Urban residential carbon emission interval prediction model based on bidirectional Mamba integrated with Koopman-based feature extraction and enhanced quantile regression

Abstract

Accurately predicting urban residential carbon emissions is crucial for addressing global climate change. However, urban residential carbon emissions are influenced by multiple factors, exhibiting significant volatility, nonlinearity, and uncertainty. Conventional point prediction models struggle to effectively capture these characteristics, thereby limiting their application in urban carbon emissions management. In response to these limitations, this study proposes a novel interval prediction model named Koopman-BiMamba-EQR, which integrates a Koopman-based (Koopman) feature extraction module, a Bidirectional Mamba (BiMamba) modeling module, and an Enhanced Quantile Regression (EQR) module. The prediction performance of the model is comprehensively evaluated using daily residential carbon emissions data from four representative cities at the 50 % (conventional monitoring scenario), 85 % (robust regulation scenario), and 95 % (high-risk prevention scenario) quantile levels. Experimental results indicate that the Koopman-BiMamba-EQR model significantly outperforms two state-of-the-art baseline models in terms of prediction accuracy and interval stability, demonstrating excellent predictive performance. Ablation experiments further validate the complementary roles of the Koopman and BiMamba modules in enhancing overall prediction performance. Moreover, the necessity analysis, robustness evaluation, and practical implementation collectively highlight the scalability and applicability of the model in real-world scenarios. This study provides a highly precise, stable, and reliable interval prediction model for urban carbon emissions management, which can provide a theoretical basis and practical support for urban low-carbon transition, policy formulation, and risk management.