| |
 |
UNIVERSITY OF BUCHAREST
FACULTY OF PHYSICS Guest
2026-06-12 0:10 |
 |
|
|
|
Conference: Bucharest University Faculty of Physics 2026 Meeting
Section: Polymer Physics
Title:
Artificial intelligence tool for analysis, interpretation, and forecasting of environmental sensor data
Authors:
Stefan Caramizoiu1,3, Ana-Maria Iordache1, Ana-Maria Florea1,2, Marilena Claudia Stancu1,4, Bogdan-Ionut Bita1,2, Stefan-Marian Iordache1
*
Affiliation:
1) Optospintronics Department, National Institute of Research and Development for Optoelectronics - INOE 2000, 409 Atomistilor Street, Magurele, 077125, Romania;
2) Department of Electricity, Solid-State Physics and Biophysics, Faculty of Physics, University of Bucharest, 405 Atomistilor Street, P.O. Box MG-11, 077125 Magurele, Romania;
3) Department of Structure of Matter, Atmospheric and Earth Physics, Astrophysics, Faculty of Physics, University of Bucharest, 405 Atomistilor Street, P.O. Box MG-11, 077125, Magurele, Romania.
4) University of Valahia – Doctoral School of Economy and Umanist Science – History
E-mail
stefan.caramizoiu@inoe.ro
Keywords:
artificial intelligence; environmental monitoring; sensor data; anomaly detection; forecasting; air quality
Abstract:
Environmental monitoring systems produce heterogeneous data that must be cleaned, interpreted, and forecast before it can support decisions. This work presents the development of an artificial intelligence tool for real environmental sensor data stored in a PostgreSQL database. The monitored parameters include temperature, relative humidity, total volatile organic compounds, carbon dioxide, nitrogen oxides, particulate matter fractions PM1.0, PM2.5, PM4, and PM10, ultraviolet radiation, and illuminance. The software is designed to clean data, detect missing or inconsistent values, identify anomalies, interpret environmental trends, generate short-term predictions, and produce reports. The system focuses on acquiring and structuring real sensor data and defining an analytical workflow. The first methods used for implementation were statistical filtering, threshold-based anomaly detection, moving averages, linear regression, and classical machine learning models such as decision trees and random forests. These methods provide an interpretable baseline before moving to more complex time-series models, including recurrent neural networks or transformer-based architectures. The tool can support air quality assessment, detection of abnormal pollution events, and forecasting of relevant physical and chemical parameters. Future work will include model training and validation on the acquired dataset, comparison between baseline and advanced prediction methods, interactive visualization, and automatic generation of scientific reports.
Acknowledgement:
The authors were supported by the Core Program with the National Research Development and Innovation Plan 2022-2027, carried out with the support of MER, project no. PN 23 05.
|
|
|
|