バイオガス発電施設を対象としたデジタルツイン改善システムの開発

抄録

This study proposes a data-driven approach to support hypothesis formation aimed at improving the predictive accuracy of digital twin systems. By applying Lasso regression, the research successfully identifies key explanatory variables and clarifies the factors contributing to prediction errors in a methane fermentation plant—a chosen use case. The study demonstrates that the proposed approach can potentially enhance prediction accuracy by up to 17% when the suggested hypotheses are implemented in the digital twin system. Furthermore, these hypotheses undergo validation through a Human-in-the-Loop process, ensuring their practical applicability. The findings highlight the significance of hypothesisdriven methodologies in optimizing industrial processes and suggest broader applicability across various fields, including other industrial and bioengineering domains.