Evaluation of arima and ann stream analytics for air quality monitoring system
There are many environmental monitoring systems available in the market with Internetof-Things (IoT) enabled technology. However, the existing system is not equipped with online data analytics. Some of them provide analytics but are done in offline mode through third-party software or devices known...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2021
|
| Subjects: | |
| Online Access: | http://umpir.ump.edu.my/id/eprint/34942/1/Evaluation%20of%20arima%20and%20ann%20stream%20analytics%20for%20air%20quality%20monitoring%20system.ir.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | There are many environmental monitoring systems available in the market with Internetof-Things (IoT) enabled technology. However, the existing system is not equipped with online data analytics. Some of them provide analytics but are done in offline mode through third-party software or devices known as batch analytics. Pricewise, the existing monitoring system alone is expensive even though none of them are furnished with stream analytics. The thesis presents the design and development of an accurate air quality monitoring system equipped with streaming machine learning predictive analytics called Smart Environmental System (SES). The developed SES is divided into two sections End-Node Unit (ENU) and Gateway Unit (GWU). ENU consisted of calibrated sensors of NO2, CO, CO2, PM2.5, PM10, O3, temperature humidity integrated with Raspberry Pi Single-Board Computer (SBC) and Long-Range (LoRa) Transmitter (Tx) module. Meanwhile, GWU consisted of Raspberry Pi SBC, LoRa Receiver (Rx) and 4G module. The ENU transferred the data wirelessly to the GWU through LoRa communication, and GWU stored the data immediately in MySQL, which was installed in the Linux Apache MySQL PHP (LAMP) server. Investigation on evaluating senso rs’ accuracy is executed by comparing the collected data by SES vs data from the Department of Environment (DoE). The SES’s accuracy percentage error of CO, NO2, O3, PM10 are 5.1%, 7%, 6.1% and 6% correspondingly compared to DoE. Such accuracy of sensors is acceptable with an accuracy below 10%. Once accuracy has been validated, the data stored in MySQL database is successfully exported to the R query table in R-Server by using dbGetQuery() command, checked and aligned with the MySQL database. It is observed that the data in MySQL are successfully exported to the R query table based on the similar number of variables between those two tables. The data stored in the query table act as input to the analytics algorithm, which runs in R-server as well. In this thesis, two algorithms have been implemented and compared. Auto -Regressive Integrated Moving Average (ARIMA) and Artificial Neural Network (ANN). It is identified that ARIMA has better prediction accuracy (PA) percentage of 99.45%, 99.87%, 99.75%, 98.92% for CO, NO2, O3 and PM10 over ANN thus chosen as a predictive analytics algorithm for SES. Once embedded in SES, ARIMA performances are evaluated based on Mean Absolute Percentage Error (MAPE) and Prediction Accuracy (PA). It is observed that ARIMA MAPE is 1.64%,9.67%, 9.59%, 7.09%, for CO, NO2, O3 and PM10, respectively which led PA to achieve 96.78%, 90.33%, 90.41% and 92.91% correspondingly. The results proved that the proposed SES is able to precisely predict those gases for the next 24 hours above the 90% prediction accuracy. It can be concluded the proposed SES could be implemented as a future for the Air Pollutant Index (API) system. |
|---|