From data to digestion: AI-powered loop supercharges biochar for cleaner biogas

What if artificial intelligence could turn centuries of scientific literature—and just a few lab experiments—into a smarter, faster way to produce clean energy from waste?

That’s exactly what Dr. Yeqing Li and Dr. Junting Pan have achieved with their innovative “knowledge-based machine learning loop framework” (KMLLF), a breakthrough now published in the open-access journal Carbon Research (Volume 4, Article 71, December 16, 2025). Their work redefines how scientists design biochar—the charcoal-like material increasingly used to turbocharge anaerobic digestion (AD), a key process for turning organic waste into renewable biogas.

Anaerobic digestion holds huge promise for a circular bioeconomy, but its efficiency hinges on one tricky variable: biochar. Not all biochar is created equal. Its performance depends on dozens of synthesis parameters—temperature, particle size, heating rate, pH, surface area, and more. Traditionally, optimizing these factors meant months of painstaking trial-and-error.

The team’s novel framework starts by distilling decades of published research into an initial set of “smart guesses” for biochar preparation. They then test these in real-world AD experiments—and here’s the twist: the new data don’t just sit in a notebook. They’re fed back into a machine learning model (specifically, a Gradient Boosting Regression algorithm), which learns, adapts, and refines its predictions.

The results speak volumes. In their first round, guided by literature alone, they boosted cumulative methane production (CMP) by 39.6%—reaching 318 ± 16.63 mL/g VS compared to controls. But after looping experimental results back into the model, predictive accuracy sharpened dramatically: RMSE for CMP dropped by 34.8%. The updated model then pointed to precise optimization targets—like a processing temperature of 650°C, a particle size of 0.5 mm, and a biochar dose of at least 15 g/L.

When they followed those AI-refined instructions? Methane output climbed again—to 333.51 ± 6.76 mL/g VS, a further 4.8% gain.

“This isn’t just machine learning applied to science,” says Dr. Yeqing Li, corresponding author and researcher at the State Key Laboratory of Heavy Oil Processing and Key Laboratory of Biogas Upgrading Utilization, College of New Energy and Materials, China University of Petroleum Beijing (CUPB), as well as the Shandong Institute of Petroleum and Chemical Technology, Carbon Neutrality Research Institute in Dongying. “It’s a living loop where knowledge and data continuously improve each other. We’re accelerating discovery without sacrificing scientific rigor.”

Co-corresponding author Dr. Junting Pan of the Institute of Agricultural Resources and Regional Planning, Chinese Academy of Agricultural Sciences, adds: “For rural biogas plants or municipal waste facilities, this means faster deployment of high-performance biochar—turning agricultural residues or food waste into reliable, renewable energy with less guesswork and lower cost.”

Critically, the KMLLF doesn’t treat AI as a black box. Through interpretable machine learning techniques, the team identified which biochar properties matter most—giving engineers not just better outputs, but deeper understanding.

Published open access, the study offers a ready-to-adopt blueprint for researchers, biogas operators, and policymakers worldwide. At a time when scaling up renewable gas is critical for decarbonizing hard-to-abate sectors, this fusion of domain knowledge, experimentation, and adaptive AI marks a leap forward.

Thanks to the collaborative leadership of China University of Petroleum Beijing—a national hub for energy innovation—and the Chinese Academy of Agricultural Sciences, a pillar of China’s sustainable agriculture strategy, the future of bioenergy just got smarter, cleaner, and significantly more efficient.

Because in the race to net zero, the best solutions aren’t just green—they’re intelligent. And now, they learn as they go.

=== 

Journal reference:  Zhang, Y., Fu, Y., Ren, Z. et al. Building a knowledge-based machine learning loop framework to optimize biochar for anaerobic digestion performance. Carbon Res. 4, 71 (2025).   

https://doi.org/10.1007/s44246-025-00226-4 

=== 

About Carbon Research

The journal Carbon Research is an international multidisciplinary platform for communicating advances in fundamental and applied research on natural and engineered carbonaceous materials that are associated with ecological and environmental functions, energy generation, and global change. It is a fully Open Access (OA) journal and the Article Publishing Charges (APC) are waived until Dec 31, 2025. It is dedicated to serving as an innovative, efficient and professional platform for researchers in the field of carbon functions around the world to deliver findings from this rapidly expanding field of science. The journal is currently indexed by Scopus and Ei Compendex, and as of June 2025, the dynamic CiteScore value is 15.4.

Follow us on Facebook, X, and Bluesky.