GA-PSO-BP prediction of carbon emission and factor analysis

Abstract

In response to the escalating challenge of global climate change, this study presents a robust forecasting framework for carbon emissions based on a hybrid GA-PSO-BP neural network model. Leveraging high-dimensional multi-sector emission data from Sichuan Province, China, Principal Component Analysis (PCA) was employed to effectively reduce data dimensionality and eliminate noise, thus enhancing model stability. The BP neural network, optimized through the integration of Genetic Algorithm (GA) and Particle Swarm Optimization (PSO), was trained to capture the complex, nonlinear relationships inherent in carbon emission trends. The results reveal that the manufacturing sector remains the dominant contributor to overall emissions, while the wholesale and retail industry also exhibits significant emission levels, underscoring the need for targeted mitigation strategies in the service sector. Moreover, time series forecasting indicates that carbon emissions are projected to increase until approximately 2025, after which a marked decline is anticipated—a critical turning point for carbon peaking and neutrality policies. Overall, the proposed method not only delivers high prediction accuracy with low error rates but also provides actionable insights for policymakers and industry stakeholders to implement effective carbon reduction measures.