Yuliang Liu

The low-carbon transition of the power sector is fundamental to achieving “Dual Carbon” goals, where demand-side management plays an increasingly vital role in transforming flexible loads into renewable energy accommodation and active emission-reduction resources. However, existing low-carbon demand response mechanisms based on dynamic carbon emission factors only reflect average system states and fail to quantify the incremental carbon impact of marginal load changes. To address this limitation, this paper proposes a novel marginal carbon emission factor-driven low carbon demand response mechanism. Unlike traditional methods, the proposed mechanism utilizes marginal carbon emission factors as a high-sensitivity guiding signal to inform users of the real-time emission and renewable energy consumption variations caused by their consumption adjustments. Furthermore, considering the forecasting errors of high-penetration renewable energy, the uncertainty of marginal carbon emission factors is explicitly considered. Case studies are conducted to compare the proposed method with the conventional method through comparative analyses based on the modified PJM-5 system. Results demonstrate that the MCEF-driven approach provides more precise carbon-reduction and renewable energy utilization signals to achieve superior system-wide decarbonization performance and sustainable development.

To address the critical issues of lagging monitoring and inaccurate early warning in regional carbon emissions, this paper constructs a DDPR dynamic early warning model. This model is supported by “electricity-based carbon emission estimation” technology and integrates an improved Grey Model GM(1,1)-BP neural network optimized by BOA with concept drift retraining. A case study was conducted using monthly data from China’s Guangdong Province from 2020 to 2024 to verify the calculation accuracy and applicability of the early warning framework. Results demonstrate that the improved electricity-to-carbon prediction model delivers high-frequency calculation with a relative error rate maintained below 4%. The dynamic early warning model successfully identifies explicit progress deviations and latent cross-period risks by capturing three core risk scenarios: production rushes in May drive the DI to a peak of 1.257 (Emergency level). Summer peak and end-of-year production sprints trigger a surge in the RF from 90.41% to 99.31% between September and November, providing a critical early warning window several months in advance. By transforming high-frequency power data into near real-time situational awareness, this model enhances the foresight of regional carbon control and establishes a closed-loop framework for differentiated administrative and technical interventions.

Driven by the “Dual-Carbon” targets, precise, real-time, and cost-effective carbon emission monitoring has become imperative for formulating regional mitigation policies and evaluating compliance. However, conventional monitoring practices are often hindered by high costs, data latency, and the failure of static models to adapt to dynamic environmental shifts. To overcome these limitations, this study proposes an adaptive dynamic prediction framework grounded in an “electricity-to-carbon” mechanism. First, a hybrid Grey Model-Backpropagation (GM(1,1)-BP) neural network is developed to precisely model the non-linear coupling between power consumption and carbon emissions. Next, a Fuzzy-Weighted Drift Detection Method (FW-DDM) is integrated to monitor the prediction error stream in real time, triggering online model retraining immediately upon the detection of concept drift. Furthermore, Monte Carlo resampling is employed to quantify prediction uncertainty. Empirical results from Guangdong Province validate the feasibility of inferring carbon emissions from high-frequency electricity data, underscoring the proposed framework’s superior robustness in non-stationary environments. Finally, this paper discusses future research directions and the broader application prospects of “electricity-to-carbon” technologies at enhanced spatiotemporal resolutions.