Under the background of green low-carbon economy, it is of great significance to accurately estimate the future CO2 emissions of countries with large CO2 emissions for the development of the world green economy. A new Nonlinear Grey Bernoulli and BP neural network combined model (BP-ONGBM (1,1) model) has been proposed to study the CO2 emissions of China, the United States, the European Union, India and Japan. Firstly, the Particle Swarm Optimization (PSO) algorithm is optimized by using the idea of Artificial Fish Swarm Algorithm (AFSA), and then the background value of ONGBM (1,1) model is dynamically optimized. Based on the linearization of the model, the time response function is derived. Then, the ONGBM (1,1) model is combined with the BP neural network model. The combination weight and the background value coefficient are determined by improved PSO algorithm. Finally, according to the observation data from 2010 to 2021 in the Emissions Database for Global Atmospheric Research 2022, the model is established to calculate the CO2 emissions of the selected countries from 2022 to 2026, and compared with the prediction results provided by multiple competitive models. The empirical application shows that the newly proposed BP-ONGBM (1,1) model is significantly better than other competitive models.