Application of Multi-Scale Traffic Data Integration with Life-Cycle Carbon Accounting and CTMC Deep Learning Model in Travel Behavior and Emission Reduction Policies

Background California, USA, faces core challenges in transportation policy-making: insufficient data localization, poor policy effect adaptation to regional characteristics, and travel behavior prediction deviating from local habits. There is an urgent need to establish a transportation simulation and behavior prediction system integrating localized multi-source data to support precision governance(1). Methods This study integrated multi-source localized data (e.g., Caltrans public transport fares, NOAA meteorological data) to construct a micro-meso-macro multi-scale traffic data integration framework. Based on 578,000 road network data entries and 50,000 real travel behavior data entries, 9 groups of traffic policy scenarios were designed for TraCI simulation, and an enhanced CNN-Transformer-Mamba travel behavior prediction model was proposed. Results Simulation results showed that the high congestion pricing policy achieved the optimal efficiency in California's core road network, with key simulation indicators optimized by over 1.73% compared to the baseline scenario. The carbon emission accounting results matched California's actual data with a consistency rate of 88.3% (5). In terms of behavior prediction, the NIO model exhibited the best comprehensive performance in static scenarios (accuracy = 0.9039), while the enhanced model maintained stability during dynamic peak hours (recall = 0.8984). Conclusions This study successfully built a data-driven decision-making framework, providing reliable support for the precise evaluation of California's transportation policies and localized travel behavior prediction. The research results can be directly applied to the optimization of intelligent transportation systems in cities such as San Francisco and Los Angeles.