CO 2 Emissions Forecasting in Multi-Source Power Generation Systems Using Dynamic Bayesian Network

Climate change is one of the significant challenges that the planet is facing nowadays. CO 2 emission is the largest contributor, and it is mainly released by human activities. In Europe, the energy sector is responsible for roughly two-thirds of all greenhouse gas (GHG) emissions and the amount of CO 2 emitted from electricity production can greatly vary in time as a function of sources used to generate it. An accurate prediction of CO 2 emissions not only provides a basis for policymakers, but it can also assist the management of carbon emissions in making efforts towards limiting emissions generation and global warming as a consequence. For such a purpose, researchers have proposed the use of traditional algorithms for forecasting CO 2 emissions from the energy sector. However, there still are challenges yet to be overcome as regards forecasting CO 2 emissions from the energy sector. Power dispatch problem consists in planning the use of all available sources for minimising the environmental impact while at the same time satisfying the energy demand, stressing the necessity to dealing with this topic in multi-source power generation systems. In this context, this paper presents the use of discrete Dynamic Bayesian Networks (DBN) to forecast CO 2 emissions in a multi-source power generation system. The proposed methodology has been evaluated using the multi-source Germany grid data. The results were benchmarked against Multilayer Perceptron (MLP), K-nearest neighbor algorithm (KNN) and Random Forest (RF), and it was found that DBN achieved a significantly better performance due to reducing average NRMSE by 16.57%, average MAE by 19.88 $gCO_{2}eq/kWh$ and average MedAE by 27.48 $gCO_{2}eq/kWh$ in comparison with the second best method.