Latest trends and dynamics in green, sustainable and smart biorefineries

• Bibliometric mapping of smart biorefinery research from 2014–2025 • AI and ML uses grouped into eight smart biorefinery application domains • Random forest, ANN, SVM and RSM emerge as key optimisation techniques • Lignocellulosic waste and microalgae dominate sustainable feedstock use • Smart methods enhance efficiency, evaluation and design of biorefineries Smart biorefineries, integrating artificial intelligence (AI), machine learning (ML), chemometrics, and digital monitoring technologies, have gained attention as tools to address limitations in large-scale biorefinery implementation. Here, a decade-long (2014–2025) bibliometric analysis was conducted to evaluate their evolution and identify current trends. The results indicate an increase in research activity since approximately 2019, with AI and ML applications becoming progressively more frequent. Smart methods are applied across eight main purposes, including process modelling, optimisation, parameter selection, sample characterization, decision making, climate or spatial monitoring, and automation. Artificial neural networks are the most frequently applied tools, together with established chemometric approaches such as principal component analysis and response surface methodology, while methods such as random forest and support vector machines have gained relevance in recent years. The analysed applications mainly target lignocellulosic biomass, agricultural residues, and microalgae, focusing on the production of platform chemicals and biofuels. The reviewed studies show that smart methods can support process optimisation and predictive control and may contribute to sustainability-oriented strategies. However, most applications are concentrated in process optimisation and modelling, with limited implementation in automation and system-level integration, and only a minority of studies incorporate quantitative sustainability assessment tools such as life cycle analysis or techno-economic evaluation. Key challenges remain related to data availability and standardisation, as well as limited consideration of model robustness, interpretability, and environmental trade-offs.