Digital twin for machine-learning-based vehicle CO2 emissions concentration prediction in embedded system

In this paper, we describe the design, implementation, and installation of a digital twin version of a physical CO 2 monitoring system with the aim of democratizing access to affordable CO 2 emission measuring and enabling the creation of effective pollutant reduction strategies. The presented digital twin acts as a replacement that enables the measuring of CO 2 emissions without the use of a physical sensor. The exhibited work is specifically designed to be installed on a low-powered Micro Controller Unit (MCU), enabling its accessibility to a broader base of users. To this end, an optimized Artificial Neural Network (ANN) model was trained to be capable of predicting CO 2 emission concentrations with 87.15% accuracy when performing on the MCU. The ANN model is the result of a compound optimization technique that enhances the speed and accuracy of the model while reducing its computational complexity. The results outline that the implementation of the digital twin is 86.4% less expensive than its physical CO 2 counterpart, whilst still providing highly accurate and reliable data.