Scholarlink

<p>The addition of biochar is a key strategy to enhance anaerobic digestion (AD) performance. However, optimizing biochar synthesis to improve biogas production is labor-intensive and time-consuming. To address this, we developed a knowledge-based machine learning loop framework (KMLLF) to optimize biochar’s role in AD. Initially, biochar preparation was guided by scientific literature, and the AD experiments showed a 39.6% increase in cumulative methane production (CMP), reaching 318 ± 16.63 mL/g VS compared to the control group. New experimental data were integrated into the original Gradient Boosting Regression (GBR) model for retraining, resulting in improved predictive accuracy. The root mean square error (RMSE) for CMP decreased by 34.8%, and for Rmax, it decreased by 1.9%. Interpretable analysis of the new GBR model highlighted key factors affecting biogas production, including biochar preparation temperature, pH, specific surface area, and addition amount. Partial correlation analysis suggested optimal biochar particle size (0.5 mm), addition amount (≥ 15 g/L), heating rate (15°C/min), holding time (3 h), and processing temperature (650°C). Based on these insights, iterative optimization experiments were conducted, further increasing CMP to 333.51 ± 6.76 mL/g VS, an improvement of 4.8% over the literature-guided approach. This study demonstrates that KMLLF enhances the predictive accuracy of machine learning models in AD experiments, and facilitates the iterative optimization of biochar conditions, accelerating the development of optimal biochar for AD through artificial intelligence.</p> Graphical Abstract&#xa0; <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Building a knowledge-based machine learning loop framework to optimize biochar for anaerobic digestion performance

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月16日

查看全文
  • Yi Zhang,
  • Yu Fu,
  • Zhonghao Ren,
  • Yeqing Li,
  • Yijing Feng,
  • Zheng Hao Leong,
  • Junting Pan

摘要

The addition of biochar is a key strategy to enhance anaerobic digestion (AD) performance. However, optimizing biochar synthesis to improve biogas production is labor-intensive and time-consuming. To address this, we developed a knowledge-based machine learning loop framework (KMLLF) to optimize biochar’s role in AD. Initially, biochar preparation was guided by scientific literature, and the AD experiments showed a 39.6% increase in cumulative methane production (CMP), reaching 318 ± 16.63 mL/g VS compared to the control group. New experimental data were integrated into the original Gradient Boosting Regression (GBR) model for retraining, resulting in improved predictive accuracy. The root mean square error (RMSE) for CMP decreased by 34.8%, and for Rmax, it decreased by 1.9%. Interpretable analysis of the new GBR model highlighted key factors affecting biogas production, including biochar preparation temperature, pH, specific surface area, and addition amount. Partial correlation analysis suggested optimal biochar particle size (0.5 mm), addition amount (≥ 15 g/L), heating rate (15°C/min), holding time (3 h), and processing temperature (650°C). Based on these insights, iterative optimization experiments were conducted, further increasing CMP to 333.51 ± 6.76 mL/g VS, an improvement of 4.8% over the literature-guided approach. This study demonstrates that KMLLF enhances the predictive accuracy of machine learning models in AD experiments, and facilitates the iterative optimization of biochar conditions, accelerating the development of optimal biochar for AD through artificial intelligence.

Graphical Abstract