Dynamic Carbon Emission Accounting for Electric Buses: A Probabilistic Deep Learning Framework Considering Grid Uncertainty

The electrification of public transit has emerged as a pivotal pathway for deep urban decarbonization. However, prevailingcarbon accounting methodologies predominantly rely on static grid emission factors and deterministic energy models, fre-quently overlooking the critical roles of spatiotemporal volatility and uncertainty propagation. This limitation biases life-cycleassessments and impedes precise fleet management. To bridge this gap, this study establishes a dynamic, uncertainty-awareframework for the rigorous carbon accounting of electric bus systems. We propose a novel hybrid deep learning architecture,integrating Temporal Convolutional Networks (TCN), Long Short-Term Memory (LSTM) networks, and an Attention mech-anism to capture multi-scale temporal dependencies in energy consumption. Furthermore, moving beyond deterministicassumptions, we formulate a probabilistic grid emission model utilizing Monte Carlo simulations to systematically quantifyand propagate uncertainties across the assessment chain. Validated on high-frequency operational telemetry from the Shen-zhen electric bus fleet, our proposed model achieved a coefficient of determination (R2) of 0.9610 and an RMSE of 0.0523,demonstrating superior predictive fidelity compared to both ensemble learning algorithms and traditional deep learning base-lines. Additionally, the results reveal a distinct heteroscedasticity in emission profiles, where prediction uncertainty significantlyexpands during high-power transient events. This study contributes to the field of precision decarbonization by presenting arobust, data-driven approach capable of generating confidence intervals for carbon footprints, thereby enhancing the reliabilityof decision-making for urban transport policy.