An advanced data-driven spatiotemporal fusion discrete grey forecasting model for CO2 emissions prediction in urban agglomerations

Purpose Accurate predicting CO2 emissions from urban agglomerations is vital for optimizing CO2 emissions reduction policies and energy structure. Existing grey models typically treat CO2 emissions units as independent units, extracting features in temporal dimension for forecasting. However, they neglect features in spatial dimension that exist among CO2 emissions units in urban agglomerations. Therefore, we will formulate an advanced spatiotemporal fusion discrete grey forecasting model in this paper. Design/methodology/approach Firstly, an asymmetric time-varying spatial weight matrix is constructed using a combination of data-driven and prior knowledge to quantify the asymmetric time-varying spatial correlation intensity between spatial units. Subsequently, a spatial correlation term, in combination with the proposed matrix, is designed to capture the correlation feature among spatial units. Then, a nonlinear time term driven by a power function is established to model the differential nonlinear trend of each spatial unit over time. Through incorporating the spatial correlation term and the nonlinear time term into the grey difference equation, a spatiotemporal fusion discrete grey prediction model (STFDGM(1,1,m)) is developed. Further, particle swarm optimization (PSO) is employed as an effective tool for nonlinear parameters optimization, thereby enhancing the model adaptability to diverse datasets. Findings A case study about CO2 emissions forecasts in Yangtze River Delta (YRD) demonstrate that the R2 and MAPE of the STFDGM(1,1,m) are enhanced by 8.53%–147.73% and decreased by 48.40%–97.40%, respectively, relative to nine comparison models of three categories and surpass the comparison models in terms of the tightest limit of agreement. Furthermore, Monte Carlo simulation experiment, ablation experiment and robustness test are conducted to verify the effectiveness of PSO in solving nonlinear parameters, the necessity of model optimization and the model stability, respectively. Originality/value In light of the spatial correlation feature and nonlinear time development trend amongst CO2 emissions of urban agglomerations, as well as the asymmetric and time-varying characteristics of the intensity of the spatial correlation, the STFDGM(1,1,m) model is proposed by integrating an asymmetric time-varying spatial weight matrix, spatial correlation term and nonlinear time term. This model provides a valuable tool for policymakers and researchers to accurately forecast CO2 emissions and develop effective carbon reduction strategies in urban agglomerations.