Smart solutions for sustainable energy: Machine learning powers biochar production from aquatic biomass

The increasing global demand for sustainable energy and carbon materials, alongside pressing environmental concerns, necessitates innovative approaches to resource management. Biomass, a versatile renewable resource, offers significant potential for conversion into biochar, an alternative fuel and valuable carbon material. However, efficiently transforming diverse biomass types into high-quality biochar remains a challenge. A recent investigation, conducted by Zhilong Yuan, Ye Wang, Lingfeng Zhu, Congcong Zhang, and Yifei Sun from Beihang University and Hainan University, addresses this by developing a sophisticated machine-learning framework to optimize biochar production from aquatic biomass. This work bridges a crucial gap, as previous modeling efforts largely overlooked aquatic sources.

Data-Driven Conversion for Cleaner Energy

This comprehensive study compiled an extensive dataset of 586 data points from existing literature, detailing the properties of hydrochar (from hydrothermal carbonization, HTC) and pyrochar (from pyrolysis carbonization, PLC) derived from aquatic biomass, including macroalgae, microalgae, and duckweed. Researchers trained and evaluated five tree-based machine learning algorithms to predict biochar yields and properties such as nitrogen recovery, energy density, energy recovery, and residual sulfur degree. The chosen input parameters spanned 10 feedstock and process variables, encompassing elemental compositions, industrial components, and reaction conditions. This robust approach enabled the identification of the Random Forest Regression (RFR) and Extreme Gradient Boosting (XGB) models as top performers for predictive accuracy.

Precision in Hydrochar and Pyrochar Properties

The RFR model showcased exceptional predictive accuracy for hydrochar, achieving R2 values between 0.89 and 0.98 for hydrochar yield, nitrogen recovery degree, energy recovery degree, and residual sulfur degree. Analysis of feature importance revealed that beyond process parameters like temperature and time, feedstock elemental compositions, particularly nitrogen and sulfur content, critically influenced biochar properties. Similarly, the XGB model demonstrated strong performance for pyrochar, with R2 values ranging from 0.84 to 0.94 for energy density of hydrochar, pyrochar yield, and nitrogen recovery degree of pyrochar. Key factors impacting pyrochar properties included pyrolysis temperature, ash content, and carbon content of the aquatic biomass. Understanding these relationships is fundamental for tailoring biochar for specific applications, such as solid fuels or catalysts.

An Innovative Iterative Learning Method

A distinctive aspect of this investigation involves the development and demonstration of an iterative learning application method. Initially, the models showed moderate generalization ability when directly predicting new data. However, by incorporating even a small amount of new experimental data (as few as six samples) into the original dataset and retraining the models, predictive accuracy for subsequent new data significantly improved. For example, the retrained XGB model achieved an R2 of 0.97 for new pyrochar yield data, and the RFR model reached an R2 of 0.98 for new hydrochar yield data. This remarkable enhancement in generalization ability provides a highly efficient and cost-effective strategy for researchers, substantially reducing the need for extensive experimental trial and error.

Advancing Sustainable Bioresource Utilization

This research not only fills a critical knowledge gap in modeling biochar production from aquatic biomass but also offers a powerful, data-driven framework for optimizing these processes. While the current study focused on specific biochar properties due to data availability, future work could expand to include other critical parameters like surface area and pore volume as more data become accessible. The iterative learning approach developed here holds immense promise, offering versatility for adoption across various machine learning applications beyond biochar research.

"Our work signifies a crucial stride in leveraging machine learning models to transform aquatic biomass into high-value biochar products," states corresponding author Yifei Sun. "The demonstrated iterative learning method provides a practical and efficient pathway for researchers to quickly adapt and enhance predictive models with minimal new data, thereby accelerating the development of sustainable biochar production processes and reducing the associated labor and financial burdens in the pursuit of cleaner energy solutions."

Corresponding Author: Yifei Sun

Original Source: https://doi.org/10.1007/s44246-024-00169-2

Contributions: All authors contributed to the study conception and design. Material preparation, data collection and analysis were performed by Zhilong Yuan, Ye Wang, Lingfeng Zhu, Congcong Zhang and Yifei Sun. The first draft of the manuscript was written by Zhilong Yuan and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.