The Spatiotemporal Prediction Model of Urban Carbon Emissions Based on Graph Neural Networks

Introduction

In recent years, the problem of global climate change has become increasingly prominent, and its impact on the earth’s ecosystem and human society has been increasing. Since 2022, extremely high temperatures in the northern hemisphere have spread to Europe, North America, the Middle East, Asia, and many other regions, causing unprecedented extreme heat in many countries. This has seriously affected the local ecological environment and the normal life of human beings. This extreme climate phenomenon is a prominent manifestation of global warming. Global warming has become a major environmental problem that affects hundreds of millions of people around the world and has aroused widespread concern from all walks of life (Gu & Li, 2024).

One of the leading causes of global warming is the massive emission of greenhouse gases, in which carbon dioxide occupies a dominant position (Yang et al., 2023). In the process of energy consumption, such as oil and natural gas, most of their carbon-containing components are finally discharged into the atmosphere in the form of carbon dioxide. When these carbonaceous substances react with oxygen, they will generate carbon dioxide and water (Ji, 2024). With the development of the global economy, the energy demand is rising, and the carbon dioxide content in the atmosphere continues to increase due to excessive energy use (Chang et al., 2025). The rapid increase in carbon dioxide emissions is undoubtedly the chief culprit leading to the global greenhouse effect. In this context, responding to global warming and reducing carbon dioxide emissions has become an unavoidable responsibility and an obligation that all countries in the world must take the initiative (Ye et al., 2024).

Accurate spatiotemporal prediction of urban CE is of great significance in dealing with global warming. It can provide a scientific basis for environmental pollution control and help relevant departments to locate the key areas accurately and time periods of CE, to formulate more targeted control measures (Mou et al., 2023). Furthermore, the accurate prediction results also provide strong support for the formulation of energy conservation and emission reduction plans, and help to rationally plan energy use, optimize industrial structure, and reduce carbon dioxide emissions from the source (Zhao et al., 2021). Spatial and temporal prediction of urban CE can play a key role in slowing global warming and provide a guarantee for the sustainable development of human society (Zhou & Chen, 2021).

At present, there are still many challenges in the field of spatial and temporal prediction of urban CE. The existing forecasting methods are mainly based on measurement data or power flow results for post-assessment calculation (Cui et al., 2021). This method cannot accurately determine the power CE factors of different load nodes in the power grid in the future, and it is difficult to provide direct guidance signals for users to implement low-carbon regulation, which limits the effective implementation of energy-saving and emission reduction measures (Li et al., 2024). In view of this, this article proposes a spatiotemporal prediction model of urban CE based on a graph neural network (GNN). It aims to break through the limitations of traditional methods, predict the spatial and temporal distribution of urban CE more accurately, provide a more reliable basis for controlling environmental pollution and making energy-saving emission reduction plans, and contribute to the global response to climate change.

Compared with the existing research, the spatiotemporal prediction model of urban carbon emissions based on GNN proposed in this paper shows obvious innovation in many aspects. On the one hand, by constructing the graph structure between power grid nodes, the spatiotemporal coupling relationship of urban carbon emissions is deeply explored, which breaks the limitation that traditional methods only focus on a single time or spatial dimension. On the other hand, a graph transformation method considering passive no-load node elimination is introduced, which effectively improves the calculation efficiency and prediction accuracy of the model.