Modelo de predição de impactos ambientais de produtos baseado na avaliação do ciclo de vida integrada à inteligência artificial

Abstract

It is estimated that 78% of American consumers and 70% of Brazilian consumers already value more sustainable lifestyles, primarily due to personal well-being and concerns for future generations. However, climate risks are already a reality and complex to address, requiring swift and effective responses. This underscores the importance of integrating techniques such as Life Cycle Assessment (LCA) and Machine Learning (ML) to support environmental decision-making in a rapid and reliable manner. LCA, regulated by ISO 14040 and ISO 14044, requires significant time and large volumes of data for its execution. On the other hand, ML, with its data processing capabilities and predictive modeling, offers a solution to significantly reduce this analysis time. This study aimed to propose and test decision-support models using ML trained with LCA datasets to predict environmental impacts. This exploratory research is structured into two main articles. The first reviews the literature on computational techniques and modeling languages for environmental impact prediction, with the objective of assessing the current state of LCA+ML integration. The second article applies ML models in three different tests, utilizing LCA datasets and comparing prediction results. The primary challenge of this research was managing the heterogeneity of LCA data, as different databases use varied formats, such as ecoSpold and ILCD, impacting interoperability. Open-source software, such as OpenLCA, was employed to ensure compatibility and portability across the 1,219 datasets evaluated using ML. The key outcome was a refined decision-support model that enables the entire training pipeline, from sample generation through data export from LCA databases to ML model performance analysis. It is expected that the developed support model will guide LCA and ML users, offering a replicable approach for other sectors and contexts.