Research on small sample carbon emission prediction based on improved TimeGAN: A case study of the Yangtez River Delta urban agglomeration in China.

Abstract

Carbon emission prediction at the urban level is essential for effective reduction strategies. However, in the research on carbon emission prediction of the Yangtze River Delta urban agglomeration in China, it faces the challenges of difficult carbon emission calculation at the city scale, difficulty in carbon emission data collection, and insufficient data amount, which together hinder accurate forecasting of future urban emissions, presenting a classic small-sample dilemma. To address this, this paper first proposes a new carbon emission calculation model that integrates socio-economic data and nighttime light data to calculate urban carbon emission, categorizing cities into four types: high-tech, industrial support, private economy, and resource support. The multivariate regression model is used to calibrate the fitting coefficient of the nighttime light data of each city to correct the carbon emission calculation model, which significantly improves the accuracy of carbon emission calculation at the city scale. Furthermore, this paper proposes an improved carbon emission data augmentation model (TimeTGAN) based on time-generation adversarial network by introducing bidirectional temporal convolutional network (BiTCN) and dilated causal convolutions to effectively capture long-term dependencies in time series data, thereby generating more accurate and coherent carbon emission data. Comparison with conventional data augmentation methods shows that the TimeTGAN model offers a higher-quality experimental dataset for carbon emission prediction models. Finally, spatial autocorrelation analysis is used to reveal the spatial correlation of carbon emission. Based on this, a ConvLSTM spatiotemporal sequence prediction model is used to predict city-scale carbon emission and conduct an analysis of the prediction results.