Data-driven prediction of pollutants emission from small-scale heating units using temporal deep learning

Result code in IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F61989100%3A27650%2F25%3A10258557" target="_blank" >RIV/61989100:27650/25:10258557 - isvavai.cz</a>

Result on the web

<a href="https://www.sciencedirect.com/science/article/pii/S2590174525004544?via%3Dihub" target="_blank" >https://www.sciencedirect.com/science/article/pii/S2590174525004544?via%3Dihub</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1016/j.ecmx.2025.101322" target="_blank" >10.1016/j.ecmx.2025.101322</a>

Result language

angličtina

Original language name

Data-driven prediction of pollutants emission from small-scale heating units using temporal deep learning

Original language description

Artificial intelligence (AI), particularly its subfield of machine learning (ML), has gained increasing attention in the field of environmental modelling and energy systems. These data-driven techniques offer robust tools for handling high-dimensional, nonlinear, and noisy datasets that are common in combustion diagnostics and emission prediction. This study investigates the use of advanced machine learning models for predicting flue gas emissions from residential heating systems under real-world operating conditions. Three types of solid-fuel boilers – automatic pellet, down-draught lignite, and gasification with hard coal – were analyzed using time-series data collected during full combustion cycles. Emissions of carbon dioxide (CO2), carbon monoxide (CO), nitrogen oxides (NOx), sulfur dioxide (SO2), and organic gaseous compounds (OGC) were modelled using two deep learning approaches: a neural network with long short-term memory (NN-LSTM) and a hybrid convolutional LSTM (CNN-LSTM). In addition, Random Forest analysis was applied to identify the most influential operational parameters driving emission formation.The results show that CO2 emissions are predicted most reliably, especially in the gasification boiler using NN-LSTM (R2 = 0.72). CNN-LSTM outperforms NN-LSTM in predicting CO and OGC in boilers with high variability, such as the down-draught system. However, both models face limitations when modelling NOx and SO2, suggesting the need for additional variables or physics-informed modelling. Feature importance analysis confirms oxygen concentration, flue gas temperature, and boiler heat output as key emission predictors.The findings demonstrate the feasibility of applying AI-based models for real-time emission estimation and optimization of small-scale combustion systems. This study provides a realistic baseline for future integration of predictive emission models with adaptive boiler control systems in residential energy applications.

Czech name

—

Czech description

—

Type

J_imp - Article in a specialist periodical, which is included in the Web of Science database

CEP classification

—

OECD FORD branch

20704 - Energy and fuels

Project

—

Continuities

O - Projekt operacniho programu

Publication year

2025

Confidentiality

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Name of the periodical

Energy Conversion and Management-X

ISSN

2590-1745

e-ISSN

2590-1745

Volume of the periodical

28

Issue of the periodical within the volume

october 2025

Country of publishing house

NL - THE KINGDOM OF THE NETHERLANDS

Number of pages

11

Pages from-to

101322

UT code for WoS article

001594645800001

EID of the result in the Scopus database

—