부산대학교 도서관

In this paper, we first use the BP model optimized by particle swarm optimization (PSO) to integrate DMSP-OLS nighttime light data and NPP-VIIRS nighttime light data to establish a long-time series of nighttime light data sets were obtained. And using the integrated nighttime light data, temperature data and geographical data, based on the Bayesian optimization algorithm (BO) optimized XGBoost model, a carbon dioxide emission prediction model was constructed. In this study, the dataset based on nighttime light data shows a significant positive correlation to carbon emissions. As a result, the dataset can be used to estimate carbon emissions. And when using the BO-XGBoost model to invert carbon emissions, the R 2 is 0.9791, and the RMSE is 29.9118, which is the most accurate predictive outcome compared to other models. The results of this study will help simplify the work of carbon emission accounting and provide an important basis for regional carbon emission policies.