Spatiotemporal pattern evolution and quantitative prediction of electrical carbon emissions from a demand-side perspective in urban areas.

Amid global climate change, analyzing spatiotemporal patterns and predicting urban demand-side electrical carbon emissions is vital for regional low-carbon transitions. This study focuses on a developed coastal region in Guangdong, China. Utilizing high-frequency monitoring data from  3000 distribution network stations (May-Sept 2018), it creates an integrated 'spatiotemporal evolution-data driven prediction' framework to reveal emission dynamics and enhance forecast accuracy. Breaking through the limitations of traditional single-scale analysis, the study innovatively integrates monthly, daily and hourly time series with standard deviation ellipses and Kriging spatial interpolation technology, achieving a combination of spatial and dynamic spatiotemporal evolution analysis. The study found that the center of gravity of carbon emissions showed a significant southwest-northeastward migration trajectory, and there was a spatial differentiation feature of central urban agglomeration and peripheral area dispersion. The logarithmic mean divisia index analysis shows that finance and taxation are the primary positive driving factors, while the impact of values of industrial output and commercial consumption shows significant spatiotemporal scale differences. On this basis, the study proposed a prediction method that integrates feature engineering and bidirectional gated recurrent unit (Bi-GRU) to effectively capture carbon emission fluctuations, with an accuracy of 82.83%. The analysis framework and prediction model can provide methodological support for formulating emission reduction policies in the region and have significant application value.