|
Project Team:
|
Marko Blagojevič
|
|
Duration:
|
36 Months
1.3.2026 – 28.2.2029 |
|
Project Code:
|
J7-70260
|
|
Lead Partner:
|
UL FGG
|
|
Project Leader:
|
Sabina Kolbl Repinc
|
|
Other project Partner’s Organization:
|
Kemijski inštitut Slovenije
Fakulteta za elektrotehniko, Univerza v Ljubljani |
|
Source of Finance:
|
 ARIS |
|
Key words:
|
biogas/methane, machine learning, industry, biomethane potential, methane yield, methane yield database, optimization
|
Description:
The project aims to advance the science and practice of renewable energy, particularly in the field of anaerobic digestion (AD) and biogasproduction. Biogas is a renewable energy source produced through the anaerobic breakdown of organic materials by microorganisms, resulting inmethane (CH₄) and carbon dioxide (CO₂). Recognizing this, the EU has included biogas in its strategic energy and climate goals, with biomethanebeing a cornerstone of the REPowerEU plan. By 2030, the EU aims to produce 35 billion cubic meters of biomethane annually, significantlycontributing to the green energy transition and the circular economy. Regulatory frameworks, including RED II, RED III, and the Fit for 55 package,provide further support by promoting biogas production from waste. Biogas production faces challenges that include the complexity and variability of substrates, which hinder consistent methane production, and thelack of accurate predictive tools for assessing biomethane potential (BMP)/ methane yield (MY). BMP/MY is a critical parameter that estimates themethane production capacity of different organic materials under ideal conditions. However, BMP/MY values are influenced by numerous factors,such as nutrient content, total solids, pH, temperature, the presence of inhibitory substances, etc. The project takes an innovative approach to address these challenges by combining experimental data, advanced spectroscopic techniques, andmachine learning (ML). Spectroscopic methods such as Fourier Transform Infrared (FTIR) and Raman spectroscopy enable detailed analysis ofsubstrate composition and structural changes during the AD process. Furthermore, ML offers a transformative solution for analyzing complexdatasets, uncovering nonlinear relationships between substrate characteristics, process parameters, and methane yields. This approach eliminatesthe need for prior assumptions about system dynamics, allowing for more accurate and robust predictions.
Main Goals:
By incorporating an interdisciplinary approach that combines experimental data, spectroscopic techniques, and machine learning (ML), the research has the potential to significantly improve BMP/MY prediction. The application of ML models and methane yield metadata to enhance BMP/MY prediction remains relatively rare. The advantage of using ML models together with data from the Methane Yield Database (MYDB) to improve BMP/MY prediction lies in addressing a key research and practical gap in this field. This gap represents an opportunity for innovation and leadership in both science and industry. The uniqueness of this approach—integrating a harmonised and expanded MYDB with ML—lies in the fact that such an application of ML to datasets of this scale does not yet exist. The project offers an opportunity to develop new methodologies and gain novel insights into BMP/MY prediction. By incorporating metadata and advanced ML algorithms, the project can increase the accuracy and reliability of predictions by uncovering complex relationships between substrate characteristics, process parameters, and methane yields. In line with these objectives, the proposed research aims to: Expand and upgrade the infrastructure of the existing Methane Yield Database (MYDB) (http://methane.fe.uni-lj.si/ ); Significantly increase the complexity of the core dataset by adding new data and variables, such as detailed physicochemical parameters of substrates and inoculum, using advanced spectroscopic tools (FTIR and Raman); Determine and report new methane yield values that will provide reference data for further research and database expansion; Map existing data onto newly generated, extended datasets (including methane yields, FTIR, and Raman data) to ensure consistency and harmonisation; Perform machine learning analyses incorporating metadata from the expanded MYDB to develop predictive models of methane yield for different substrates and validate them at laboratory scale; Apply successful models to optimise methane yield prediction in biogas plants, with the aim of increasing biomethane production efficiency. The integration of experimental metadata, physicochemical parameters of substrates and inoculum, spectroscopic techniques (FTIR and Raman), the MYDB database, and ML methods will enable a novel and reproducible approach for investigating and optimising BMP/MY prediction and anaerobic digestion (AD) processes.
Work Packages:
Within Work Package WP1, the main objective is to establish the scientific foundation of the project through a systematic literature review on BMP/MY, anaerobic digestion (AD), advanced spectroscopic methods such as FTIR and Raman, and machine learning (ML) techniques for data analysis. The review will cover research on the anaerobic digestion of various types of substrates, including lignocellulosic residues, bioplastics, industrial and agricultural waste, and municipal wastewater sludge. Particular emphasis will be placed on methanogenesis kinetics, inhibitory effects, and the influence of physicochemical properties of substrates. In addition, this work package will analyse the application of advanced spectroscopic techniques and their relationship with methane yields. Finally, WP1 will include the collection of relevant BMP/MY data from existing databases, such as MYDB, and from other studies, with the data standardised for further integration into the expanded database.
Within WP2, the existing MYDB database will be upgraded with new physicochemical parameters of substrates and inoculum, as well as results from spectroscopic analyses. The database will be enriched with new data on substrate chemical composition and spectral data obtained using FTIR and Raman spectroscopy. The outcome of this work package will be a robust and expanded MYDB, enabling the development of advanced ML models in subsequent work packages.
WP3 focuses on experimental BMP/MY measurements for new types of substrates. Measurements will first be conducted on pure polymers to establish a baseline reference framework. More complex substrates will then be included. Substrate preparation will involve measuring key physicochemical parameters and performing spectroscopic analyses (FTIR and Raman), as well as conducting analyses before and after AD, enabling the monitoring of structural changes in substrates. BMP measurements will be carried out using the AMPTS II system, which enables precise real-time monitoring of biomethane production under controlled conditions. The measured data will be used in ML models.
In WP4, the focus will be on the development of advanced ML models for BMP/MY prediction. The first step will involve data preparation, including cleaning, normalisation, and integration of physicochemical parameters, spectroscopic data, and BMP/MY results from MYDB. Initially, baseline predictive models will be developed for pure polymers using different algorithms. Subsequently, advanced models will be developed that integrate complex datasets, including FTIR and Raman results, and investigate the influence of interactions between substrates and process parameters. Model validation will be based on data generated in WP3, ensuring accuracy and robustness. Finally, the models will be adapted for application in industrial biogas plants, enabling optimisation of AD processes, increased biomethane production efficiency, and improved reliability of BMP/MY predictions. The fifth work package includes overall project management, coordination among partners, and dissemination of results.