Preliminary project plan

Forecasting of PM2.5 with Neural Networks

Transcript

1. FORECASTING 2.5 OF PM WITH NEURAL NETWORKS
2. What is particulate matter? PARTICULATE MATTER (AEROSOLS) A mixture of solid particles and liquid droplets found in the air. Some particles, such as dust, dirt, soot, or smoke, are large or dark enough to be seen with the naked eye. Others are so small they can only be detected using an electron microscope.[EPA US]. 2
3. Classification 3 Physical form: -Dust -Fume -Smoke -Smog -Fog and mist Method of Generation: -Nucleation mode -Accumulation mode -Coarse mode Process of emission: -Primary aerosols -Secondary aerosols Source: -Natural source -Anthropogenic
4. Particle Size Categories • Nanoparticles(< 50nm) •Ultrafine particles (<100 nm) • PM2.5 (< 2.5 µm) • PM2.5–10: known as PM coarse. • PM10 (< 10 µm) Photo(University of Sao Paulo / University of Maryland) 4
5. Why do we bother about the size?
6. What can we do against pollution? -Regulation -Dust collection systems -Notify the population
8. Can we model particulate matter with NN? 8
9. Model proposed
10. 10 Data set The particulate matter data contain the taken measurements for atmospheric particles up to 2.5um hourly from 2010-01-01 to 2014-12-31.
13. Relationship between Features 13
14. Radial Basis Function Multilayer Perceptron Support Vector Machine 14 Neural Networks Types
15. Splitting-up Dataset Raw Data 43824 hourly Samples 1826 daily Samples Testing (30 ) Data % 535 Samples Learning (70 ) Data % 1248 samples 15
16. Assessing the performance of the model 16
17. Results so far 17
18. Ongoing Work 18 -Optimizing the neural network in order to get better performance. -Comparing the three models. -Generating set of features -Generate a linear and nonlinear weighted model on the basis of three neural network models. -Aapply methos of features selection which may allow to improve the accuracy such as dropout, PCA Stepwise fit. Further Work
19. THANKS! Any questions? 19