equation

2. Machine Learning - LSTM

Dataset

[1] 기상자료개방포털, https://data.kma.go.kr/cmmn/main.do  
[2] 에어코리아, https://www.airkorea.or.kr/index