Artificial Intelligence Applications in Biomass Pyrolysis: A Systematic Literature Review

The integration of artificial intelligence (AI) techniques into biomass pyrolysis research has attracted increasing attention in recent years; however, the existing literature remains fragmented across diverse methodological approaches and application domains. This study presents a systematic literature review of AI applications in biomass pyrolysis, combining bibliometric and qualitative analyses to map the current state of the art, identify prevailing research trends, and highlight existing knowledge gaps. Following a structured search conducted in the Scopus database, 33 peer-reviewed journal articles published in English between 2003 and 2026 were selected according to predefined eligibility criteria. The final portfolio was prioritized using an adapted version of the Normalized Index for Ranking Papers (NIRP 2.0), while the review procedure followed, whenever applicable, the recommendations of PRISMA, PRISMA for Abstracts, and PRISMA-S guidelines. The ranking methodology incorporated four scientometric indicators: Field-Weighted Citation Impact, average citations per year, SNIP, and CiteScore. The bibliometric analysis revealed a significant intensification of research activity during the last five years, with China, India, and Pakistan emerging as the most productive countries in the field. Machine learning techniques, particularly ensemble learning methods such as Extreme Gradient Boosting, Random Forest, and Gradient Boosting Decision Trees, were identified as the dominant approaches, especially in applications related to product yield prediction (biochar, bio-oil, and gas), kinetic and thermodynamic modeling, co-pyrolysis optimization, and process parameter estimation. Recent studies have also demonstrated growing interest in explainable artificial intelligence methods aimed at improving model interpretability and supporting physical understanding of pyrolysis systems. Despite the promising predictive and optimization capabilities demonstrated by AI-based models, important challenges remain, including limited dataset sizes, data heterogeneity, inconsistent terminology, reduced model generalizability, and the absence of physically informed constraints in many machine learning frameworks. The findings of this review indicate that future advances in the field will strongly depend on the development of standardized and publicly accessible databases, harmonized reporting protocols, and the integration of physics-informed artificial intelligence approaches capable of providing reliable, interpretable, and transferable predictions for biomass pyrolysis processes.