Einträge gefunden: 2693 von 2766. Filter zurücksetzen
 
  Titel: Predicting and improving indoor air quality based on IoT sensor data and machine learning
  AutorIn: Manuel Berger
  Typ: Bachelorarbeit
  ÖFOS 2012 Code: 102019 Machine Learning
  Institution: Ferdinand Porsche FernFH, Wiener Neustadt, WIBA
  Betreuung: Tom Gross
  Datum: 2022
  Abstract (de):
  Abstract (en):
Indoor air quality is affected by many variables, such as temperature, humidity, particulate matter, or gases like carbon dioxide. Considering that European citizens spend, on average, 90% of their time indoors (cf. EPA 2021, WHO 2013, 9), the indoor air we breathe must be healthy and free of pollutants. Even though most people are aware of the harmful health effects caused by indoor air pollution, many do not take the necessary steps to prevent them or to improve indoor air quality (cf. Osagbemi, Adebayo and Aderibigbe 2009). Based on this background, a prototype was developed to help occupants of a room to improve indoor air quality. Therefore, a system was built to record crucial indoor air quality parameters. Sensor measurements were recorded every 10 seconds over five months, from May 2022 until September 2022. The focus was on the carbon dioxide concentration, a parameter easy to measure that increases when humans exhale air (cf. Palmer, 2015) into the room and that decreases when windows or doors are opened. One crucial question was whether machine learning could reliably predict indoor air quality parameters such as CO2 to notify room occupants at the right moment, for example, before a parameter exceeds its limit or in further development to improve heating costs in winter. Therefore a machine learning model was trained with the historical recordings of CO2 concentrations and covariates such as the state of open doors and windows. The other, more important question was whether the system’s suggestions could help room occupants to improve indoor air quality significantly and efficiently in a user-friendly and practical way. As a result, a notification- and recommendation service, a web application, and a voice interface were developed. Three methods were used to find answers: Historical forecasting to test the performance and accuracy of the machine learning model, a user survey to receive feedback about practicality and user-friendliness, and an experiment to evaluate whether the system can help improve indoor air quality. Based on previously recorded sensor data, historical forecasting showed a prediction accuracy of 94.58%. The user survey concluded that 96.9% of participants found the system’s suggestions user-friendly and practical. The experiment showed that indoor air quality significantly improved thanks to the system’s notifications and recommendations. The results show an excellent potential of the system and an overall positive acceptance of users towards IoT devices and sensor data helping to improve indoor air quality.
  Keywords (de):
  Keywords (en): Indoor air quality, Machine learning, IoT, CO2, sensors, AWS, Virtual Assistant, Cloud
 
PDF-Dokument Berger_Manuel_51910535_SS22_Gross_2022-09-14.pdf