The project was developed as a solution to the burgeoning issue of air pollution in the OneM2M hackathon held at University of Texas at Dallas.
It utilizes machine learning along with IoT to provide real-time data on the website whether or not to wear a pollution mask that day, and which areas in the city to avoid (because of high pollution).
So in effect, it replaces a very costly pollution sensor with a set of cheap sensors and machine learning technology.
Why Machine Learning?
- Assume the pollution level is decided by independent pollution level values (eg: level of gas A>130; level of gas B>100 => polluted)
- But in real life, values are not independent, the pollution level cannot be determined from a single variable alone. Example- more traffic during morning and evening time of the day, and the affect of temperature on the pollution caused.
- Mask it uses machine learning model. It calculates weights to be assigned to different variables like temperature, humidity, gases etc. using perceptrons: Equation: A.w1 + B.w2 + C.w3 + D.w4+ E.w5
- The intention is to get a set of data verified and categorized as "pollution causing" or "non pollution causing" by the costly pollution sensor, so that the perceptron model can train on it and appropriate weights can be calculated.
- The current data streaming in is test data (real time) and gets classified based on weights calculated using machine learning model
- A map is also included which updates the regions which are safe to go to as per the pollution levels.
Why IOT?
- The interconnection of a number of sensors are used to supply real time data to the web application.
- In the demo, an LCD screen was connected to the server using oneM2M protocol and was displaying "Polluted" or "Non polluted" based on the web-app output.
- The web app was connected to 2 servers through the oneM2M protocol, one server was a dummy server providing values for gases, temperature & humidity. The second server was a laptop to which a luminosity sensor was connected to with this procedure.
Architecture
Self Improving Technology
- Can use the costly device and calculate validation values for one week of real time data to improve models
- This data when added to the machine learning model, updates weights and gives more accurate classification for further values
- This script can be run as and when needed
Real Time Data: oneM2M
- Real time data from the sensors displayed on website using OneM2M protocol
- We get data from 5 different sensors onto the server using the OneM2M protocol
- This data will be used by our model to compute the pollution status: Polluted/ Non polluted – from weight multiplication with real time values
- Pollution state is sent to the server and oneM2M protocol will send that back to the LCD
Area wise air quality detection
- Implemented color coded area display on website using Google Map API
- Can divide the city into quadrants and detect average pollution ratings for each quadrant
- Create a weighted vector using area value as well and show in map that area which has maximum pollution
Website Screenshots
Thank you!
Comments