Answers are grounded in this paper and cite their sources.
Biochar properties are influenced by various factors, including raw material type and pyrolysis conditions. However, existing methods for predicting biochar properties exhibit significant limitations, particularly in the integration of structured and unstructured textual data. To address these challenges, this study proposes a multimodal data fusion framework based on large language models (LLMs) to improve the accuracy of prediction of biochar properties. First, we constructed a multidimensional biochar dataset incorporating the raw material composition, pyrolysis parameters, and biochar performance indicators. A customized multimodal algorithm was used to impute missing values, achieving an average R² value of 0.8863 across multiple features, indicating the effective restoration of data regularity. The data were then converted into question–answer pairs using a specific instruction format and fine-tuned for two rounds to characterize the model’s parameter adaptation under a continuous sample supplementation scenario. Among the models, InternLM2.5-7B-BiocharPrediction-v2 (InternLM2.5-7B-BP2) achieved the highest R² value of 0.9552, demonstrating its ability to form stable predictive mappings for gradually expanding in-distribution samples. In addition, compared with traditional machine learning methods, InternLM2.5-7B-BP2 showed clear advantages in handling multidimensional prediction tasks, with an average prediction accuracy improvement of 16.33%. Overall, this study demonstrates the potential of LLMs for biochar property prediction and provides a new framework and theoretical support for optimizing biochar performance prediction. Synopsis This study integrates large language models with multimodal data to improve prediction of biochar properties, as well as advance sustainable management and environmental remediation.
No takes yet. Share an insight, caveat, or question.
Peng et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: