Prediction of Beijing carbon emissions utilizing machine learning techniques with enhanced PLO integration

Carbon dioxide ( CO 2 ) emissions have profound and far-reaching effects on the Earth’s ecosystems, climate patterns, and human societies. In many cases, particularly for specific regions or countries, the scarcity of carbon emission data makes accurate prediction extremely difficult, which significantly hampers effective governance and underscores the urgent need for models capable of handling small sample sizes without sacrificing accuracy. Traditional large-scale data-driven prediction models often perform poorly in small sample contexts, especially in data-scarce regions or during the early stages of carbon emission monitoring, limiting their applicability. This paper presents a hybrid model for predicting CO 2 emissions, which combines a Tent chaos mapping, Sobol sequence, and an update mutation mechanism-enhanced Polar Lights Optimization (TSPLO) algorithm with a multifaceted ensemble BDGRU model, and a backpropagation (BP) neural network. First, the Polar Lights Optimization (PLO) algorithm is improved by integrating Sobol sequences, Tent chaotic mapping, and collision mutation strategies. The performance of the TSPLO algorithm is then evaluated using the IEEE CEC 2022 international standard test set. Finally, the TSPLO-BDGRU-BP model is assessed through a case study utilizing carbon emission data from Beijing. The experimental results demonstrate that the TSPLO algorithm significantly enhances the stability and predictive performance of the TSPLO-BDGRU-BP model, achieving the highest prediction accuracy among all the tested models. This model provides broader methodological support for carbon emission forecasting and offers strong technical support for promoting sustainable development.