To address the elevated calculation deviations of carbon emission intensity (CEI) caused by the neglect of thermal network transmission delays and adsorption effects in conventional static carbon flow models, along with the exacerbated challenges in low-carbon synergistic optimization due to heterogeneous multi-agent decision conflicts in electro-thermal coupled systems, a dynamic carbon flow tracking model (DCFTM) was developed. Thermodynamic transport equations were coupled with Langmuir adsorption kinetics to quantify transmission delay and adsorption buffering effects. A hierarchical master-slave-coalition game framework was constructed, where elastic carbon quota constraints of the power grid were coordinated with dynamic thermal storage regulation through spatiotemporal decoupling strategies and adaptive learning rate mechanisms. Simultaneously, structural equation modeling (SEM) was employed to analyze the dynamic correlation mechanisms among the dynamic carbon efficiency index (DCEI), economic cost sensitivity (ECS), and collaborative equilibrium degree (CED).Experimental results demonstrated that the normalized root mean square error (NRMSE) of CEI is reduced from 22.1% (static model) to 2.9% by the DCFTM, validating the necessity of adsorption kinetics modeling. In a 30-node power grid and 6-node thermal network system, the proposed algorithm achieves a 37.5% reduction in total carbon emissions and 15.6% operational cost savings, with the Pareto solution set surpassing traditional algorithm performance boundaries. SEM path coefficient analysis revealed that the enhancement effect of collaborative equilibrium on carbon efficiency is significantly stronger than the inhibitory effect of economic regulation. Moreover, the synergistic strategy combining higher carbon pricing with wind power penetration is shown to alleviate the low-carbon economic conflict, reducing the path coefficient of carbon efficiency to total cost by 17.3%.